Liquid cooling is fast becoming the default for artificial intelligence (AI) data centers, as the chips inside each server rack generate more heat than air can carry away. The shift has accelerated sharply over the past year.

AI training racks draw around 70 kilowatts (kW) on average today. Coming generations of Nvidia hardware could push that to 500kW, a load that conventional fans, floors and electrical distribution were never designed for.

“Liquid cooling is the only solution for these rack densities,” Matthew Baynes, vice president of Schneider Electric’s Secure Power and Data Centre division for the UK and Ireland, told TechJournal.uk in an interview. “Mechanical cooling technology design has rapidly moved from air to liquid in the past 12 to 18 months.”

“The next evolution of chip could be up to 500 kilowatts per rack. A traditional data center just cannot cater for that type of design or type of rack load,” he said.

A rack’s kilowatt rating measures the power it draws continuously, not its hourly use. A 500kW rack running around the clock would consume about 4.4 million kilowatt-hours (kWh) a year, as much electricity as roughly 1,750 UK homes, based on the typical household benchmark of 2,500kWh a year set by the energy regulator Ofgem.

By comparison, the average enterprise server rack draws about 9kW, or roughly 79,000kWh a year if run nonstop, enough for about 32 homes. An AI rack packs dozens of graphics processing units (GPUs) wired to work as one machine. Nvidia’s current GB300 NVL72 holds 72 GPUs and draws about 140kW.

Its Rubin Ultra rack, due in 2027, is designed to draw around 600kW.

A GB300 rack is no bigger than a standard server cabinet, measuring 600 millimeters (mm) wide, 1,068mm deep and 2,236mm tall. Loaded, it weighs about 1,580 kilograms (kg), and Nvidia says its Vera Rubin rack weighs about 1,800kg.

Clusters of such racks train large AI models on vast datasets. The same hardware also runs finished models to answer users’ requests, a task known as inference. In an industry benchmark, a single GB300 rack generated about 674,000 tokens, or word fragments, per second on DeepSeek’s R1 model.

Baynes said Nvidia’s move from GB300 to Vera Rubin pushes densities to 200kW and 250kW, with the generation after that possibly reaching 500kW.

“Their pace of research and development (R&D) change is nothing like the industry’s seen before. Every six months, what we talked about in terms of rack density changes by a factor of whatever it might be,” he said.

JLL, the commercial real estate firm, expects average rack density to triple to 45kW by 2030, with 80% of new facilities adopting liquid cooling.

“You may have seen Schneider with a strategic acquisition of Motivair, an American liquid cooling company, to position ourselves strongly in the world of liquid cooling,” Baynes said.

Schneider agreed in October 2024 to buy a 75% stake in Motivair for $850 million in cash and plans to acquire the remaining 25% in 2028. Founded in 1988 in Buffalo, New York, Motivair makes coolant distribution units (CDUs), rear door heat exchangers, cold plates and chillers.

On September 23, Schneider launched the WCDU, the largest CDU in the Motivair range. A single unit can cool up to 3.5 megawatts (MW).

Up to 20 units can run together to serve data halls of 30MW to 40MW. The WCDU sits in the technical corridor outside the server floor and combines liquid cooling with fan-wall air cooling, so operators can shift the balance between the two without redesigning the hall. It ships in select regions in October.

Months to weeks

Baynes took the UK and Ireland role in January after more than 20 years at Schneider.

He said a newer group of customers, AI labs and neoclouds that rent out GPUs as a cloud service, wants speed above all. A traditional data center takes 18 to 24 months from design to ready for service, but these operators want to be running within six to 12 weeks.

“They are challenging the industry to move data center builds from months to weeks,” he said. “They’ve got [GPUs] on order with Nvidia. They need to deploy them and convert that into revenue as soon as possible. These GPUs are expensive. You need deep pockets.”

He said a lot of Schneider's recent demand has come from AI labs and neoclouds in Europe and beyond, and the company is in discussions with them in the UK and Ireland.

Britain’s problem is that most of this AI training work is heading elsewhere.

“The cost of energy in the UK is around four times the cost in the Nordics, which means the UK struggles to attract that type of IT (information technology) load,” Baynes said. “We’ve struggled to attract the large-scale AI training and AI lab load market in the UK because, firstly, we are challenged with having access to energy, but of course our cost is prohibitive too.”

He said AI training campuses range from 50MW to one gigawatt (GW), and much of that training need not stay within national borders. It therefore follows cheap power to:

• The Nordics, with abundant hydroelectricity

• Iberia, with solar power

• France, with nuclear capacity and national AI ambitions

“In the UK, we have the AI Growth Zones, and there’s an element of power capacity there, and a slight incentive on cost of energy. It still doesn’t get close to the cost of the Nordics or Iberia,” he said.

He said UK power costs about three times Iberia’s, and discounts of 10% to 15% in the growth zones still do not make the economics of training work.

Queuing for the grid

Access to power is the other obstacle. Baynes said a new UK grid application now faces a wait of seven to 10-plus years, and press reports suggest some new connections will not be available until 2040.

He said established operators that have filed requests year after year will see connections arrive steadily, while newcomers wait longest. Reform from Prime Minister Andy Burnham’s government, which took office in July, could speed up the queue.

“If you’re stepping in now as a new logo and requesting a connection outside of the [AI Growth Zones], it could be quite lengthy. That’s one of the biggest challenges to hyper-scale data center growth in the UK, but steady growth will be maintained,” he said.

As AI shifts toward inference, Baynes expects that steady growth to come from enterprise and government workloads that need to stay in the country.

“When AI becomes more enterprise and more sovereign, we’ll see more demand come into the UK, and we are seeing interest in neoclouds landing in the UK and building some sites in the UK,” he said. “The cost of energy would not be number one in that regard.”

JLL expects inference to overtake training as the dominant AI workload in 2027, and AI to account for half of all data center workloads by 2030.

Operators are also looking for ways to get around the grid altogether.

Baynes pointed to Era4, a UK developer that uses waste to generate power, though he said such partnerships are not available to everyone. Era4 operates 44 sites and is building AI data centers on landfill, including one in Canterbury linked by private wire to a landfill-gas plant.

He also cited on-site microgrids such as Pure DC’s 110MW system in Dublin. He said microgrids can cut energy costs compared with the national grid, though they require upfront capital spending that can be restructured as an operating cost. He called them a viable option.

Small modular reactors (SMRs), such as those Rolls-Royce is developing, are further off, he said.

“In the UK, we’re a nuclear country, so there’s a possibility for us to do it. But there are quite a few hurdles the industry needs to go over before maybe we’ll see a live SMR in the UK,” he said.

Schneider ships the WCDU in October, but its UK customers still need the new government to shorten the power queue.

Keeping skilled workers has become a daily task for factories racing to build power equipment for artificial intelligence (AI) data centers, according to a former electrical apprentice with more than 30 years in the industry.

Demand for engineering skills has tipped the balance toward workers, and employers now have to earn their loyalty.

“When I was young, you thanked everyone that you had a job. You were lucky to have a job. Nowadays, the employer is lucky that they’ve got that person there to employ, because it is an employees’ market,” Michael Beagan, managing director of TES Power, told TechJournal.uk in an interview in Ballykelly, Northern Ireland.

“What we have to do is make sure that we retain the staff. We have to make sure that the staff are all happy here,” he said.

He said people want to see progression and understand the career paths open to them, so the company is clear about routes forward and promotes from within. He added that a team spirit, in which staff work hard but also enjoy the job, goes a long way toward keeping them.

Beagan also runs an open-door policy, so anyone in the company can talk to him at any time.

“When information isn’t there, people make up their own information, and that’s normally a lot worse. So we try and be very open with everyone,” he said.

He said attrition is extremely low, but commissioning engineers, who are in short supply across the industry, are harder to keep. Once they are fully trained, the company is lucky to hold on to them for 18 months to two years.

"We spend all the time investing in and growing these people, and then you send them out into the big bad world, and then the wolves swoop in," Beagan said at a company briefing on the same day.

TES Power is the data center arm of TES Group, which French electrical group Legrand bought in April 2026.

The company took over the Ballykelly site in County Londonderry, then a derelict former military building, four years ago. It now employs nearly 200 people there, making low-voltage switchgear and modular power rooms for data centers across Europe.

Fridays off by vote

The interview took place on September 23 during a two-day media tour of Northern Ireland.

The clearest test of that openness came a few months ago, when staff asked to squeeze their 40-hour week into four days, from Monday to Thursday.

“We said, look, okay, we’ll look at this. We’ll do a vote among everyone, but at least 80% of the staff needs to be for this, because it can’t just be 50% plus one. It’s such a big change,” Beagan said.

Some 97% voted in favor, and the four-day week went ahead.

“Even the few people that didn’t want it at the time have now come and said, it’s pretty good now,” he said.

The same instinct shapes how the company trains its newest recruits.

“Thirty-odd years ago, when I was doing my apprenticeship, things were different. You were released into a company and just kind of tagged in. You had no thought or planning for your progression,” he said.

TES Power now employs an in-house technical training officer and a dedicated apprentice mentor to support close to 30 apprentices.

This year it took 16 recruits onto a new panel wiring apprenticeship at North West Regional College, the first course of its kind on the island of Ireland. It plans to take on a similar intake every year.

“I always tell people when they first come in here, don’t panic. You don’t need to understand the bigger picture. Understand this small part of it, and then this small part of it, and eventually the whole picture will connect,” Beagan said.

He said training has to explain why each step matters, not just what to do. A driver who follows a sat nav learns nothing about the route, while one who learns the journey remembers it next time.

He does not believe young people have it harder today, as apprentices can earn while they learn without university debt.

Robots for repetitive work

Automation is part of the plan, but it targets tasks, no

t people. A new “lights-out” sheet-metal punch lets one operator run two machines, while an automated powder-coating line does in a morning what two painters once did in a week.

“We automate the tasks that carry little value for a human, and allow the humans to carry out the tasks that have a great value,” Beagan said.

He said some repetitive jobs are still done by hand only because the capital expenditure to automate them cannot yet be justified. Once it can, he would move those workers to more meaningful work, and he does not rule out humanoid robots taking on the most repetitive tasks.

In the meantime, staff are rotated so that no one does the same repetitive job every day.

"There'd be nothing more soul-destroying than if your job was to come in and do the exact same process all day, every day," he said.

Beagan said the workforce is not split 50-50 between men and women, and the company sets no hiring targets.

Women work as apprentices, shop-floor staff, project managers and designers. On the shop floor, many have gravitated toward wiring, which he described as dextrous work demanding fine detail, while mechanical fitters bolt in heavy copper. He said this was not planned, as some people simply lean toward electrical work and others toward mechanical.

“Society isn’t made up of one type of people, so why should a workforce be made up like that too? It allows the free flow of ideas and thoughts and processes,” he said.

Every morning, production staff from purchasing, design, fabrication and assembly meet in a huddle to review the previous day’s problems. Anything fixed comes off the board, while anything unresolved by the next morning moves to a formal tracker with a named owner.

Staff also tell the company what bugs them about their jobs, and the company puts a process or system in place to fix it.

He said every worker needs to know three things to do the job well:

• What to do

• How to do it

• When to do it

“That sounds so simple, but the reality is it’s a lot more difficult when you’ve got all these different departments that are interconnected. This is a fast-paced environment, and it’s essential that everyone’s well-being is considered all the way through,” he said.

His own career follows the practical path he preaches. He said academic study was never his strength and that he learned on the factory floor. As a child, he took apart every toy he owned.

He started as an electrical apprentice in 1993, later set up a US factory making switchgear to both European and North American standards, and joined TES Power in 2022 to run the Ballykelly site. He became managing director in November 2025.

“The difference that I have is probably 30 years of mistakes that I’ve made and learned from, which is very difficult to replicate unless you’ve felt the pain of those mistakes,” he said.

He said he is happy to hire graduates with more technical knowledge than he has and loves being challenged. He said the same appetite for learning runs through the company, which will keep training and upskilling its own staff as it expands to prepare the next wave of engineers and electricians.

TES Group, a Northern Ireland-based water treatment systems builder and power equipment supplier, is riding a shift in data centers away from on-site power equipment and toward modular rooms built in factories.

These rooms, known as pods and skids, arrive packed with switchgear, uninterruptible power supplies (UPS), cooling units and batteries, cutting the labor needed on site.

“The whole industry has flipped now. Traditional data centers used to be constructed. Now they’re manufactured,” said Michael Beagan, managing director of TES Power. “The modular rooms are all everyone’s talking about, which is fantastic for us.”

“One reason you can’t build data centers like I used to is that you can’t find 1,000 people in 20 locations. That’s why it’s a modular building factory, which means you need 200 people, not 1,000, to build the site,” said James Peacock, chairman of TES Power.

Pods are clad units that sit outside a data center, while skids carry the same equipment unclad inside the data hall.

TES works to about 80% standardized and 20% customized design. A client building in Frankfurt, Warsaw and Milan wants the same product in each city, but every site needs changes to voltage, space or operation.

“Building switchgear and putting copper in boxes has never changed since the 1960s, and it’s not going to,” Peacock said. “There’s always going to be some sort of manual stuff. It’s about how we make sure that skill is used for the real skillful bit.”

Each switchgear board is bespoke but assembled from standard subassemblies. Units are now built and fully tested on their own mini skids before being lifted into the main skid, which has room for only a few workers.

“The critical thing here is about speed to market. That’s what people care about,” Beagan said. “They place an order, they want that gear tomorrow if they could.”

Peacock said data center builders used to wait about 24 weeks for a backup generator to arrive after ordering it. The wait grew to a year and now runs to 18 months. Circuit breakers, delivered in four to six weeks today, could soon take 10 to 12 weeks.

“There is going to be another supply crunch,” he said. “Particularly in Europe, there’s going to be an avalanche of work.”

Since the pandemic, supplier relations have turned into partnerships, and Schneider Electric, ABB and Siemens have set up stock hubs in Northern Ireland, he said.

TES was founded in 1999 by Brian Taylor and Noel McCracken to build motor control centers (MCCs) for water utilities.

It grew into a contractor that designs, installs and commissions the mechanical, electrical and control systems of water treatment works, leaving civil engineering to partners. Over the past few years, it has transformed itself into a maker of modular switchgear and power rooms, spinning off its data center work as TES Power.

It now aims to apply its water pumping experience to data centers, which need their own firefighting and cooling water systems.

French electrical group Legrand acquired TES in April from private equity firm Foresight Group. On the same day, it also bought Keydak, a data center rack maker based in Guangzhou, China.

TES then had about £72 million in revenue and around 300 staff, with data centers accounting for about half its sales.

Europe’s leasing record

The comments came during a media tour of TES Group’s sites in Northern Ireland on September 22 and 23, organized by TES. Journalists visited TES Water’s headquarters in Cookstown, County Tyrone, and TES Power’s campus in Ballykelly, County Derry, to see how the group serves the data center and water markets.

David Sandars, regional director for Europe, the Middle East and Africa (EMEA) at datacenterHawk, presented the industry outlook. The Dallas, Texas-based market intelligence firm was acquired by S&P Global on September 1.

“The first half of this year has been the most incredible and unprecedented six months in the data center industry in Europe,” Sandars said. “The market has grown in six months more than it has ever done in previous years.”

Leasing hit a record 678 megawatts (MW) in the first quarter of 2026 and 1.6 gigawatts (GW) in the second. The half-year total of 2.2GW beat every full year from 2023 to 2025.

The second quarter also brought Europe’s first deal above 300MW.

Planned capacity jumped from 58GW to 66.8GW in three months, though Sandars said there may not be enough energy to deliver it all.

Growth is shifting away from the FLAP-D hubs of Frankfurt, London, Amsterdam, Paris and Dublin, which absorbed under 200MW in the second quarter. The Nordics and Baltics absorbed more than 700MW in the same period.

“The FLAP-D markets are slowing, whereas the traditional tier two markets, the Nordics and the Baltics, are accelerating,” Sandars said.

He said new capacity in those markets is due between 2028 and 2029, against about 2030 in the FLAP-D hubs. Norway’s market has grown almost 70% a year since 2020, with most of that growth coming after ChatGPT launched at the end of 2022.

TES Power is chasing that demand from a former military base on the shore of Lough Foyle. The 770-acre site was run by the Royal Air Force (RAF) and then the British Army, and TES moved into its 1940s aircraft hangar four years ago.

“We came into this derelict building with no staff, no nothing, and just built this,” Beagan said. “It has been a very fun four years.”

TES has used the whole hangar, about 300,000 square feet, since early 2025. It fits about 60 pod-and-skid builds at once, and the company calls it Europe’s largest cantilever building, with no pillars in the way of 18-meter skids.

Gary McLorn, director of business development and marketing at TES Power, said turnover has grown from zero to about £80 million in a few years, all of it exported.

He said Northern Ireland’s post-Brexit status gives TES access to both the UK and European Union (EU) markets without customs checks.

“Next year we are set to more than double. Nothing is set in stone. There’s a very high probability that we double again,” Beagan said.

TES Power avoids hyperscalers and targets colocation operators only.

“Hyperscale would just swallow us up,” McLorn said. “We don’t want to work for one company, and we don’t want hundreds of companies either. We want five to 10 really good customers that want to scale and we can scale with.”

The company plans to start building a 600,000-square-foot expansion at Ballykelly by 2027.

The shift to DC

Beagan said switching from alternating current (AC) to direct current (DC) at rack level brings two advantages. Power conversion can move outside every rack, and conductors can shrink.

“In an AC system, you’ve got your fluctuating voltage and magnetic field, and you create what’s called the skin effect in conductors,” he said. “With DC, that eliminates that completely, so now you’re getting to utilize the full conductor.”

He said a hyperscaler he worked with 14 years ago shelved a low-voltage DC project at the last minute because nothing was driving demand. That driver now exists.

DC is now on TES Power’s research and development (R&D) roadmap, and part of the hangar will host a new R&D center.

“In the next three to five years, you’ll see some people adopt 800-volt DC totally, and you’ll see some hybrid designs,” Peacock said. “Nobody will want to be the guinea pig. But eventually one of the hyperscalers will adopt a staggered approach.”

He said solid-state transformers and breakers, now in early development with suppliers such as Schneider Electric and ABB, will eventually appear in TES panels.

The skids on the hangar floor also use nickel-zinc batteries from ZincFive instead of lithium.

“It doesn’t have the same fire properties as lithium, so you can store the batteries beside the machinery without having separate fire compartments,” McLorn said.

Asked whether the boom could burst, Peacock said construction is always lumpy, but he could not see data center demand stopping in his lifetime.

“Data centers are now a piece of global infrastructure. The world cannot survive without them,” he said.

TES Power’s growth also depends on people. It has created about 200 jobs in two years and expects to double that within the next couple of years, McLorn said.

Training that workforce is its own challenge. TES Power has about 30 apprentices and took 16 recruits onto a new apprenticeship course at North West Regional College this year, though Beagan said trained commissioning engineers are often poached soon after.

Founders Taylor and McCracken were apprentices themselves. TES now recruits at universities, colleges and even primary schools, and is trying to attract more women.

Water follows the servers

TES Group runs TES Power and TES Water as two wholly owned divisions, and the water division is now following the power division into data centers.

“The water utility companies are imposing the same thing on the data centers,” said Mark Geddis, pre contracts director at TES Water. “If you’re building a data center that’s going to use a lot of water for cooling purposes, basically build your own treatment plant.”

New sites now harvest rain from roofs and car parks into tanks and treat it on site.

McLorn said visiting data center customers had long asked TES to build their water systems too.

About 18 months ago, TES took its first such project, for PolarDC in Norway. It built potable water and firewater pumping pods, with storage tanks and diesel pumps for sprinklers and hydrants, designed to carry heavy snow loads.

“We were able to use our supply chain in the water industry to deliver this project faster than what we could with data center suppliers,” McLorn said.

Water demand is rising at home too. UK water utilities spend in five-year asset management plan (AMP) cycles regulated by Ofwat, and TES Water’s clients include Thames Water, Anglian Water, Northumbrian Water and Uisce Éireann.

“The current AMP cycle, which started last year, was roughly double the previous AMP cycle in terms of spend, and the next one’s going to be about double again,” McLorn said. “The volume of work that has to be done is going through the roof.”

The shortfall is already biting in Northern Ireland, where Geddis said most treatment works are at capacity.

“Housing development within Northern Ireland has largely stopped, because Northern Ireland Water are saying we don’t have room,” he said.

Barriers to entry stay high. Contractors need at least 80% in an annual two-day audit to bid for work, and some frameworks now demand 90%. TES scored 100%, McLorn said.

The group’s biggest recent test is at Ringsend, the landlocked plant that treats Dublin’s wastewater.

The upgrade lifts capacity to 2.4 million population equivalent within the existing site, while the plant stays live. Geddis said its 24 double-stacked treatment basins, each about 65 by 45 meters, are being converted to aerobic granular sludge (AGS), a Dutch technology licensed by Royal HaskoningDHV. TES retrofitted four of them.

“The concrete’s moved a little, so the walls aren’t nice and straight and the floor is not nice and level,” McLorn said. “With a laser scan survey, we were able to find all the movements, so it goes to site, and it fits the first time.”

TES’s contract, part of the FaCT3 consortium, was worth €45 million of a €54 million total and ran from 2021 to 2024.

McLorn said more than half of TES’s share went on metalwork, and that TES models every project in building information modeling (BIM) whether or not the client asks.

With the next AMP cycle expected to double spending again, TES Water is now talking to more data center operators as it looks beyond the utilities.

Akamai Technologies, a Cambridge, Massachusetts-based cloud and cybersecurity company, announced on September 24 that it signed a seven-year, $11.6 billion agreement to support Anthropic's growing central processing unit (CPU) workloads, the developer of the Claude artificial intelligence (AI) models.

The commitment could rise by a further $9 billion, lifting the relationship’s potential value to about $20 billion.

Anthropic will tap Akamai Cloud’s distributed AI infrastructure and software to meet its accelerating CPU workload demands. The contract adds to more than $2.8 billion in multi-year cloud infrastructure services (CIS) commitments that Akamai announced earlier this year across its customer base.

Tom Leighton, co-founder and chief executive of Akamai, said Anthropic was advancing the AI revolution and had chosen his company’s capabilities to build and operate AI infrastructure at scale.

“Our expanding global footprint, combined with our years of experience serving the world’s largest enterprises, positions us to be the infrastructure provider for secure and responsible AI applications and workloads,” he said.

Akamai Cloud supports computing from the core to the edge on a network spanning thousands of points of presence. Built on diversified hardware, it lets customers build and run applications and optimize how they serve users and AI agents.

It is designed to keep applications fast, reliable and secure across their full lifecycle, as customer demand for deploying AI workloads grows.

Akamai has also given Anthropic the right to buy up to about 5% of its shares at $111.33 each. About 2% is tied to the initial commitment. Anthropic can unlock the remaining 3% in 1% steps for every additional $3 billion it spends on Akamai’s cloud services over seven years.

Akamai estimates capital expenditure tied to the contract at about $5.5 billion. It expects no impact on its 2026 revenue guidance but plans to raise this year’s capital spending by about $1.7 billion to secure and pre-purchase critical supply chain components, including memory.

Guarding nonhuman identities

The deal came two days after Akamai published its latest State of the Internet (SOTI) security report on September 22, which warns that agentic AI is fundamentally changing enterprise risk.

Titled The Agentic Frontier: Speed, Scale, and Nonhuman Identity and aimed at chief information security officers (CISOs), the report said securing a business has evolved beyond managing people's identities and access to behavioral governance over nonhuman entities that execute tasks across application programming interfaces (APIs) and systems.

It identified two major challenges for CISOs. One is setting guardrails for autonomous agents. The other is how quickly AI models discover and chain vulnerabilities, often outpacing human patch cycles.

Key findings include:

• Model Context Protocol (MCP) exposure ranks last among CISO security priorities, even though the protocol lets AI agents act autonomously across multiple software systems and security leaders expect rogue agents to become a top cyberthreat by 2030

• More than 6% of enterprise AI chatbot conversations contain sensitive corporate data, mainly personally identifiable information, with 47% of interactions taking place via unmonitored personal accounts

• More than 40% of enterprise users have installed AI-powered browser extensions, a quarter of which changed their permissions within 12 months; these tools are 60% more likely than standard extensions to carry known common vulnerabilities and exposures (CVEs)

• AI bot traffic rose 300% in 2025, with the commerce industry hit hardest

• The average annual cost to an enterprise of resolving API breaches has reached $700,000

Boaz Gelbord, chief security officer at Akamai, said companies face a triple threat from internal AI use, such as AI-generated code and productivity tools; AI built into customer-facing products and cloud workloads; and AI-driven attacks on the enterprise.

“AI presents an ‘everything, everywhere, all at once’ moment for the security ecosystem as a whole. Security programs need to adapt to this rapidly evolving reality, and there’s a lot of pressure in the system due to the unprecedented speed of these changes,” he said.

The report urges CISOs to:

• Deploy edge-native runtime protections, API filters and isolation to neutralize vulnerabilities while back-end patching takes place

• Build visibility and behavioral controls inside the browser to secure staff use of unmanaged AI extensions and web-hosted models

• Use generative engine optimization (GEO) and machine-readable data at the network edge to stop AI crawlers and synthetic shoppers misrepresenting brand data

• Grant AI agents autonomy according to how easily their actions can be verified and how reversible a failure would be, keeping humans in the loop for high-stakes actions

Gelbord said the issue will be a central topic for boards, customers and regulators for the foreseeable future.

The fusion industry’s biggest financing obstacle is not a shortage of capital but a shortage of proof. Insurers and lenders will not back a first-of-a-kind power plant until someone has shown, in hardware rather than theory, that the technology behaves as promised.

That distinction between science risk and engineering risk is reshaping how developers court financiers. Banks are drawing a sharp line between backing the companies building fusion technology and backing the projects that will eventually generate power, and only one of those two categories is ready for debt.

Chris Mowry, chief executive officer of Type One Energy, said that bar is different from the demonstrations fusion has already delivered.

A test that proves fusion happens, such as the laser-driven ignition achieved at the US National Ignition Facility (NIF), is not the same as proving a specific reactor design can run reliably inside a power plant.

“What insurance and finance need to see is technology demonstration, something close enough to what you’re actually going to do in a power plant that you can say the technology risk is bounded and therefore you can price it,” he said. “If you can’t do that, you can’t get insurance, and without insurance, you’re not going to get it financed.”

Financiers apply a similar distinction from the banking side. Iain Smedley, global chairman of banking at Barclays, said equity investors already back early-stage fusion developers, but debt financing, the kind needed to build an actual plant, will not arrive until a technology has moved past a single successful attempt.

“You have to differentiate between equity finance and the debt finance we need for projects,” he said. “You need to have got beyond first-of-a-kind for it to be structured in a way that finance will be there, without explicit backstops from government, because until you’ve got there, you’re not going to get the insurance market behind a true first-of-a-kind.”

Financing challenges are seen as largely a product of the underlying technical, engineering and business hurdles the industry still faces, rather than a shortcoming in the finance industry itself.

He said the market’s next signal to watch would be another major funding round for a fusion developer.

“I’d like to see another landmark capital raise in the equity markets for a fusion player, and I suppose the most likely place for that is in the US. But I wouldn’t mind seeing an equivalent investment here,” he said.

Both bankers and developers agreed that closing that gap means shifting risk away from untested parts of a project and toward participants with a track record.

A financeable blueprint

The comments came at the UK Fusion Forum 2026 in London on September 15, organized by the Fusion Industry Association.

The panel, moderated by Jennifer Jordan-Saifi, chief executive officer of the Sustainable Markets Initiative, examined how the UK Infinity Fusion Consortium plans to fund and build a commercial fusion power plant.

The consortium was formed by Type One Energy, Tokamak Energy and AECOM to combine an American reactor design with British and American engineering and manufacturing capacity. It centers on Project Infinity, a planned commercial plant in Tennessee that the Tennessee Valley Authority (TVA) will operate.

Mowry said the consortium is designed to remove risk layer by layer rather than ask investors to underwrite an unproven project outright.

“Having TVA as an operator effectively takes operating risk off the table, and having confidence that you have an operator with the experience, the capabilities and the track record adds comfort to anybody thinking about putting equity into a project,” he said.

Warrick Matthews, chief executive officer of Tokamak Energy, said the company has shifted its own strategy to fit that model. Rather than building its own plants, Tokamak Energy now focuses on high-temperature superconducting magnet technology and supplies magnet systems to programs including the UK government’s STEP (Spherical Tokamak for Energy Production) program.

“Ridgway makes machines that wind the cables and wind the magnets, and if you are anchoring that order book, there are different financing routes open to you,” he said. “When you have that certainty of demand, you can move into debt financing and project financing.”

Jennifer Obertino, global energy practice leader at AECOM, said the firm’s role is to make the design scalable rather than bespoke to one site.

“Type One is advancing the technology, and with the agreement they have with TVA, they are a fantastic partner,” she said. “We’re able to bring the engineering and the power plant design, the civil site, to make this a scalable technology we can take from the US to the UK.”

Panelists said interest from artificial intelligence (AI) data center operators is beginning to change how that financing case gets made.

“We saw it just last week, the Finnish utility announced a deal with Google for a large-scale power purchase agreement, and the share price went up 10% on the day,” Smedley said.

Mowry said hyperscale companies are almost uniquely positioned to pay first-of-a-kind electricity prices because their profit margins are large enough to absorb the premium, though they have historically planned only 18 to 24 months ahead.

“It’s easy to get an offtake agreement that’s just a piece of paper, but something with real teeth requires a belief that you have a pathway to deployment, partly from the technology side and partly from the supply chain,” he said.

Educating the market

Smedley said much of the caution around fission is misplaced when applied to fusion. He said the risks investors actually worry about in fission projects are civil engineering and construction risk, plus the risk of policy reversal, rather than anything specific to nuclear technology itself.

“A lot of the risk people worry about with fission isn’t really nuclear technology risk. The risks we spend a lot of time on are fundamentally civil engineering and construction risk, along with risk from government policy change and discontinuation,” he said.

Mowry drew the same line from the developer side.

“If this were nuclear fission, we’d be having a very different conversation. You’re probably not able to finance a first-of-a-kind fission project without a significant government backstop, because of nuclear liability and a track record of construction delays driven by regulation,” he said.

Both said closing that understanding gap, for insurers, investors and the public alike, would happen faster once fusion has a small number of working projects to point to rather than more explanation.

Matthews framed the stakes more broadly.

“When you look at what’s going on around the world, most of the common denominators of conflict are energy,” he said. “The biggest prize is not just that we need more baseload dispatchable power, but the role this plays in a more peaceful world by democratizing power-dense systems for electrification.”

Mowry pushed back on comparing fusion’s moment to the rise of SpaceX, a parallel raised earlier in the discussion.

“SpaceX isn’t necessarily a good analogy. They didn’t invent the product, and there was zero science risk in getting into space. They put together a system as an integrator, but in the way fusion is first-of-a-kind, that’s not a good reference case for what we’re trying to do,” he said.

Matthews said the industry is moving away from developers trying to build every component themselves.

“Type One is leading in that system-level thinking as a private company, on the public-private partnerships,” he said. “You can’t be vertically integrated to do something this hard perfectly across all the technologies, so you want the best system provider for magnets, diagnostics, and a big engineering company to help put it together.”

Obertino said the model also depends on identifying an operator and an end customer early in the process, rather than treating those roles as afterthoughts once the technology itself is settled.

“It’s bringing in who is going to operate and who is going to be the end client, making sure what we’re designing is able to be utilized and will, at some point, be one of the cheapest options,” she said.

The next test comes as the Tennessee project moves toward construction, with the UK consortium aiming to repeat the model at home.

A commercially viable fusion market will hinge on supply and demand, not on scientific breakthroughs alone.

Reaching that balance means building supply chains capable of delivering hardware at scale while lining up buyers willing to pay for power before the technology has proven it can run without subsidy.

Data centers, where the cost of energy is less of a barrier, are seen as fusion’s likely first paying customers.

“You need the supply chain to be able to build what needs to be built, and you need the demand there to enable the supply chain to work effectively,” said Adam Baker, head of fusion policy at the UK’s Department for Energy Security and Net Zero (DESNZ). “It’s just pure economics, but it’s far more complicated than that. Otherwise none of us would have jobs.”

“Harmonization is what’s needed, including supply chain, technology, finance and cooperation between public and private players,” said Kiyoshi Seko, president and chief operating officer of Kyoto Fusioneering. “Everything has to work together instead of competing with each other. That’s the stage the fusion industry is at, and we have to recognize that.”

Scott Hsu, fusion partner at Lowercarbon Capital, said he and colleagues at the U.S. Department of Energy (DOE) once calculated that fusion would need to hit $50 per megawatt hour (MWh) to compete in most power markets, in a paper published in 2021.

He said the goalposts have since moved. Energy costs are higher, and clean, firm electricity now carries a bigger premium.

“You need to get into a market first in order to have an opportunity to start going down the cost curve,” Hsu said. “The industry should be very attuned to what the market opportunities are, and also to geographical differences.”

“At the beginning it would be more rational to have a mixed industrial supply to data centers, for instance, where the cost of energy is not necessarily the first concern,” said Francesca Ferrazza, global head of magnetic fusion initiatives at Eni. “Different schemes could follow depending on the cost of energy in different places, and that should be chosen carefully.”

The view represents a rare point of alignment among developers, suppliers, financiers and regulators occupying very different positions in the fusion value chain.

Public demand as validation

The comments came during a panel titled “Unlocking a Sector: Policies and Market Models Enabling Fusion” at the UK Fusion Forum 2026, hosted by the Fusion Industry Association in London on September 15.

The session was moderated by Steve Lawler, UK director of Helixos and chair of the Nuclear Industry Association’s Fusion Business Group.

The panel examined the policies, market structures and financing needed to attract private capital into commercial fusion projects. Lawler said unresolved questions remain around financing, insurance, supply chains, skills and testing facilities, even as the sector’s own message is that the UK is open for fusion business.

Hsu said the clearest example of public attention translating into political weight came in 2022, when scientists at the U.S. National Ignition Facility (NIF) achieved fusion ignition for the first time.

“Continued execution against truly significant de-risking milestones toward a fusion pilot plant is what’s going to move everything,” he said. “When NIF achieved ignition, nothing was more effective at getting worldwide attention on fusion. Suddenly people’s mothers and grandmothers were asking about it.”

“We don’t exist in a bubble. In democratic societies, it’s the public who sets the priorities,” Baker said. “What solves that isn’t big cross-party agreements, it’s public demand and public input. That’s what speaks to governments, not the promise of a long-term silver bullet, but jobs, regional investment and new skills.”

Kyoto Fusioneering announced a commitment of up to £130 million the day before the panel for what it calls Project Albion, a three-year effort to build breeder and blanket manufacturing capacity and a domestic supply chain in the UK for the UK Atomic Energy Authority (UKAEA)’s Spherical Tokamak for Energy Production (STEP) program.

“STEP needs breeder and blanket manufacturing capacity and a domestic supply chain, and someone has to build those capabilities, along with the workforce, here in the UK,” Seko said.

“There are still many missing technologies and missing infrastructure, and someone has to fill that gap to deliver fusion power at commercial scale,” he said. “That’s the contribution Kyoto Fusioneering is making, a commitment to the entire fusion industry.”

He said the company is still weighing whether to share supply chains globally or build them separately in each country where it operates, including the United States, Japan and the UK. He said the fusion market is not yet large enough to justify duplicating investment everywhere, so priorities have to be set.

Capital and government roles

“The investors’ greatest concern, what keeps us up at night, is if our companies run out of capital,” Hsu said. “That’s the existential problem. No matter how good the technology or how talented the staff, if they run out of money, it’s pretty much game over.”

He said the amount of non-dilutive funding, meaning grants and awards that do not dilute a startup’s equity, has to scale with each company’s stage, since capital needs vary widely across the sector.

“For a company like Commonwealth Fusion Systems (CFS), built on decades of published tokamak research, that validation matters less,” Hsu said. “For concepts that haven’t had as much public support, milestone validation from investor diligence is extremely valuable.”

He said the arrangements also signal to the wider market that policy and demand both exist, which helps attract customers.

Lawler opened the panel by citing a joint US-UK statement, signed the previous day, in which the two countries committed to harmonize fusion regulation. He also cited a DESNZ investment prospectus that values the global fusion market at between £3 trillion and £12 trillion.

“Regulation is an example of a role government has to take, because no one else will,” Baker said. “Our job is to create the environment in which the market can flourish.”

He said the sector still has to work out who bears the cost of decommissioning fusion plants at the end of their life, and whether companies have adequate access to capital.

He pointed to the National Wealth Fund, the British Business Bank and the government-seeded Starmaker One fund. He added that the biggest single thing that would move the dial in the UK is deeper utility involvement, not just in funding but in expertise and capability. Asked to picture the sector a decade from now, he had one image in mind.

Ferrazza said Eni’s experience varies sharply by region, shaped by differences in market understanding, technology maturity and regulatory openness across the US, the UK and Europe.

“We saw the future in the US investing in CFS,” she said. “We saw the dynamic and the potential, and it moved very fast and very well.”

Eni has also maintained a roughly 20-year relationship with the Massachusetts Institute of Technology (MIT), is listed on the U.S. stock exchange and has other long-running UK activities including carbon capture and storage (CCS) and offshore wind.

“In the UK we have a very solid relationship with UKAEA,” Ferrazza said. “The experience we built there with JET (Joint European Torus) is so solid that we fitted in very well. We’ve been in the country as a company for about 60 years.”

“For fusion, [Europe] was a very scientifically fragile community, and it’s now evolving into a more industrial approach,” she said. “It’s picking up, and I hope it picks up fast, because we need the Western fusion system working.”

She said continued investment depends on predictable regulation and sustained political support, and on raising public awareness.

With Project Albion underway and government funds like Starmaker One now live, the industry’s next test is whether that capital arrives fast enough to get first-of-a-kind fusion plants connected to the grid this decade.

UK and US government agencies are rewriting how they fund private fusion companies, replacing straightforward grants with joint ventures and milestone-based funding.

In the UK, that has meant new joint ventures between the national fusion authority and private companies. In the US, it means tying federal funding directly to technical milestones companies have to hit before they get paid.

“When we talk about public-private partnerships, we’re talking about trying to do something more specific together, where both parties bring something different to the table and try to construct something new,” said Heather Lewtas, interim chief development officer at the UK Atomic Energy Authority (UKAEA). “You’re actually working with a goal, and together you’re trying to do it.”

UKAEA launched two joint ventures over the past year built on that principle. One, with French utility EDF, pairs UKAEA’s tritium expertise from the Joint European Torus (JET) with EDF’s industrial manufacturing experience to commercialize the fusion fuel cycle. A second, with TAE Technologies, combines UKAEA’s neutral-beam experience from JET with TAE’s own intellectual property.

“We’re not industry, and we shouldn’t be trying to do that part,” Lewtas said. “But together, both sides bring what the other doesn’t have, and we have a much better chance of commercializing.”

Jennifer Arrigo, senior advisor at the US Department of Energy’s (DOE) Office of Fusion, described partnerships within the department as running along a spectrum. At one end is a solicitation with a private sector cost share. At the other are deals shaped from a shared need with every actor at the table from the start.

“What we’re moving towards are true partnerships, where there are deep value propositions on both sides, and it’s a process of shaping the partnership together, starting from a shared need,” Arrigo said.

Linda Werner, head of fusion at St1, a Nordic energy transition company whose fuel stations are a familiar sight across the region, said that starting point matters because it signals commitment before a project takes shape.

“Having a partnership between the private and public sector as a starting point shows that this is a priority,” Werner said. “That’s the first signal we want to see, that both parties are engaged and committed.”

She said industry also carries its own obligation once a partnership begins, taking on commercial and industrial risk and translating a shared roadmap into steps both sides can explain to each other.

“We have a big obligation in these partnerships, taking on industrial and commercial capital and working out how this will function in reality,” she said. “Getting a common language, and dividing the responsibilities and roles, is a job in itself.”

Funding only on milestones

The panel was held at the UK Fusion Forum 2026 on September 15 in London. Organized by the Fusion Industry Association and sponsored by UKAEA, the session was titled “Public-Private Partnerships: Catalysing Growth” and moderated by Valerie Jamieson, fusion cluster development manager at UKAEA.

Arrigo described the DOE’s own answer to that challenge in fusion, a milestone program that ties funding directly to technical progress.

“Our keystone public-private partnership program is the milestone program,” she said. “Private companies propose technical milestones, and they only get the DOE share of the funds once they meet that milestone.”

Setting it up required DOE to re-examine its own legal authorities and processes, including how it handles intellectual property. Arrigo said the lesson was that good faith on both sides matters as much as the paperwork itself.

“One lesson learned is that it takes longer than we expect once you start writing down those details,” she said. “It’s important to have good faith on both sides to get to an appropriate outcome.”

Arrigo said the value had to be clear on both sides of the deal. For DOE, that meant supporting multiple technical approaches across the US fusion industry. For companies, it meant earning a DOE stamp of approval through the negotiated milestones.

She said an unplanned benefit of the program was the technical expertise the milestone review boards brought to companies, giving them direct access to national laboratories, universities and international collaborators.

“National labs are a national asset, the crown jewel of our research ecosystem,” she said. “They are the ones out there every day, engaging with companies and identifying the technical opportunities.”

That funding model has not solved every friction point. Robbie Scott, chief scientist at the UK Programme of Laser Inertial Fusion Technology for Energy (UPLiFT) at the Science and Technology Facilities Council (STFC), said UK government procurement remains one of the clearest obstacles to faster deals.

“UK government procurement rules need to be revised. It’s a nightmare trying to buy anything in the public sector,” Scott said. “Revising that would be very helpful for big projects like STEP (Spherical Tokamak for Energy Production), and for us.”

Lewtas said UKAEA’s own early joint ventures bore the brunt of that learning curve, working out which government checks and approvals were required before later deals moved faster.

“Don’t underestimate how quickly we are learning how to do this properly and at pace,” she said. “Next time, we knew when to ask and what we needed to agree upfront, and it sped things up.”

Arrigo said DOE faces a similar internal challenge as it shifts from funding research to supporting commercialization directly.

“We need to rethink where we are in terms of the transition from research to commercialization,” she said. “DOE has a lot of experience with commercialization in other parts of the department, so it’s about rethinking the processes we use for deployment.”

From lab to industry

Scott said the split between public and private responsibility becomes clearest once a concept has been proven and needs to scale.

“The role of national labs, and potentially universities, should be to develop the technologies to prove the concept,” he said. “After that, it should be sent into industry, one way or another.”

He pointed to the US National Ignition Facility (NIF), a laser fusion project run by the government’s Lawrence Livermore National Laboratory, as the model he wants the UK to follow for its own inertial fusion program.

“NIF is a good example of this. They spent 50 years developing laser fusion from concept to getting Q greater than one,” he said, referring to a reaction that produces more energy than it consumes. “Instead of giving Livermore a load of money to develop inertial fusion energy itself, there was a scheme to encourage private companies to work with the national lab. That’s a really good model.”

Scott, whose own UPLiFT program develops UK laser fusion technology, said he wants that model applied closer to home, tied to a UK strategic need that already has government backing.

“I would like to see the UK investing in power-level inertial fusion,” he said. “The UK has strategic defense needs that likely involve high-energy lasers, similar to NIF, so working across government could develop national fusion capability while meeting those needs.”

Asked what these partnerships will have made possible by 2036, panelists gave four different answers:

• A cross-concept supply chain serving every fusion technology, not just one, according to Lewtas

• Q economic greater than one, the point at which fusion becomes commercially viable, according to Scott

• A Nordic role in the wider fusion supply picture, according to Werner

• Electrons reaching the grid, which Arrigo said would not happen without both sectors working together

Lewtas said the clearest sign of a working partnership is when it removes a problem from a fusion developer’s list entirely.

“Have we taken things off the plate of the fusion developers, so that they do not have to worry about it, that they know it can be there when they need it,” she said. “Has TAE UK made it so that if you need a neutral beam, you can buy one, no problem, you don’t have to worry about it?”

DOE’s milestone program is now being tested against its next phase of commercialization funding, while UKAEA and its partners continue expanding the joint ventures built over the past year.

Fusion developers are pressing governments to move faster, warning that 10-year roadmaps no longer match the technology’s pace.

The argument centers on public-private partnership as the structure best suited to absorbing fusion’s enormous cost and technical risk, with backers pointing to past state-industry collaborations that delivered breakthrough technologies once dismissed as decades away.

“We look at public-private partnership as being the path,” said Jonathan Toretta, chief executive of TAE Global GmbH and chief revenue officer of TAE Technologies Inc. “Fusion is extremely hard, and this is an endeavor not for the modest.”

“Roadmaps out 10 years on paper make me nervous, because the urgency around what we’re trying to deliver is greater than a 10-year timeline,” Toretta said.

He said fusion is capable of reshaping economics, the defense sector and global geopolitics, while also serving as a way to lower energy costs and shorten supply chains worldwide.

Building that sense of urgency, he said, was the second half of what he wants from the public sector.

He compared the dynamic to the US space race, calling public-private collaboration one of the few examples of the two sectors working materially together at this scale, adding that TAE’s own five-generation machine program reflects that model in practice.

He described the goal simply as putting a pin into decision-making that too often treats fusion as a distant bet.

He said the fusion industry is still forming its own market structure, spanning a supply chain, consumers, producers and ancillary businesses that will take time to coalesce, and that TAE deliberately treats other fusion developers as allies rather than rivals because the sector cannot afford internal competition given how difficult the underlying science remains.

He said the biggest mistake in the industry is when one company criticizes another, given how seriously difficult the underlying science remains.

“Nobody in the fusion industry is here by mistake,” he said.

“I do expect consolidation, radical success and radical failures,” he said. “Some of the presumed winners will be losers, and some of the presumed losers will be winners.”

TAE Technologies, founded in 1998 as Tri Alpha Energy and based in Foothill Ranch, California, is one of the longest-running private fusion developers, pursuing an approach that aims to burn hydrogen-boron fuel without the radioactive byproducts associated with conventional reactors.

The company has raised more than $1.2 billion in private funding since its founding, according to public records.

Betting on Culham

The exchange took place at the UK Fusion Forum, held in London on September 15, 2026 and organized by the Fusion Industry Association (FIA). The session, moderated by David Capper of the UK Department for Energy Security and Net Zero, examined how public-private partnerships are shaping the UK’s fusion sector.

TAE’s UK presence dates to an acquisition years ago in the West Midlands, its beam technology base, later paired with a move onto the Culham campus in Oxfordshire, home to the UK Atomic Energy Authority.

“We already had 100 people in the West Midlands,” Toretta said. “Right place, right people, right technologies, and we expect to build out that supply chain and add labor to the workforce there.”

“We have brought together 40 years of experience at the Culham campus with 25 years of beam technology expertise in the US that we felt practically belonged here,” he said.

The Culham move became TAE Beam UK, a joint venture with the UK Atomic Energy Authority announced in December 2025, under which the authority is investing £5.6 million to develop the technology for fusion, cancer treatment and other applications.

Toretta said Southern California’s venture-heavy culture, where TAE has been based since 1998, is not suited to the company’s next, fully commercialized phase, and that a site needs PhD-level researchers, hundreds of megawatts of power and heavy water infrastructure concentrated in one location, something he called difficult to replicate anywhere in the world.

“The labor pool that we are accessing on that campus is highly unique, and we find it very difficult to replicate anywhere else,” he said. “A nod to the UK, you see that happening here. The UK may be in front.”

Toretta said fusion remains TAE’s ultimate goal, but the company has also built two revenue-generating businesses alongside it: EmFi Technologies, a UK-based life sciences unit with operations across Europe, the US and Asia, and a power management business he described as having large addressable markets.

“TAE has a north star in delivering fusion, no doubt,” he said. “But the company has also built technologies that are intrinsic to its growth.”

“We’ve built TAE Life Sciences, a company with operations in Europe, the United States and Asia that is revenue generative, and also a revenue generative power management business that has large addressable markets and notable revenue,” he said.

He said the life sciences business shares a production facility in part with an Italian partner also attending the forum.

Toretta said most of TAE’s workforce is now international, split between the US, Europe and Asia and built around recruiting specialists wherever they are strongest, pointing to smaller specialist offices in Italy, the Czech Republic and Slovenia alongside its larger UK operations.

“More than 50% of our employees are not Americans,” he said. “We will recruit people from anywhere in the world who are the right people to help us de-risk fusion technology and the derivative technologies we build.”

He declined to disclose specific revenue figures for either business but said they represent a meaningful differentiator for TAE with investors and government backers.

Splitting fission from fusion

Toretta said he wants a clear regulatory separation between fission and fusion, arguing the physics carries different risks.

“I would really like to see a true separation of fission and fusion,” he said. “You can site a TAE reactor in a city or next to a population, and that should not be viewed the same way as a radioactive neutronic fusion system.”

“Continental Europe should really catch up in this regard,” he said.

Toretta said interest in the sector has grown sharply since he attended his first fusion industry gathering in 2018 or 2019, when he recognized nearly everyone in the room within minutes.

He said the rise in attendance and public coverage since then reflects the field’s shift toward a mainstream investment case.

TAE’s technical case rests on five generations of machines built since 1998, using a field-reversed configuration (FRC) design that lets the company switch fuels rather than being locked into one. Toretta noted that Helion, another fusion developer that spoke earlier the same day, has described building seven generations of machines in 12 years.

The latest machines operate in temperature regimes up to 70 million degrees Celsius, Toretta said.

“We’ve been around since 1998,” he said. “I think we’re the longest-standing fusion company, certainly at the commercial level, but maybe in the world.”

“These are FRC-focused machines that are fuel agnostic,” he said. “We can burn DT [deuterium-tritium], deuterium-helium-3 and hydrogen-boron, the three fuels targeted by the industry.”

Toretta said he expects fusion’s timeline to keep beating expectations, pointing to how quickly net energy gain has moved from a distant prospect to an active discussion.

“My sister is an MIT nuclear engineer,” he said. “She said one of the most arresting things in her own life is that fusion was studied as a 2050 or 2060 event in the early 2000s, and she’s shocked there’s now discussion of net energy gain in the 2030s.”

“We hope that we deliver net positive energy by the early 2030s,” he said.

Toretta, whose career in the oil and gas sector dates to 2007, before he joined TAE, said he sees the same pattern of underestimated progress playing out across other fields, pointing to supercomputing, drone technology and AI as areas that advanced faster than expected.

“We’re going to be equally surprised at what we see in nuclear, in both fission and fusion gains,” he said.

TAE is targeting net positive energy in the early 2030s while continuing to expand its life sciences and power management businesses alongside its core fusion program.

Fusion’s private sector in Europe has shifted from small research teams into fast-scaling industrial operations within a single year, as funding, hiring and public financing have all accelerated sharply across the continent.

Companies building lasers, magnets and reactor components say the past 12 months mark a turning point after years spent proving the underlying physics. Staff numbers have tripled at some firms, funding rounds have grown far larger and governments have begun committing hundreds of millions of euros to the supply chain behind the technology.

“Last year in September, we were 60 to 65 people. Today, we’re above 200, nearly 240, and we have 60 open jobs,” said Guenter Kraft, chief communications and government affairs officer of Focused Energy.

“The public financing is falling into place. The private financing is coming in as well. The partnerships, both on the public and the private side, are there,” said Lucio Milanese, co-founder and chief external affairs officer of Proxima Fusion. “Now we just need to deliver.”

Milanese said Proxima’s funding more than tripled over the past year, backed by a €400 million financing deal with Bavaria for its Alpha demonstrator.

Proxima also secured further German federal funding and signed a separate agreement with utility RWE. Germany’s Hubs for Fusion program is separately distributing an initial €125 million across three centers covering magnetic confinement, laser fusion and fuel-cycle research, further evidence of the country’s fast-moving public support.

Saskia Horsch, director of corporate affairs at Marvel Fusion, said construction began on the company’s US technology demonstrator this year.

Sam Guilaume, chief executive officer of Renaissance Fusion, described the shift toward supply chain work as evidence of an industry maturing beyond pure science. He said the company’s focus on materials expertise, including high-temperature superconductor (HTS) tape and liquid lithium, will combine with engineering and fusion-company collaboration to deliver simplified reactor designs.

“We are living in a Darwinian era to some extent. Access to capital has evolved, access to policies has evolved, and we need to adapt,” he said.

Peter Roos, chief executive officer of Novatron Fusion Group, said political engagement in the Nordic region has grown quickly this year.

Securing the supply chain

The panel discussion took place at the UK Fusion Forum 2026 on September 15 in London, organized by the Fusion Industry Association. Melanie Windridge, founder and director of Fusion Energy Insights, moderated the session, which focused on how private fusion developers in Europe are bringing their technology to market.

Kraft said most fusion technology, whether magnetic or laser-based, already originates in Europe, calling the sector the continent’s next major industrial opportunity beyond energy alone.

“This whole thing is a sovereignty game, a geopolitical game and an industrial game-changer for Europeans,” he said. “We have to make a decision: is it China or is it the Western world?”

Milanese said Europe is already 80% to 90% of the way toward manufacturing all the components needed for a stellarator-based demonstrator domestically, thanks partly to industrial capacity built around the ITER (International Thermonuclear Experimental Reactor) project. The main remaining gaps are in HTS tape manufacturing, concentrated in Asia, and in tritium enrichment, he said.

“We just signed last week a memorandum of understanding with the government of Saxony for an investment of €140 million in the first large-scale manufacturing plant for fusion-grade HTS tape,” he said.

Horsch said sovereignty does not mean total self-sufficiency, since interdependencies are unavoidable in a globally connected world.

“Sovereignty really means making sure that we have control of the critical technologies here, that we know how to build them, keep the know-how here, build the industrial capability and then start to export,” she said.

Kraft said the sector’s next challenge is industrializing production at a scale the optics industry has never attempted, comparing today’s methods to building a satellite by hand.

“We need 1,000 lasers. We need to build gigafactories for laser production. This is the goal,” he said.

Focused Energy spent three years negotiating with RWE to convert a former fission site into a fusion site, a deal that has drawn other utilities’ attention to fusion startups and served as a signal that helped move the wider industry forward.

Milanese said European utilities have historically been less willing than American counterparts to embrace fusion, though RWE’s move has shifted that, he said.

Roos said he expects more utility deals in the coming years, in a different form than seen in the US.

Investors want returns

Horsch said effective de-risking requires two things: a predictable regulatory path kept separate from fission rules that does not overregulate before the sector’s risk profile is understood, and co-funding mechanisms that let private investors back a facility with some prospect of return rather than a pure science project.

“A private investor will not invest in a science project, but a private investor will invest in a facility that will have some sort of return on investment at the end. It does not need to be the final machine,” she said.

She pointed to demonstration facilities with a university-style user model as one option, with the state acting as co-investor. Horsch said Germany has made a good start, and that she hopes the European Union’s forthcoming fusion strategy reflects similar ideas.

Panelists said governments could do more to clarify how public money will support the sector. Milanese said the UK has tremendous potential in talent and stability for business, pointing to its £2.5 billion fusion funding commitment as one example where clarity is still needed.

“We would need, as a private sector, clarity on what the UK really intends to do engaging with the private sector with the funding that is being allocated,” he said.

He said clarity on how the money is spent, whether on factories or financing, could boost cross-Channel collaboration.

Kraft said fusion cannot be solved by any single country, noting that Focused Energy employs people from 40 different nations across its sites.

“Fusion will only happen if we look across the borders, and this is something where the UK should look into Europe and not just inside it,” he said.

Horsch pointed to Marvel Fusion’s partnership with Colorado State University as a model that could work in the UK too.

Guilaume said the UK’s policymaking and public funding commitments have already sent a positive signal likely to be followed elsewhere in Europe, pointing to similar momentum building in France.

The five companies represent different bets on how fusion reaches the grid, with shared testing facilities and transatlantic supply chain cooperation cited as needs across the sector. Novatron’s mirror-machine design addresses the plasma-leakage and stability problems that pushed the approach out of favor relative to tokamaks in the 1980s.

Renaissance Fusion is pursuing magnetic confinement built around a simplified stellarator design, with a fully operational pilot line producing HTS tape and liquid lithium, both targeted for production at scale by 2028.

Horsch said Marvel Fusion’s laser-driven approach builds on the achievements of the National Ignition Facility (NIF), pairing high-gain targets with high-performance lasers and a simplified fuel cycle developed alongside power plant partner Siemens Energy. She said the company used the past year to deepen those industrial partnerships as it moves toward commercial deployment.

With funding, staff numbers and government commitments now moving together, the panelists said the industry’s next test is turning that momentum into reactors connected to the grid.

Private fusion developers are splitting into two camps on how to bring their reactors to market.

Some plan to build and operate their own power plants and sell electricity directly to customers. Others are positioning themselves as equipment suppliers, licensing reactor technology to utilities and industrial buyers instead.

“From the beginning, we’ve had the vision to be the power company,” said Jennifer Ganten, chief global affairs officer of Commonwealth Fusion Systems (CFS). “The first-of-a-kind is the hardest, and you can’t outsource that, so we’re going to build and operate that first plant.”

CFS is vertically integrated, running its own magnet factory beside the assembly hall for its demonstration plant outside Boston. The plant is more than 80% complete and CFS plans to turn it on next year, alongside a separate 400 megawatt (MW) plant that will connect to the grid in Chesterfield County, Virginia.

“We plan to build and operate as well,” said Andrew Proffitt, Helion's regulatory policy lead. “That is the business we are in, to sell electrons to the grid, directly to a customer, whatever that means.”

Helion’s first power plant is already under construction, with permits and licenses in place. Its first two customers are Microsoft, supplied through the grid, and Nucor Steel, supplied directly to its facility.

Matt Miles, senior vice president of external affairs at Type One Energy, has spent nearly a decade in fusion after 12 years in the nuclear fission industry. He said his company is taking the opposite approach as an original equipment manufacturer (OEM), leaving vertical integration to partners.

“Our model is OEM. We don’t plan a lot of vertical integration towards manufacturing, and we rely on companies like Commonwealth for that,” Miles said.

“We’re fusion prime, OEM within enabling technologies,” said Mark Thomas, chief executive officer of First Light Fusion. “We have our own system architecture, but it’s a very deliberate partnership play. Our technology enables machines to be simpler and cheaper, and buildable by people in this world.”

First Light’s approach centers on target amplification technology, physical devices that create extreme conditions for fusion with applications in space and defense. The company launched a fast ignition architecture called Flair last year.

Takaya Taguchi, co-founder and chief executive officer of Helical Fusion, said his company’s business model will change as it matures. For now it is focused on building its own plant, and once that technology is commercially viable it plans to license it for use elsewhere, a strategy he said reflects the company’s limited resources.

Contracts validate the market

The comments came during a panel titled “Private Developers, Bringing Fusion to Market in the US & UK” at the UK Fusion Forum 2026, hosted by the Fusion Industry Association in London on September 15. Melanie Windridge, founder and director of Fusion Energy Insights, moderated the session.

“It provides a validation that there’s a market, and that’s an important validation across a number of stakeholders,” Ganten said of CFS’s power purchase agreements (PPAs). “We’ve sold nearly all the power for that Virginia facility when it comes online, under a power purchase agreement with a price.”

She said those customer negotiations directly shaped the final design of CFS’s power plant, and that customer engagement, along with the underlying technology, separates a commercial fusion venture from a government research project.

Miles said a commitment validates the technology differently for an OEM like Type One Energy, which is not selling electricity directly.

“It validates the technology; you have people willing to offtake the product,” he said. “We talk to the Tennessee Valley Authority (TVA) weekly, actually several times a week.”

Type One Energy signed a letter of intent with the TVA last September to co-develop its first project. Miles described the first unit as a “minimum viable product.”

“We’re very focused on the economics, but we come at it from a different direction,” Thomas said. “We’ll be building a dual-use facility that can be monetized from day one, so this commercial viability piece can be accelerated if you’ve got the right technology and the right partnerships.”

First Light Fusion is targeting a gain of 1,000 for its inertial fusion target technology, which he said is the single biggest lever for cutting electricity costs in this approach. He said the company reached a significant milestone the day before the panel.

“We recently announced that we reached 150 million degrees Celsius plasma temperature and did meaningful deuterium-deuterium fusion, the first for a fusion energy company,” Proffitt said.

Helion has built seven fusion machines over roughly a decade, each more complex than the last. Its current machine, Polaris, is the seventh, using a pulsed approach called field-reversed configuration, magnetic inertial confinement and deuterium-helium-3 fuel.

“We chose the tokamak design. It was intentional,” Ganten said. “The physics largely worked out for our machine. It’s always more to do, so then it became a systems integration and engineering challenge.”

She said CFS chose a tokamak because more than 50 have already been built worldwide, giving the industry broad familiarity with the design. CFS recently completed a preconceptual design for its follow-on plant, ARC, which it expects to begin supplying power to the grid in the early 2030s.

“We are building the world’s most advanced stellarator in the Tennessee Valley,” Miles said. “Our target is to have a fusion stellarator power plant integrated at scale in the first half of the 2030s.”

The approach builds on stellarator research at the Max Planck Institute in Germany, which demonstrated a non-destructive, steady-state plasma.

Building fusion at scale

Ganten said CFS designed the company from the outset by studying how the aerospace and automotive industries develop and deploy products quickly, aiming for plants that share a common design so components can be assembled on site.

“If you wanted not 400 megawatts but a gigawatt, you can put two units next to each other and share subsystems for economies of scale,” she said. “We’re manufacturing a magnet for SPARC right now. We’ve already decommissioned the line for a toroidal field [magnet] because that’s done. We’re going to retool that factory to do the ARC magnets starting next year.”

ARC refers to “affordable, robust and compact,” while SPARC means the “smallest possible ARC.”

“To deploy at scale, you have to have a standardized product that can be built on an assembly line,” Proffitt said. “There’s no reason fusion generators can’t be a similar commodity to aerospace.”

Helion’s headquarters in Everett, Washington sits across the street from Boeing. He said the fusion economy beyond cheaper power includes the whole supply chain, from skilled workers to machine shops and manufacturing foundries.

“It’s a relatively tiny amount of investment to get inertial fusion moving in a more coherent way in the UK,” Thomas said. “It’s pretty close to oven-baked already. Inertial fusion is dual-use, so we’re talking not just about energy but about defense and security as well.”

He said most UK inertial fusion work has so far been privately funded, and that the country already has strengths in pulse power, lasers, modeling, simulation and target design that need to be brought together.

“The Infinity Fusion Consortium came together a couple of months ago to try to replicate what we were doing in the United States with the TVA,” Miles said. “We’re not making any effort to supplant the STEP (Spherical Tokamak for Energy Production) program, we’re willing to build on it.”

He said the consortium, which also includes Tokamak Energy and the engineering firm AECOM, is looking for a UK “fast follower” project, pointing to the country’s existing regulatory framework and supply chain ecosystem as an advantage.

Taguchi said Japan’s approach sits between the UK and the US models, combining a government-led project with a private-sector acceleration program.

“One is a government-led project, now called QST Demo. The other is a private companies’ acceleration program, which is quite similar to the milestone-based program in the US,” he said. QST refers to quantum science and technology.

He said the government-led track traces back to a project originally based on the International Thermonuclear Experimental Reactor (ITER) program and resembles the UK’s STEP, while the private-led track resembles the US’s milestone-based approach.

Helical Fusion already partners with Japanese manufacturing and assembly companies as part of its own plant development, he said.

Commonwealth Fusion Systems is aiming to turn on its demonstration machine next year, the first step toward a wave of commercial-scale fusion plants that developers across the industry are targeting for the early to late 2030s.

Europe’s push to keep pace with data center demand is hitting a hard wall: not enough power, and not enough land already connected to the grid to build on.

One proposed fix is drawing mixed reviews across the industry. Utilities and energy companies are bundling land with a ready-made power connection, built specifically for a data center to move in on top. Some operators see a shortcut through years of grid queues. Others see a new dependency replacing an old one.

“It’s positive in the initial phases, and [the energy companies] are the owners in the beginning. But the models I’ve seen have them holding a minority stake once a site is commercialized,” said Rune Hogga, director of power at Bulk Infrastructure. “I see the collaboration with energy companies positively, but maybe not as the owner of the data center buildings.”

“Powered land can be a very complex project, but I would also like to see the flip side,” said Carlos Alves, manager of energy and capacity at Digital Realty. “Even if from a hundred opportunities there is one that becomes feasible, that can already mean something for us.”

Laurens van Reijen, chief executive of LCL Data Centers, said European Union rules block that combination outright in Belgium. A distribution system operator (DSO) or transmission system operator (TSO) must remain a neutral network manager rather than a market player.

Van Reijen said the alternative carries its own risk. LCL once turned down a 15-year biomass power deal over fuel-supply risk.

Similar offers from older industrial sites with spare capacity carry a geographic mismatch, he said. Those plants are rarely located near the internet exchange points and ports that data centers depend on, and the land beneath them is often held on concession rather than owned outright.

He pointed instead to the European Commission’s plan for “data center acceleration zones,” which would designate areas with spare grid capacity in each of the bloc’s 27 member states and fast-track permitting inside them.

The commission wants the region’s data center capacity to grow from 12 gigawatts last year to 28 gigawatts by 2030, and van Reijen said he expects the European Parliament to vote on the zones policy within months.

Already acting like utilities

The panel, titled “Are data centers becoming Europe’s new energy partners?”, was held in London on September 16, at the DCD Connect London conference organized by DCD. Elsa Fucile, the company’s portfolio director for North America, moderated it.

Joining her were Rune Hogga of Bulk Infrastructure and Goran Arya, director of energy origination and structuring at Equinix.

Some operators are already behaving like power companies without calling themselves one. Van Reijen said LCL now produces 40% of its own energy from company-owned wind and solar assets, largely to meet customer demand for corporate power purchase agreements (CPPAs), long-term contracts that let a buyer purchase electricity directly from a generator.

“We are already energy companies because we do CPPAs for our customers,” he said. “I can imagine that in five or ten years we will be back here, and we will be a data center and energy company.”

Hogga said the scale of new sites is forcing similar questions even where operators have resisted the label. Two-gigawatt campuses tied to long-term power purchase agreements carry balancing costs that few utilities want to absorb alone, pushing developers to weigh whether to hold assets they could use to manage that risk themselves.

“It forces you to think about getting closer to being an energy company,” he said.

Not every operator wants to go that far. Arya, who spent 15 years at an energy company before moving to Equinix, said the firm has no interest in owning power generation, even though its sites run backup fuel cells and diesel generators.

“I don’t want to be an energy company again,” he said. “We don’t want to own generation on site. That is not our core business.”

Alves struck a similar note, saying the company’s core commitment is reliability, not energy supply. He said third-party firms can bridge operators and grid operators, provided the contractual boundaries between them are clearly defined.

Meanwhile, grid operators are tightening the rules for who gets to enter the connection queue in the first place. Van Reijen said Belgium introduced dedicated capacity lanes after its own queue became congested, splitting demand into fixed shares: 85% for battery storage projects; 10% for data centers; and 5% for industry, with unused capacity reclaimed after a year.

“The biggest thing that helped get the queue shorter was you had the proof you had the land,” he said. “Speculators didn’t have it, and that alone solved 30% of the queue.”

Other countries are moving further still.

Hogga said Finland now requires developers to hold a building permit before they can even apply for a power connection, up from no land-zoning requirement at all until recently. He said the same pressure now applies to existing Nordic industrial power allocations, freeing further grid headroom.

Flexibility now pays

Flexibility has quietly become its own market inside the industry. Hogga said the value system operators are willing to pay for flexibility in Norway has risen tenfold in four years, a price signal operators can no longer afford to ignore.

“If you don’t act toward those price signals, the system operators will have to implement other actions,” he said.

Alves said flexibility works best when designed into a data center from day one rather than retrofitted later, when adding equipment to a live site demands a much deeper risk analysis. He cited a European Union directive already requiring operators to report aggregated performance data to regulators.

“Retrofitting is really expensive,” he said. “The good thing about Belgium is that flexibility legislation only applies to new lines, not something existing, because retrofitting something old is very difficult and risky.”

Arya linked the same trend to public trust. He said the calculus goes beyond electricity price alone, pointing to water use and being a good neighbor to nearby communities as part of what he called the industry’s “social license to operate.”

“We’re all downstream of policy,” he said. “Once these are built, we are helping reliability of the grid where we operate.”

The type of workload also shapes how much flexibility a site can offer. Alves said a colocation facility aggregating dozens or hundreds of customers reads as more stable to the grid than one built purely for artificial intelligence (AI) training, which can demand large blocks of power from day one.

Arya said the distinction matters most between training and inference. Equinix focuses on inference, where firm delivery is contractually required and there is little room to flex. He said regulatory changes in the United States have already let operators deploy batteries and distributed generation to unlock additional grid capacity, calling it a model he wants Europe to match.

“That is much better geared towards what we can do for the grid operators,” he said. “But we need a partnership to understand where the constrained points are on their grid, otherwise we’re sort of shooting in the dark.”

New capacity is being built along similar lines. Van Reijen said cheap power and available land make the Nordics the obvious home for AI training, while inference needs to sit close to population centers such as Belgium’s telecom hubs.

“I’m always very jealous. I look at electricity prices in the Nordics, and it’s a tenth of our price,” he said. “I don’t believe inference will go to the Nordics.”

Alves said the two pulls will coexist rather than cancel each other out, since financial and real-time applications still need to sit within milliseconds of end users regardless of where power is cheapest.

With the European Parliament expected to vote on the acceleration zones proposal in the coming months, the panel’s operators said the industry’s next test is showing regulators it can be trusted with the additional capacity, not simply requesting more of it.

The definition of tech sovereignty in Europe is shifting away from owning every layer of the artificial intelligence (AI) stack.

Sovereignty, in this view, is not about building chips, models and software entirely within Europe. It is about operational resilience, staying in control without relying on a switch someone else can flip. The term “sovereign AI” only entered wide use as generative AI took off in 2023, and its meaning is still being worked out in practice.

“For those workloads you need jurisdictional immunity,” said Manuel Greisinger, general manager of Google Distributed Cloud EMEA. “You don’t want the Cloud Act to apply, and you need it to be cryptographically and post-quantum safe.”

“It would be delusional to think Europe could have the whole stack, from building chips that compete with the latest GPUs (graphics processing units) through to the software layer on top,” he said. “You cannot rely on infrastructure that has a kill switch built in, or that doesn’t provide full operational resilience.”

He said the distinction that matters is between workloads that are non-critical and elastic, and those that are highly mission-critical and cannot tolerate a break in the supply chain. He said running mission-critical workloads on an air-gapped system costs significantly more than putting non-critical ones on standard hyperscale infrastructure, a tradeoff every operator has to weigh.

That framing does not mean shutting out non-European tools, Greisinger said. Sovereign AI, in his view, means operating models in a resilient way that keeps innovation firmly in the operator’s own hands, regardless of where a given model was built.

“Sovereign AI doesn’t mean you don’t use other models,” he said. “As long as you can operate them in a resilient way and drive innovation from them, that’s key.”

Mark Pestridge, managing director of Telehouse, said the practical test for customers is less about geography than certainty.

“The physical infrastructure doesn’t need to be European, as long as you know where the data is held and where it’s flowing from and to,” he said.

The same principle works in reverse, Greisinger added.

“If something happens, can you push a button and evacuate the data to a globally distributed infrastructure?” he said. “That may be the only way to continue operating.”

He cited the war in Ukraine as the kind of crisis that makes an air-gapped, mission-critical setup risky without that fallback.

Power beats permits

The discussion took place at DCD Connect London 2026 on September 16, during a panel titled “Europe’s Precision Strategy: Can a purpose-built digital continent still compete globally?” George Rockett, co-founder and board member of Yotta Events and DatacenterDynamics, moderated the session, which examined whether Europe’s data center industry can keep pace with the speed of artificial intelligence.

Also on the panel were Bruce Stephenson, chief development officer at HSCALE, a data center developer, and Greg Roberts, data center sector lead at Ramboll, the Danish engineering firm.

Stephenson said the industry's success comes down to three Ps: power, permits and people. Asked directly whether he ranks them, he confirmed power comes first, with people a close second.

“Without power, it’s clearly a massive understatement to say we don’t have a data center. It’s about power reform for the better,” he said. “Every developer in the room has had a power-on date slip by two, three, four, five years. To call that a business case killer is an understatement.”

“Just look at the backlogs of the hyperscalers,” Greisinger said. “Every single one has a triple-digit billion figure in backlog that needs to be built out into capacity to be turned into revenue.”

Greisinger said the primary constraint he hears from customers is power availability in existing data centers. His own prescription for closing that gap has little to do with engineering.

“In order to be successful moving forward, Europe needs to be way less fragmented,” he said. “It’s more or less a political issue, with individual countries competing with each other.”

He named Mistral, ElevenLabs and Lovable as the kind of homegrown AI companies he wants Europe to produce more of.

“We cannot act as independent countries. It needs to all come together at the end of the day,” he said. “If that happens, by 2030 you will see enough sovereign AI infrastructure here.”

Building without stopping

Pestridge said the constraint is not only power. It is the physical difficulty of upgrading a data center that never stops running.

“Retrofitting an interconnected data center is really difficult,” he said. “If you say to a customer, we’re going to take that down and retrofit it to provide, say, 45 to 100 kilowatts of rack, they’re going to say no.”

He said the difficulty is one of scale. A single interconnected facility can carry thousands of cross-connects, so customers won't accept taking it fully offline to raise power density.

“What we’re looking at is how you can do it in pockets of those buildings, floor by floor, rather than the whole building,” he said. “We already have the planning permission and access to the power. It’s about making sure customers are comfortable with what you’re going to do to them.”

He said the most difficult example was Telehouse’s old Thomson Reuters building, now known as Telehouse South.

“It’s the most difficult project we’ve ever done,” he said. “You have to constantly balance keeping the lights on and keeping the interconnection going.”

He said many of the connections in older buildings were never fully documented, so identifying one sometimes means physically removing it, a process that can trigger serious outages if it goes wrong.

Greisinger said Google takes a similar approach on the software side.

“We take the approach of containerizing the solution, a fully certified container for specific applications, at least to get customers started while they retrofit or build new locations,” he said. “It’s all about the time to value.”

Roberts offered a different picture of where the industry is heading, literally closer to home.

“There’s a rise in inference, and we’re going to start to see that integrated within our cities,” he said. “These are opportunities to use stranded power, and the waste heat these sites produce for community heating schemes.”

“Data centers seem remote, stuck out somewhere. Going forward, I think communities are going to start to see these smaller data centers within cities supporting our economies, research, manufacturing, healthcare and social benefits.”

“We need to start choosing the locations where we want data center development to happen,” he said. “That means making sure there is power, water, connectivity and skills available in that location.”

He said this kind of planning, which European Union (EU) legislation already allows member states to pursue, would also help attract companies that share Europe’s own values around data sovereignty and sustainability.

The panel agreed that this vision of embedded infrastructure only works if local communities are on board.

“The third thing to look at is enhanced public perception,” said Pestridge. “We need to be more proactive about engaging with the public, and understanding the good our industry enables, rather than just the energy or water it uses.”

He said knocking down a building and starting again causes disruption, traffic, construction work, dust and pollution that a data center operator cannot inflict on the community it operates in.

Panelists said that trust will decide how much of Europe’s AI capacity gets built at home, rather than imported.

Enterprises trying to scale artificial intelligence (AI) workloads are running into a hardware market where graphics processing units (GPUs) are effectively rationed and colocation space is increasingly reserved for the largest cloud providers.

That scarcity is reshaping infrastructure planning. Some companies are stockpiling capacity before they know what they will use it for, while others discover late in the process that they lack the power or cooling to run what they have already committed to build.

“All capacities, all GPU cards, are running at full scale, so none are wasted,” said Yitian Xu, head of solutions for Alibaba Cloud’s UK and Ireland, Nordics and Cyprus region. “All customers who want to deploy GPU cards have to wait.”

Xu said customers now have to commit to one- or two-year contracts to reserve capacity, as demand from both external customers and Alibaba Cloud’s own internal model use keeps climbing. He linked the strain to a wider constraint: the UK’s limited electricity supply, which he called a national growth problem.

Retika Gupta, a solutions architect at Enterprise Rent-A-Car, said buying GPUs has become difficult for the same reason. Suppliers routinely quote six to 12 months for delivery, she said, and the timeline can shift again after an order is placed.

She said the uncertainty raises a harder question before any purchase: whether it is worth buying high-capacity GPUs at all, given they are likely to be outdated within 18 months, rather than leaning on cloud capacity or existing infrastructure instead.

“When COVID happened, we all went into hoarding essentials,” said Uma Mudigonda, vice president of solutions at Kyndryl. “That has to stop. It is going to bite us pretty soon.”

She described a recurring pattern among her customers: businesses declare themselves ready to move from testing into private AI infrastructure, only to discover during the design process that they lack the power to support it.

“The private AI vendor gives you the kit, but you have to host it,” she said. “When there is a colocation and a service provider, there is an opex (operating expense). A lot of things are playing, and then when the business case stacks up, the question is, do I need this?”

Hybrid by design

The panel took place at DCD Connect London on September 16, organized by DatacenterDynamics (DCD).

Titled “Hybrid cloud in the AI era: best practice or a model under pressure?” and moderated by Dan Swinhoe, editor-in-chief of DCD, the session examined how enterprises are deciding where to place AI workloads as cost, capacity and sovereignty pressures mount.

Kyndryl describes itself as the world’s largest infrastructure services provider, managing mission-critical systems for clients across 60 countries. Alibaba Cloud is both a cloud provider and a model developer, with its own custom AI chips built as an alternative to Nvidia hardware.

Gupta said hybrid infrastructure is not fading away in her experience, despite years of industry rhetoric about moving everything to the cloud.

“It’s not dying. It’s our reality, so it’s not going anywhere,” she said. “You do have a capital expenditure which you have already spent. You’re not going to get rid of it at one moment.”

Enterprise Rent-A-Car runs a mix of owned data centers, colocation facilities and both private and public cloud. The company briefly pursued a cloud-first strategy but abandoned it once AI workloads made the case for hybrid clearer. Gupta now designs each application first, establishes where its data lives and what compute it needs, and only then decides where it should run.

John Bradshaw, field CTO for Akamai Technologies in Europe, the Middle East and Africa, described a telecommunications client in South America that connected WhatsApp to its customer relationship management (CRM) and provisioning systems, letting customers resolve issues without calling a contact center. The workflow reached production and touched live network provisioning and customer data.

“Their agentic workflows were delivering real value to them very quickly, and they loved the outcomes they were getting,” Bradshaw said. “What they stopped loving were the bills from the frontier LLMs (large language models), because they were working out to be more expensive than having an agent able to answer the phone.”

He said organizations are increasingly willing to commit to specific workloads once the value is easy to validate, but large capital outlays remain risky given how fast the field moves. Eighteen months ago nobody used the term “agentic,” he said, and in another 18 months something equally unforeseen is likely to reshape the calculus again.

Akamai provides GPUs as a service along with model routing and gateway infrastructure delivered globally.

Xu said Alibaba Cloud has built a routing gateway that judges each incoming request by task and complexity before sending it to the most cost-appropriate model.

“Normally, customers are okay to have the copilot, so they adapt quite quickly,” he said. “For the autonomous side, they are always thinking about the scenario.”

He said he advises customers to keep a human in the loop for autonomous agents, and to avoid letting them run unsupervised on financial transactions or other mission-critical systems.

Sovereignty enters the calculus

Data sovereignty has become as pressing a question as cost or capacity for many enterprises weighing where to place AI workloads.

“Conversations used to be about which cloud to use. But now it is, what am I getting, and where am I allowed to run,” Mudigonda said. “There is a lot of confusion around sovereignty, around user data, where the data is.”

She described a case in which a US company’s acquisition of a Netherlands-based company was blocked by the Dutch government because the target hosted citizen data in its facilities and the deal was judged a sovereignty risk.

Xu said European customers are increasingly pressing Alibaba Cloud on where their data and models are physically hosted.

“Are you sure that all the data, all the models, are deployed on the EU data centers?” he said. “This is an increasing question to us.”

He said some customers in regulated sectors such as education and finance are choosing to compress large open-weight models to run on smaller, on-premises hardware rather than rely on cloud-hosted large models, partly for governance reasons.

The panel split over how much the open-versus-closed choice still matters.

“Open weight is not the decision factor for me,” Mudigonda said. “Enterprises are picking a small model, not the large model, and the situation is, they have already started long back with AI and ML (machine learning).”

“I don’t agree with that part, because as an architect, we decide that,” Gupta said. “Even if it has been decided by someone that we are going to use that model, we can go and challenge it.”

Gupta said no single model works for every use case, so architects need to keep re-evaluating the choice as needs change; a decision that was right a year ago may not hold after 18 months. Mudigonda said she agreed with the underlying point but that in her own customer base the decision was often already locked in by the time she became involved.

Asked whether colocation providers increasingly prefer to sell entire buildings to hyperscalers rather than retail space to individual enterprises, Bradshaw said the shift toward cloud will continue by necessity wherever local capacity runs short.

“If there isn’t sufficient capacity in the geographic region you’re operating in, the supply chain isn’t there. Then by necessity, it’s going to shift,” he said.

With sovereignty rules tightening and hardware still scarce, the panel’s shared expectation was that hybrid infrastructure, rather than a clean move to any single cloud, will keep defining how enterprises build for AI in the years ahead.

Canva is folding artificial intelligence (AI) into the everyday tools people use to build presentations, documents and videos, while drawing a sharp line against the flood of generic AI-generated content spreading across the internet.

The distinction is deliberate. AI that helps finish a specific piece of work, whether a slide deck, a video edit or a brand-consistent design, is treated as a feature worth building. AI that produces interchangeable content at scale, with no clear owner or voice, is treated as the problem worth fighting.

Cameron Adams, cofounder and chief product officer of Canva, described that ambition in practical terms.

“You can start a strategy doc, move that into a presentation, create a whiteboard for your team to brainstorm on, and create videos to go out to your social channels,” Adams said.

In his presentation, Adams said the biggest threat to design isn’t a rival company, but sameness.

“You ask an AI tool for a Spring Fair flyer, and you’ll get something like this,” he said. “When you want to stand out in a sea of content like that, where do you go?”

He said the answer is inside Canva itself, where AI-generated slides and video can be fine-tuned and kept consistent with a company’s brand kit, its logos, fonts and photography, rather than published as-is.

TechJournal.uk raised competition from rivals such as Adobe Express in the Q&A session. Adams’ answer was blunt.

“We don’t see a lot of competition from that part of the market,” he said.

He pointed to Canva’s brand-kit tools and its ability to span multiple design types, from presentations and documents to videos and websites, at an accessible price, as what keeps enterprises on the platform.

Adobe, the company behind Photoshop and Acrobat, built Adobe Express as a free rival powered by its Firefly generative AI engine. The tool sits inside a unit Adobe calls Acrobat plus Express, which passed 900 million monthly active users last quarter, up more than 25% year over year.

A less direct rival is CapCut, the free video-editing app from ByteDance, owner of TikTok. It had more than 323 million monthly active users as of mid-2024 and remains one of the world’s most downloaded apps.

Despite the crowded field, Canva’s own numbers point to strong momentum. Enterprise revenue grew 100% year over year, and its AI tools have now been used 34 billion times, a sign that building AI into finished work, rather than generating content wholesale, is resonating.

“People often don’t need the mega-expensive frontier model to run the work they’re planning on doing,” Adams said. “They can take a second-tier model that’s five times faster and 30 times cheaper, and use that to do a better job because it matches their needs better.”

That mix of smaller and open-source models is central to how Canva funds the AI features driving its strategy. The company’s machine learning platform runs on Anyscale, built on the open-source Ray framework, which Canva has said cut its cloud computing costs by about 50%.

The stakes are real. Canva cut its 2026 growth forecast from around 30% to 20% in August, after AI costs ran ahead of plan, eroding the near-zero marginal cost that once made software so profitable.

On the prospect of an initial public offering (IPO), Adams stuck to a familiar line.

“We’re always IPO ready,” he said. “But in terms of the timeline, we’re in no rush.”

He added that staying private lets Canva make long-term moves that publicly listed rivals cannot, an advantage he said has become more pronounced over the past two years.

New tools cut costs

Adams was speaking at a Canva media briefing in London on September 15.

The briefing, organized by Canva to outline its product roadmap and AI strategy, centered on Canva AI 2.0, a conversational design tool launched in April that remembers a user’s brand, past documents and previous presentations, weaving them into anything new a user asks it to create.

“The demand exploded before the economics were ready,” Adams said, describing a surge in usage that forced Canva to pause the product’s wider rollout.

Users of the tool were three times more likely to use AI generally, pushing Canva’s AI users to 75 million and straining the company’s infrastructure. About 120 in-house researchers spent four months retuning models and shifting workloads across global compute providers to bring costs under control.

“We’ve managed to reduce cost per task by 90%, which then enables us to roll out to even more users, and also to include free users,” Adams said.

Duncan Clark, cofounder of the data-visualization tool Flourish and head of Europe, Middle East and Africa (EMEA) at Canva, said Canva’s product lineup splits along similar lines to its approach to AI.

“The visual suite is for the vast majority of people with no specialist design skills,” Clark said. “Then we have the ProSuite for people who need extra firepower for their craft.”

He said much of Canva’s enterprise resilience comes from adoption that starts long before any formal contract is signed.

“We just have this incredible bottom-up adoption,” he said. “Often when we go and speak to a company for the first time, there might already be 10,000 users in that company signed up who haven’t yet been consolidated into a full agreement.”

The centerpiece of that ProSuite is Affinity, a UK-based challenger to Adobe’s professional design software that Canva acquired in March 2024 for roughly $1 billion. Canva made Affinity’s core tools free in 2025, and more than 7 million advanced users, including professional photographers, have downloaded it since.

“We said free forever. We really meant it,” Clark said. “Our model with Affinity has always been that the core price should be free.”

Canva paired Affinity with Cavalry, a Manchester-based motion-graphics company it acquired in February 2026, and Flourish, the visualization tool Clark co-founded before joining Canva in 2022. Together they form the Canva ProSuite.

“This is going to be the most powerful designer ever on iPad,” Clark said of an Affinity app for Apple’s tablet yet to be released.

UK growth funds giving

The UK has emerged as one of Canva’s most important markets.

Users there created 224 million designs over the past 12 months, roughly 750,000 UK small and medium-sized businesses now use the platform every month, and Canva employs about 300 people in the country, with three buildings now clustered around a London square.

Its UK enterprise customers include National Grid, a 700-office estate agent chain and Coca-Cola’s Euro-Pacific division, alongside a growing community of independent trainers and creators who run Canva workshops across the country.

Getting the product right in each market means paying attention to local quirks, Clark said.

“In most of the world, if you search for hemp, you actually are looking for a farmyard picture,” he said. “But in the UK, hemp (cannabis) obviously means something quite different.”

Canva extends that same local focus into education, partnering with the Scottish Digital Learning Program to let every local authority in Scotland give schools free access to the platform, and building Learn Grid, a library launched six months ago with more than 50,000 curriculum-aligned lesson materials.

Britain also punches above its weight in Canva’s non-profit base.

“5% of the non-profits that use Canva are based here, which I think shows the real social conscience and social impact that’s happening,” Adams said.

Two UK non-profits show how that plays out on the ground:

• The Scottish Drug Forum uses Canva to produce its own education materials in-house, saving time and the cost of outsourcing.

• Stop Hunger Now, a global charity, localizes Canva templates so volunteers worldwide can adapt them for local campaigns.

That local work sits inside a wider philanthropic model Adams called Canva’s “two-step plan”: build one of the world's most valuable private companies, then use that value to do good.

“Build one of the most valuable companies that does the most good,” he said, describing the plan's philosophy.

More than 1 million non-profit organizations worldwide use Canva for free, part of a giving program funded by a 30% equity stake set aside for the Canva Foundation, which has delivered $150 million to date.

A separate $150 million commitment goes to GiveDirectly, which gives cash directly to people in poverty rather than dictating how they should spend it. Canva said 90% of recipients rose above the extreme poverty line within three months, and school absenteeism among their children fell 60%, with $100 million of the commitment already deployed.

Canva has also given $4.7 million to disaster and conflict relief since 2021, covering natural disasters and wars.

It has funded a solar farm in the United States that its print providers use to cut their carbon emissions, and under a program called One Print One Tree, it plants a tree for every print order placed through the platform, a tally that has reached 18 million against a target of 30 million.

As Canva’s private valuation keeps climbing, so does the fund it has set aside to give away.

Delivery robots that once stalled every few seconds on crowded sidewalks are now navigating chaotic city streets in real time, thanks to an artificial intelligence (AI) shift borrowed directly from large language models (LLMs).

For years, robotics relied on rule-based systems: sensors built a map, a robot located itself on that map, detected every object around it, then followed a hand-written rule for each one.

That approach couldn't keep pace with real streets, where pedestrians, cyclists and cars move unpredictably, and it left machines stuck waiting rather than moving.

“The traditional robotics stack would have you use LiDAR (Light Detection and Ranging) or cameras to create a map, then localize yourself within it, detect every object and build rules to create a path around it,” said Zach Rash, chief executive and co-founder of Coco Robotics.

“But you’re always surrounded by people on e-bikes, people in cars, people walking,” he said. “It’s very dynamic, so creating those rules is impossible.”

“The architecture behind the LLMs translates into robotics,” he said. “We have millions of hours of driving in these environments, and that can be used to train a reliable self-driving model on the robots.”

“There is this Cambrian explosion of robotic capability that’s being driven by advances in the language models and the kinds of AI we all use to do knowledge work,” said John Hanke, executive chairman of Niantic Spatial. “It’s led to this revolutionary period where robots are capable of doing things today that they couldn’t do a year or two years ago.”

Rash said his team already held the driving data those models needed, because Coco set a high quality-of-service bar from the start. In the company’s first few years, humans teleoperated the vehicles rather than deploy them untested, quietly building the real-world dataset the models would later train on.

“Teleops has historically been this taboo thing that no one wants to talk about, but you need a great dataset in the real world, and the best way to create that dataset is to provide actual value,” he said. “So you can continue to collect that data and iterate.”

He said Niantic Spatial’s mapping is the other half of the equation, since the maps a robot needs differ from the maps a person needs, and its team also has to solve for something closer to human intuition about drop-offs.

That combination of driving data and precise maps is why Niantic Spatial and Coco Robotics formalized a partnership in March, putting Niantic Spatial’s mapping technology directly into Coco’s delivery fleet.

Building the living map

The pair discussed the shift at The AI Summit London in June, during a panel titled “Modeling Reality: When AI Leaves the Cloud and Starts Walking the Streets.” Informa Tech organized the event in London, and Stephanie Hare, a researcher and co-presenter of the BBC series “Artificial Intelligence: Decoded,” moderated.

Rash co-founded Coco Robotics to build small vehicles for last-mile delivery in dense cities. Hanke spent a decade at Google building out Google Earth and Google Maps before founding Niantic, the company behind Pokémon Go.

Niantic sold its games division to Scopely for $3.5 billion in 2025. The remaining business became Niantic Spatial, a geospatial AI company Hanke now chairs.

Hanke said the trajectory toward precise maps began at Google, adding satellite imagery, aerial photos and Street View’s ground-level detail. Pokémon Go sat on that foundation, but he said its real origin was personal.

“The genesis of Pokémon Go was me as a parent. I had three kids who were into computer games, and I wanted to get them outside, so I thought it’d be awesome to combine game playing with being outside,” he said.

Exploring augmented reality, layering digital objects into glasses, later demanded even more precision.

“You want to be able to place things in the world within a centimeter. You want to be able to measure that error in centimeters,” he said.

Hanke said that extra detail came from players who opted in to scan an area and upload images, which were turned into 3D maps accurate enough to place game characters exactly, without collecting anyone’s data without consent. That pipeline, now expanded to drones and other imagery, feeds Coco’s delivery fleet.

Rash said the appeal is sharpest in extreme weather.

“When we met with the Finnish regulators, their comment to us was that these winters took down the Russian army. Are you sure you want to launch here? And we confidently said yes, we’re sure,” he said.

Coco has since expanded into three or four Finnish cities, alongside Los Angeles, Chicago, Miami, North Jersey and the Bay Area, running more than 1,000 vehicles as of the panel.

“We’re the most useful for our partners in markets with extreme weather,” Rash said. “When it’s cold and dangerous to be outside on a bike, demand for the service goes up because people don’t want to leave their house.”

“But courier supply goes down because it’s unsafe or unpleasant to deliver in those conditions. You get this big supply and demand imbalance, and robots can be helpful to keep a reliable service running,” he added.

He said the vehicles are built to be fully submersible, which matters in Miami’s routine heavy rain, and that the same logic held during the Los Angeles wildfires, when poor air quality made it risky for human couriers to work outside.

Winning trust on sidewalks

Rash said the vehicle’s design was engineered around trust as much as logistics: shoulder width so it will not block a sidewalk, narrow enough for bike lanes at speed, and sized for about four grocery bags or eight to ten extra-large pizzas.

“The robot’s pink. We named the company Coco after a ten-second naming exercise: we looked up cute dog names, and Coco was number one,” he said. “We wanted it to feel like this warm, friendly part of the community, more like a Wall-E, without feeling like this imposing piece of technology.”

“We’ve been in Los Angeles the longest, and it’s completely normalized there: people don’t call it a robot. They call it a Coco, and that was intentional in how we designed it.”

“It’s locked, first of all. We’ve done millions of miles of deliveries across LA and Chicago, and some of these areas have higher crime rates, and we just haven’t seen much vandalism,” he said. “We’ve never had any thefts of the vehicles. They weigh about 150 pounds, so it’s a little awkward to pick up, and they have dual locks on the lid.”

Cities often ask whether Coco will see the vandalism that hit micromobility devices like e-scooters. It has not, largely because the vehicle shuttles between merchant and customer rather than sitting on the street, and because people treat it as part of a business they already trust.

Rash said the deeper problem is capacity, not manners: delivery keeps growing faster than the pool of human drivers, especially for perishable orders, and robots fit a city’s infrastructure better than more cars or e-bikes because they are:

• small and lightweight

• quiet

• extremely energy efficient

• inherently safer by design

“On a sidewalk, we’re designed to be the most yielding and most respectful pedestrian. We have maps of these cities to identify the most underutilized parts of the infrastructure to use,” he said.

“We’ve had this 4,000-pound solution of the automobile for the past 100 years, a one-size-fits-all vehicle that, for safety and other reasons, takes a certain amount of space and a lot of energy to move,” Hanke said.

He said smaller, lighter electric vehicles could free that space for pedestrians and parks, a shift that policymakers, the public and companies must work through together.

An audience question about data collection pressed both men on what it means for privacy. Hanke said separate companies building similar mapping data is probably unavoidable, since citywide open data projects rarely happen quickly.

“Companies need to be transparent about what they’re doing, how they’re handling the data, and discard data they’re not using, so it’s not there for somebody to exploit or request at some future date,” he said. “If you don’t have it, it’s not a vulnerability.”

Rash said cities generally want a better understanding of their own infrastructure. Coco shares fleet data on sidewalk conditions with the European Union and the city of Helsinki, which uses it to maintain sidewalks in real time, and with an app called Blind Square, which warns blind and visually impaired pedestrians about hazards and obstructions.

He said that shared benefit, not secrecy, is how Coco intends to keep expanding into new cities as the fleet grows.

Self-driving cars are usually sold as a software problem. For one of the world’s biggest carmakers, the harder problem now is not the technology. It is the regulators.

The company’s autonomous system already blends lidar, cameras and infrared sensors into a single stack built into every car, rather than leaning on one type of sensor the way some rivals do.

“Our God’s Eye combines hardware, software and artificial intelligence into one integrated driving system designed to make mobility safer, smarter and more accessible,” said Stella Li, EVP and CEO of BYD Americas, Europe, Middle East and Africa. “If you want to be the hero of yourself, a Formula One driver, you have autonomous drifting capability.”

Everyday features built on the system already include automated parking, Li said, while the Z9 GT adds the autonomous drifting mode for track driving.

She said Chinese regulation currently caps the technology at Level 2-plus autonomy, but that Beijing is expected to approve Level 3 rules in July 2027. She said BYD’s hardware is already built for Level 3 and Level 4.

She said BYD is working with European regulators on approval timing, while the technology itself is already ready. More than 3.15 million BYD cars already use some level of the feature, generating over 200 million kilometers of driving data every day.

BYD also builds a voice assistant, called BYD Super Agent, into its DiLink AI Intelligent System, with memory that improves the more it is used.

“You can talk to them, and they will talk to you naturally, like you have a very good conversation with your best friend,” she said.

She described the assistant scheduling a trip, checking the news or stock market and opening the windows by voice, handling several requests at once and connecting to services outside the car. She pictured a future version that switches on the air conditioning and a rice cooker as soon as it senses a driver heading home.

BYD is the world’s largest maker of new energy vehicles. Li said more than 5,000 software engineers work on autonomous driving alone, with more than €12.7 billion invested in the technology so far, and that data from those cars keeps improving the system.

Faster than a fill-up

Li made the comments during a session called The Role of AI in Driving Sustainable Mobility at London Tech Week in June. Founders Forum Group and Informa organized the event jointly, with support from London & Partners and Tech London Advocates.

BYD installed its first flash charging station in London, at the parking lot of the company’s own office there.

“A lot of people ask whether 1.5 megawatts of capacity brings a big crash to the grid,” she said. “But we have a secret, because we are not just a car company. We combine flash charging with battery storage.”

Branded “Ready in 5, Full in 9, Cold Plus 3,” the system charges a battery to 70% in five minutes and to 97% in nine, adding about three minutes in extreme cold.

Li said the charger itself is designed to be simple: light enough to lift with one hand, unlike the heavy cables used by many rival systems.

She said each site needs only a modest 100-kilowatt grid connection, because an on-site battery buffer stores electricity in the evening and discharges it to power the car at peak hours, smoothing the load on the local network. BYD has installed more than 6,100 flash chargers worldwide so far, and aims to have about 20,000 running in China alone by the end of the year.

BYD signed a deal with Sinopec, China’s largest gasoline station operator, just days before the London event, agreeing to add chargers at many of its fuel stations nationwide. Its target outside China is to install 6,000 flash chargers across Europe by the first quarter of next year.

BYD’s answer to lingering range doubts is a new plug-in hybrid, the Dolphin G.

The hybrid system, which BYD calls DM-i, runs on electric power for daily trips and uses a combustion engine for longer journeys, delivering what the company says is the lowest fuel consumption of any hybrid on the market.

The Dolphin G also comes with Google’s Assistant, Maps and Play Store built in, part of what Li called BYD’s openness to outside technology partners.

“This will really be the disruptor, using this new technology to really challenge the combustion-engine car industry,” Li said.

She said a single charge plus a full tank of gasoline can carry the car more than 1,000 kilometers, enough to drive from London to Zurich, a distance of 941 kilometers, without stopping.

From startup to giant

BYD’s scale today contrasts sharply with its roots. The company started more than 30 years ago with 20 employees and $350,000 in investment.

BYD’s growth accelerated as it went. It took 13 years to sell its first million new energy vehicles, then each subsequent million came faster: 27 months, then 15, then 13, before the most recent million took just four.

She said BYD became the first carmaker in the world this year to build 16 million new energy vehicles in total.

The company now spans four industries: passenger vehicles; rail transit, including the Sky Rail and Sky Shuttle monorail systems it operates in China and Brazil; renewable energy, including solar panels and battery storage; and electronics.

Li said BYD frames its ambitions as three connected dreams: generating electricity from solar power, storing it in batteries and using that energy to power electric vehicles. The company’s roots in batteries and electronics mean it still supplies about 80% of the world’s tablet computer screens.

She said the company employs about 120,000 research and development engineers, who file an average of 52 patent applications a day. BYD has submitted more than 71,000 patents and been granted 42,000, according to Li.

BYD’s push upmarket has produced some of its most attention-grabbing engineering. Its Yangwang brand’s U9 Xtreme, a 3,000-horsepower electric car, set the record this year for the fastest production car in the world, reaching 496.22 kilometers per hour.

“Our crazy engineers came to us and said, ‘I need to produce a car with 3,000 horsepower.’ We said, ‘Yeah, go ahead, do that,’” Li said.

Yangwang is BYD’s flagship performance brand, built around SUVs and sports cars that Li said are the fastest in their category.

Li described this as a multi-brand strategy that runs from mass-market electric and hybrid models up through the premium Denza and ultra-luxury Yangwang brands, designed to give every price tier access to BYD’s newest technology.

BYD’s premium Denza brand launched in Europe in April with the Z9 GT, a car whose flash-charging battery and 1,036-kilometer range are fronted in marketing by the actor Daniel Craig. Denza followed in July with the Bao 5, a hybrid SUV built for off-road driving, unveiled at the Goodwood Festival of Speed.

Much of this is underpinned by BYD’s Blade Battery, a cell design the company says resists catching fire even when punctured. Li said the second-generation cell has passed 500 flash-charge cycles combined with simultaneous nail-penetration testing without catching fire or smoking.

She said the new cell also improves energy density by 5% and warranty capacity retention by 2.5% over the first generation.

Other BYD models use what the company calls its e-platform 3.0 chassis technology, which Li said lets some vehicles float briefly in floodwater or absorb pothole impacts. A separate four-motor “e4” platform lets a car rotate on the spot in what Li called a “tank turn.”

Open-weight artificial intelligence (AI) developers have spent 2026 fielding a skeptical question. If a model is free to download, how does the company behind it make money? A new infrastructure deal in the Gulf offers one answer, and it looks nothing like renting access to a chatbot.

The arrangement trades a one-time license fee for a deeper commercial relationship built on local deployment, regional customization and shared compute economics. It is an early example of how open models are expected to generate revenue at scale.

Reflection AI, the open-weight model developer led by co-founder and chief executive Misha Laskin, agreed this month to deploy its models on infrastructure run by HUMAIN, a company backed by Saudi Arabia’s Public Investment Fund. The deal was announced at the LEAP technology conference in Riyadh on September 1.

In a press release announcing the deal, Laskin described HUMAIN as a strategic partner in scaling advanced AI models, saying the deal creates an opportunity to deploy models customized for sovereign use.

The logic traces back to an argument Laskin has made for months: that ownership, not rental, is what really separates open and closed AI.

Laskin frames that distinction with a housing analogy. In his telling, closed AI models function like a rental market, a landlord who can throttle access, restrict use or raise prices at will, because the tenant never owns the underlying asset.

Open models, in the same analogy, are the buying market, an asset a customer can hold, modify and never lose.

“If you’re a large bank, you don’t want to rent something that is going to be the single most important IP (intellectual property) asset you have,” Laskin said. “If you’re a country, you don’t want to rent from a private company in another country. You want to own your intelligence.”

He said this matters because AI has moved in just a few years from what he called a nice-to-have, like chat, to something capable of doing real work, and now a source of concern for national security, for companies and for countries alike.

“With an open model, you can own the model, run it on your infrastructure, and no one can take it away from you once you own it,” he said. “You can customize it and train it for your unique workloads, the same way you can build a garden when you own a house.”

That is the commercial logic behind Reflection AI’s deployment deal with HUMAIN, in which a Gulf sovereign wealth-backed platform gains ownership of infrastructure and customization rights rather than renting access to somebody else’s model.

China’s open model race

The comments came during a fireside chat at London Tech Week, organized by Founders Forum Group and Informa. There, Laskin discussed the future of open-source AI with Iain Martin, Forbes’ senior editor for Europe.

Martin pressed Laskin on China’s dominance of open-model rankings, citing a US narrative that Chinese labs had gained ground by distilling closed Western systems. Two of China’s best-known labs, Zhipu and MiniMax, had listed in Hong Kong that January.

Laskin said strong open models from any country, including China, help prevent a dangerous concentration of AI power, and rejected the idea that Chinese progress rests mainly on distillation. He said pre-training, not distillation, consumes the bulk of the compute required to build a model.

“Pre-training is where most of your compute goes, and it’s very hard,” Laskin said. “Those companies have done a lot of things right outside of distillation.”

He said Chinese labs are extremely capable and talented, and will keep advancing regardless, limited mainly by compute access.

Laskin pointed to DeepSeek V4, a model he said was quieter than its predecessor V3 but more consequential, because its inference was optimized specifically for Huawei chips, part of China’s push to move its entire AI stack onto domestic infrastructure.

“Open models are Trojan horses to the infrastructure that they bring along with them,” Laskin said, citing Reflection AI’s own competing bid to help build a sovereign cloud in South Korea against Chinese rivals.

He compared the dynamic to the 5G standoff between the West and China and to the Belt and Road Initiative, saying a sovereign cloud built on Chinese technology also imports the Huawei chip and software ecosystem around it.

He predicted countries such as Korea will end up adopting both Chinese and Western open models rather than choosing one bloc outright.

Martin then turned to the security implications of that ecosystem, asking whether an enterprise or government could safely run an open Chinese model like DeepSeek locally.

“The Chinese models were trained under copyright laws that are much more lax than in the West, so the models were trained on illicit data,” Laskin said. “An enterprise that uses those models inherits all those liabilities.”

He said the deeper concern is behavioral, not legal. Large language models are, in his words, “mechanical brains,” and in the same way people can be taught to behave deceptively, so can a model.

“If a state actor wants to embed sleeper cell type of behavior, that is completely possible,” Laskin said. “I think that is a pretty scary thing.”

He said the risk of sending sensitive data to a Chinese company can be avoided by running the models on a company’s own infrastructure, but the risk of lock-in cannot.

“A large conglomerate might do a $10 billion or $50 billion build for its AI cloud,” Laskin said. “By the time you’ve made that investment, it’s really hard to pull out.”

From AlphaGo to autonomy

Martin opened by asking why Reflection AI ended up building a Western open AI lab.

“We’re building frontier Western open models,” Laskin said. “Today there’s a very large gap between open-model capabilities in the West and the ones in China.”

Laskin and his co-founder, Ioannis Antonoglou, were both researchers at Google DeepMind before starting Reflection AI, where Antonoglou had been a founding engineer on AlphaGo.

“What made AlphaGo special was it learned to play Go (an ancient board game) through reinforcement learning on its own and became superintelligent,” he said.

He said AlphaGo’s limitation was narrowness, brilliant at Go but unable to play tic-tac-toe, a gap he and Antonoglou set out to close as language models began to emerge.

“Our thesis was simple,” Laskin said. “If you applied the same recipe used to build AlphaGo to language models, scaling up reinforcement learning on top of them, you’d get systems capable of general autonomy.”

Reflection AI's other bet was bootstrapping on a strong open Western model. Llama 3 became what he called the most successful open-model launch in Western history. A year later, Chinese open models overtook Western ones, and Meta abandoned its own open-model project.

“We found ourselves in a position where this core piece of technology we believed needed to exist no longer existed,” Laskin said.

He called that gap Reflection AI’s non-linear path into becoming a frontier open-model developer in its own right, a path that has since taken the company to a $25 billion valuation.

Asked whether the stakes extend to national security, Laskin said he is wary of comparing AI to nuclear technology. He pointed to Reflection AI’s role powering the US Department of Energy’s Genesis Mission, a project to embed AI across 17 national laboratories and help 50,000 scientists advance scientific research.

“It also has powerful negative things you can do with it,” he said. “If a country has very powerful models, open or closed, it can have significant leverage over another country by using those models for cyber attacks.”

That is why governments increasingly treat AI as a national security priority. He said every government Reflection AI works with is thinking deeply about how to become sovereign, how to run models and keep the IP within their own country, and how to train a workforce capable of taking these models and customizing them.

The conversation closed on London.

“London is largely the source of all this AI innovation that’s been happening over the last decade,” Laskin said. “The first company to take artificial general intelligence as its mandate seriously was DeepMind, founded in 2010, when it was considered largely laughable by the academic community.”

Reflection AI now has three core locations: San Francisco, New York, and London. The company is hiring toward a team of 100 in London now, growing to 1,000 within the next few years, while also supporting UK enterprises and the government’s own AI sovereignty push.

“My co-founder [Antonoglou] lived in London for over a decade,” Laskin said. “I spent a lot of my time here working with the team, so it means a lot to us personally.”

With Reflection AI’s headcount in London set to grow tenfold, that personal history looks likely to keep shaping where the company builds next.

Startups juggling multiple currencies, entities and revenue streams are getting a new way to see where their money stands in real time, as artificial intelligence (AI) pushes deeper into corporate finance.

Airwallex, a global payments and financial infrastructure company, has launched an AI-native accounting platform called T:0 that automates bookkeeping and revenue recognition for growing businesses.

The platform automatically categorizes transactions and builds financial reports around how a business actually earns and spends, giving founders instant visibility into cash, burn rate, and runway. It went live on September 1 for U.S. startups after a period in private beta, targeting companies from early stage through scaling.

The launch follows Airwallex’s $320 million Series H funding round, which closed on June 25 this year, and valued the company at $11 billion, up from $8 billion the previous December.

Speaking in June, before the round closed, Airwallex co-founder and president Lucy Liu said the strategy was already years in the making.

“We are thinking about today’s business as being very global native,” she said. “Beyond just collecting and paying out money, you need your expense management systems and financial operations systems. It is all interconnected because a lot of the data comes from your payments.”

“That is why we expanded way beyond just payments. We have treasury management software, and with AI, we have an AI Copilot and agentic finance, everything that helps a business operate more efficiently. I think it is where the market is heading, and where we made the right bet,” she said.

Airwallex was founded in 2015 by Jack Zhang and fellow entrepreneurs in Melbourne, after a coffee shop venture of theirs ran into the cost and friction of moving money across borders.

Liu invested $1 million in the idea before later joining as co-founder and president. The company, now dual-headquartered in Singapore and San Francisco, has grown to more than 2,300 employees across 27 offices worldwide.

She said the company moved its financial operations onto its proprietary system about three years ago, adding that she personally approves every bank account and calling the resulting team the most efficient she has seen.

To make that shift visible, Airwallex has sponsored Formula One team McLaren and Premier League club Arsenal, courting the chief financial officers, finance managers and chief technology officers who ultimately choose its platform.

Betting on infrastructure

Liu made the comments in a fireside chat with Jess Jones, senior reporter at The Times, at London Tech Week in June 2026. Organized by Founders Forum Group and Informa, the London event explored how founders navigate the shift from hypergrowth to profitable, sustainable scale.

She credited Airwallex’s early decision to build as an API (application programming interface) first company, wiring its systems so customers could plug directly into them.

“Our choice to be API-first became very valuable,” she said. “It means we can become almost irreplaceable once we are integrated and once we provide the necessary data and services.”

She said pricing across the industry, from foreign exchange margins to payment collection and transaction fees, has become commoditized.

“We are not really competing on price anymore. We started to compete on service, then on scalability. We always emphasize that Airwallex is a platform that can be used from anywhere from a small business all the way up to multinational, billion-dollar enterprises,” she said.

“In the early days it was difficult to sell because people got confused,” she said. “Are you an SME (small and medium-sized enterprise) platform or an enterprise solution? Actually, you are both. You are the whole spectrum.”

Getting there took years of licensing work with regulators, she said. Airwallex now holds more than 85 licenses and registrations globally, serving more than 200,000 businesses.

That infrastructure took years to build. Liu said Airwallex was pre-revenue for about three years after it was founded in 2015, a stretch that tested investor confidence and internal morale alike.

She said the company pivoted its product roughly 18 months in, after the specific offering it was pushing found less traction than the broader infrastructure it was building underneath, including its own banking licenses and payment rails.

“We spent so much time and resources building the foundations of the business, the infrastructure, the APIs and what was going on underneath the applications, so we were able to switch our product offering to something that was welcomed by our customers,” she said.

She said growth then became a matter of speed: sticking to strict unit economics as the company expanded from Australia into the Asia-Pacific region (APAC) and, eventually, worldwide.

When the demo failed

Liu said Airwallex nearly collapsed three times in its first decade, and the closest call came during a funding round when a technical glitch, likely a firewall issue, disrupted a live demo in front of investors.

“It was not the internal team that was championing us. It was the investors who were looking to invest in us,” she said.

Tencent, which sat in on the pitch ahead of its own investment board meeting, told the team there were “greater things to look at” than the single glitch.

Liu said not everyone who joined the company in its early years stayed, but those who did shared a single vision with the founders, and the same drive.

Airwallex has been licensed in the UK since around 2020. In December 2025, it said it would invest $590 million in UK and regional growth between 2026 and 2030.

The plan includes a new Fitzrovia office of more than 16,000 square feet in London and a 60% increase in UK headcount, from about 100 to more than 160 people, by the end of 2026. Christos Chamberlain, a former Flexport general manager, took over as general manager for the UK and EU in January 2026.

“UK and EMEA (Europe, Middle East and Africa) are one of our fastest-growing regions at the moment,” Liu said. “London was a very natural expansion point for us because of the talent and its role as a financial hub, as well as the natural connection it has with where we first started, which was Australia and APAC.”

“You can tell by the sports teams that we pick. They are both London-based and UK-based, and we are receiving a lot of love from our UK customers,” she said, a reference to the McLaren and Arsenal sponsorships.

Big global players approached Airwallex with acquisition offers around eight years ago, when the company was only two years into generating revenue but already growing at hundreds of percent annually.

“We never started the business with the intention of selling it. There was so much more opportunity and potential then. It was a very generous offer, but being independent and building it ourselves was the vision that we had,” she said.

Jones, the moderator, asked Liu whether the funding environment has grown tighter for founders today, with far more scrutiny on profitability. Liu’s answer centered on a single, overused word in the startup world.

“It is a great time to be a builder, but it is also massively easy to build these days because of all the tools you have. You can code something over the weekend if you want to,” she said. “Validation comes a lot quicker these days. You might see results within the first two weeks of launch.”

She said that ease makes people matter more, not less: what a team stands for, and whether it can build trust with customers and its audience.

With the accounting platform now live and its valuation freshly reset at $11 billion, Airwallex’s next test is whether it can maintain that pace as competition in AI-driven finance intensifies.

Artificial intelligence (AI) use inside companies has tripled in under two years. Turning that activity into real business results is proving far harder.

According to Accenture’s own research, nearly 90% of companies have yet to see that surge in individual activity show up in company performance or gross domestic product (GDP). Only around one in ten organizations have managed to convert it into measurable profit and loss impact.

“We’ve got a dichotomy between individual productivity and that translating into company performance,” said Matt Prebble, chief executive of Accenture UK & Ireland. “It’s only when you redesign the work, look at the processes and engage the people in that exercise, that unlocks the productivity step change.”

“When we look at the one in ten we call ‘stayers’ that really start to get P&L (profit and loss) impact, they’re looking at the work itself. They’re redesigning the work, and they’ve got a different engagement model with their employees.”

“A narrow view of productivity, doing the same things I did yesterday but at a lower cost, isn’t a winning strategy. We don’t see people winning with that strategy.”

He said that mindset needs to work at three levels at once, reinventing the industry, the role and the individual’s own next chapter.

“If you don’t give people the tools, they’re going to find workarounds,” said Kerry Dryburgh, executive vice president for people, culture and communications at BP.

She said UK research shows only one in six UK companies are deploying agentic AI tools across their organization, even though more than 70% of the population already uses AI day to day.

The gap between individual enthusiasm and organizational results is proving to be the industry’s central challenge.

Closing the trust gap

The exchange took place on a panel at London Tech Week in June. Patricia Clarke, technology reporter at The Observer, moderated the discussion on bridging the gap between boardroom ambition for agentic AI and workforce readiness.

Founders Forum Group and Informa, alongside London & Partners and Tech London Advocates, organized the event.

Prebble said only 20% of the workforce feel engaged in the AI journey, based on an Accenture survey of thousands of employees, even though about 85% of C-suite leaders plan to keep increasing AI investment and scale.

“That lack of engagement translates into distrust,” he said. “The real innovation happens on the edge, with our clients and project teams. It’s those people on the edge who really create the innovation, not the people in the center deciding what to scale next.”

He said closing that gap starts with mapping out talent pathways for the roles agentic AI is creating.

“About 18 months ago, I had an Accenture reverse mentor who helped me,” Dryburgh said. “We’re at this point in time where you can rethink how you do your job on a day-to-day basis, and how many opportunities do you get like that in a lifetime? Not many.”

“People can’t trust what they don’t know. That’s why it’s important to think about a whole enterprise deployment overall.”

BP has deployed AI agents in its recruitment process to sift through thousands of CVs, but a human still checks for bias and makes the final hiring decision.

“You have to ensure that you’re deploying AI in a way that designs for a human-first approach, and that you’re always taking human accountability and judgment into account,” she said.

BP’s own deployment centers on three areas:

• safety, including maintenance at its facilities;

• production, such as drilling wells more effectively;

• and inventory management across its retail network.

It is working with MIT (Massachusetts Institute of Technology) in Boston on a customized program for its top 120 leaders, alongside broader training across the whole company so employees can identify where AI can remove friction from their work.

“You’ve got to have not just the tools, but a strong data foundation,” she said. “ You’re an expert in your field, and if you can understand enough, you can become a ‘super user.’”

BP calls these employees “citizen developers,” who apply AI to solve problems day-to-day without formal technical training.

“We’ve deployed something called ‘AI Edge,’ which is learning over a 12-month period, whether that’s online learning or masterclasses, across the whole company,” she said.

Doubling down on graduates

When Prebble took over leading Accenture’s UK and Ireland business, one of his first questions was how many graduates to hire this year. He said Accenture redesigned its whole graduate intake program in response, including a 12-week onboarding course built around tools from Microsoft, Anthropic and Amazon Web Services (AWS).

“There’s an easy mindset that says we need fewer graduates because the work they did previously is going to be done by AI,” he said. “That is entirely the wrong mindset to have.”

“We found there’s a whole new set of tasks those people can do that could have real client impact, and clients really want to engage with AI-native people who can drive their growth forward.”

He said Accenture is increasing its graduate intake this year as a result, rather than cutting it.

“The uncomfortable truth is that the US and China have led the way [on infrastructure],” he said. “There is a super exciting opportunity for the UK, which is how you apply the new technologies in industries like financial services, life sciences and the legal sector.”

New Accenture research puts the UK’s agentic AI opportunity at roughly £736 billion by 2038, the largest of any European country outside North America.

Prebble pointed to Amazon’s retail arm as an example of a low-cost leader that got there through reinvention, not by doing what traditional retailers already did more cheaply.

He said the only way the UK gets there is with a broader view of productivity and a real mindset to reinvent, rather than the narrow view of doing as little as possible and just trying to reduce cost.

“There’s opportunity to create new businesses and new versions of those industries, whether that’s professional services, energy or consumer goods,” he said. “That is a huge opportunity for the UK.”

“About 8 million people in the UK don’t have digital connections. They can’t open a bank account or shop online,” he said. “If you give them the tools to get started, if they cross the bridge and start using a chatbot, you suddenly see the confidence flourish.”

He cited outreach work using generative AI that had just reached a million people, as evidence of what a first successful interaction can lead to. He said closing that gap for the UK’s 8 million digitally excluded citizens, like closing the gap between individual AI use and company-wide results, will depend on whether more organizations choose to redesign the work rather than simply digitize it.

A musician who lost his singing voice to a terminal illness performed live again this year, after an artificial intelligence (AI) company rebuilt his voice from old recordings.

The moment is being offered as proof that AI can restore, rather than replace, something people value most: their own voice. It is also framed as the clearest answer to a harder question: why a public wary of AI should trust it at all.

“AI is only useful if everybody actually has access to it,” said Mati Staniszewski, co-founder of ElevenLabs. “Probably one of my proudest works of all of it was working with people who lost their voice, unfortunately, to motor neuron disease (MND) or amyotrophic lateral sclerosis (ALS), and we could bring it back.”

He was describing Patrick Darling, a folk musician diagnosed with ALS at 29 who lost his ability to sing. Using pre-diagnosis recordings, ElevenLabs rebuilt his voice so he could perform live again.

“For the first time in years, I could hear myself sing again,” Darling has said. “That changes everything.”

The performance took place in February at the ElevenLabs Summit in London and is believed to be a world first. He was also billed to perform again at South by Southwest London.

He worked with Richard Cave, a speech therapist on ElevenLabs’ partnerships team, who used old recordings to help rebuild the voice he could no longer produce himself. He performed an original song, “Ghost of a Man I Never Met,” before an audience of nearly 1,000 people.

His case sits inside a wider effort. The ElevenLabs Impact Program, run with the nonprofits Bridging Voice and the Scott-Morgan Foundation, aims to give a million people affected by ALS and MND free access to voice cloning.

“When my family first heard my voice again, they cried,” said Tim Green, a former National Football League (NFL) player with ALS who used the program.

“Despite that unfortunate event, you can still inspire, do that work and bring everyone back,” Staniszewski said.

ElevenLabs is a London-based voice-AI company Staniszewski co-founded in 2022. It reached an $11 billion valuation in February, with backers including Sequoia Capital and Andreessen Horowitz.

Agents already on call

The comments came during a fireside chat at London Tech Week in London in June, moderated by Matt Clifford CBE, chair of the UK’s Advanced Research and Invention Agency (ARIA). The event, which explored the future of human-technology interaction, was organized by Founders Forum Group and Informa alongside London & Partners and Tech London Advocates.

Staniszewski, a Polish-born mathematician, co-founded ElevenLabs after working at Palantir Technologies and BlackRock. Clifford led the government review behind the UK’s AI Opportunities Action Plan and served as the Prime Minister’s AI adviser until stepping down last year.

“We have these incredible capabilities, and they are starting to get real pull, but there is also quite a lot of public suspicion of AI. It is not a very popular technology,” he said. “It is really incumbent on the industry to explain to people what the benefit actually is.”

Staniszewski described his answer as the start of a “golden era of consumer interactions,” in which callers get seamless, personalized help instead of working through a maze of automated menus.

He pointed to companies already using the technology in production: passengers on the rail-booking platform Trainline can call in when a ticket is disrupted and get an automatic refund or rebooking, and Revolut has rolled out voice agents so customers can get help with payments through its app.

“Call centers during the alerts, when the drones are coming over, need to be evacuated and moved to the shelter. In that time, that gets swapped over to voice agents that can still take those calls,” he said.

A common thread ran through both settings: a “builder” mentality inside organizations willing to experiment with the technology, whether or not the people driving it were technical themselves.

“The shift we see is from reactive to proactive. It is not only going to be, but I also have a problem, I am calling in. How incredible would it be if you could proactively reach out and get that help,“ he said.

He offered two more examples of that shift: a delivery platform proactively calling restaurants to confirm opening hours so that information stays current for customers, and agents moving from being one part of the customer journey to becoming central to it, greeting a visitor the moment they land on a company’s website.

Reading the caller’s mood

Staniszewski said the next big leap is emotional, not just acoustic. Within six to 12 months, he expects voice models to detect a caller’s mood and adjust their tone accordingly, rather than just sounding convincing.

“The huge thing we think will happen over the next six to 12 months is that the models you interact with won’t just sound good. They will be able to understand the emotion of the other side and adjust accordingly,” he said.

He said a stressed caller would get a reassuring response from the model, while an excited one would be met with matching energy.

Clifford, who led the government review behind the AI Opportunities Action Plan, used the conversation to take stock of its first year. He said the plan rests on three pillars: getting the policy framework right, driving public sector adoption and building AI sovereignty.

He pointed to early wins: scholarships to bring AI talent to Britain, and a new AI Growth Lab to help frontier companies navigate regulation. In April, the government also launched a £500 million Sovereign AI Unit to back domestic startups.

“That’s not the same as AI self-sufficiency. No country in the world will be self-sufficient in AI. But we need, as a country and as a continent, to have a seat at the table,” he said.

Compute is where progress has lagged furthest, he said, calling it the single biggest constraint on the plan’s ambitions.

He pointed to two recent deals as evidence Britain already carries real weight in the field. Isomorphic Labs, the Google DeepMind spinout working on AI-driven drug discovery, raised $2.1 billion in May, while Ineffable Intelligence, the AI startup founded by Yann LeCun, raised a $1.1 billion seed round, the largest seed round in European history.

“We only get a seat at the table if we have companies that are actually doing important things,” Clifford said. “The way you get to decide is to build it.”

He cited British computer scientist Demis Hassabis staying in London to lead Google DeepMind after its acquisition as evidence that domestic strength shapes AI’s global norms, not just its economics.

Staniszewski said the mood in the room had visibly shifted, with more companies now spinning out and scaling in Britain.

ElevenLabs’ own UK headcount illustrates that trend. It grew from 18 to 68 employees over the year ended September 2025, part of a wider expansion that took its European workforce past 90; the company now says it has around 200 staff in Britain, with more hires planned.

“The UK economy is highly geared on AI. If we adopt fast, we will benefit. If we don’t, we’ll probably lose out to international competitors,” Clifford said.

He named professional services, financial services, life sciences and creative industries as the sectors most exposed to that transformation.

The next test, in Clifford’s own account, is whether Britain’s push to be an AI maker keeps pace with the speed at which companies like ElevenLabs are scaling.