AWS Europe Chief Warns 4 in 10 Startups Could Quit Europe

By Daniel Rao · October 22, 2026

In a single week, Europe got two diagnoses from the two companies most eager to sell it AI infrastructure, and they flatly contradict each other. On October 21, Tanuja Randery — AWS's managing director for Europe, the Middle East and Africa — told the Reuters Global Markets Forum that the continent risks a genuine startup exodus: she claimed that nearly four in ten European startups would consider leaving the region to scale, and that among the fastest-growing ones, "it's over half." That same day, Nvidia's Jensen Huang stood on a stage in Berlin and promised the precise opposite: "Every company in Europe will become an AI company."

One of them is wrong. Or, more likely, both are telling a version of the truth that flatters their own sales pipeline.

Rows of server racks in a data centre aisle — the kind of compute infrastructure at the heart of Europe's AI-infrastructure race. File photo; not an AWS facility. Photo: Derrick Coetzee / Wikimedia Commons (CC0).

Forty cents on the euro

The line that will travel furthest from Randery's interview is the one about money. "European businesses are telling us that they could spend as much as 40 cents to every euro on regulatory compliance instead of technology," she told the Reuters Global Markets Forum. "Imagine if that was reinvested in technology."

Pause on the grammar of that claim. It is not a statistic. It is what "European businesses are telling us" — a secondhand anecdote from an AWS executive, delivered without a methodology, a sample size, or even a named company. It is a devastating number precisely because it sounds like a statistic, and Randery knows it. The 40-cents figure is the hook; the policy underneath it is the EU AI Act, which Reuters in the same piece described as "widely seen as the world's most stringent AI regulatory framework."

The other headline figure is Randery's exodus claim: nearly four in ten European startups would consider leaving the region to scale — and "among the fastest-growing, it's over half." Both figures come, she said, from "an AWS survey of companies." And on funding, her message was blunter: "Startups say to us that they're reconsidering locations because they want funding and support, and they're not getting enough funding here."

The survey nobody can see

We read the Reuters interview end to end and checked every relay we could find; nowhere — not in Reuters, not in the wire relays — is the survey named, its sample size given, or a link offered. The AWS survey figures are Randery's claims about AWS's own unpublished survey. They are not independently verified, and treating them as verified would be shoddy. That does not make them false — startup surveys across the continent keep landing in the same place — but it makes them advocacy numbers, and they should be read as such.

The same caution applies to the AI-adoption figures in the Reuters piece. The article, in its own paraphrase rather than as a quotation, reported that more than half of European businesses are using AI — but that only one in five is clear about AI strategy, has an AI governance model, and is invested in processes that would fundamentally reshape operations. That is a reported figure, not a verbatim quote, and the gap it describes — experimentation everywhere, transformation almost nowhere — is the part of Randery's case that rings truest. It also happens to be the part where AWS has the most to sell: "I think inferencing is where European businesses need to really go invest," she said. Yes. Inference runs on cloud. Cloud is what she sells.

The exodus is already happening

If Randery's survey is opaque, the independent evidence for startup flight is not — and it is older and uglier than this week's headlines.

Take 11x, the London-based AI sales startup. In September 2024 it raised $50 million from Andreessen Horowitz — and then relocated to the United States, largely because its investors required it to set up shop in the Bay Area, as the Economic Times reported at the time. That was 2024, before the AI Act's heaviest obligations applied. The money moved first; the regulation followed as justification.

Atomico's State of European Tech 2024, citing an academic study, found that 6 per cent of 11,000 European VC-backed startups founded between 2000 and 2014 had relocated across borders by 2021 — and 85 per cent of the relocators went to the United States. The kicker: those relocated startups accounted for 17 per cent of total startup value at exit. A sliver of companies leaves; a sixth of the value goes with them. And once gone, the jobs follow the money: relocated firms employed 65 per cent of their workforce outside their founding country at IPO, versus 92 per cent inside it for companies that stayed.

The 2025 edition, covered by Sifted, found that 15 per cent of surveyed founders had moved their headquarters to another country and 42 per cent had considered it but stayed. Of those who left, 57 per cent went to the US; of those who merely thought about it, 73 per cent had the US in mind. Funding was the main driver — cited by 66 per cent of the considerers and 51 per cent of the movers — with customers close behind at around 40 per cent.

"Despite all the progress over the past decade, there is real concern across the ecosystem about the way forward... there are questions being asked about Europe's attractiveness as a place to start and scale a tech company," Tom Wehmeier, a partner at Atomico, told the Economic Times. And Alice Bentinck, the co-founder and chief executive of Entrepreneurs First, put the American pull in brutally commercial terms to Sifted: "American customers are very fast in terms of procurement experimentation. They have very deep pockets and it means our companies are getting to hundreds of thousands of dollars in annual recurring revenue (ARR) within three months of landing in the US." Regulation may be the complaint founders voice in surveys. Deep pockets are the reason they board the plane.

Brussels' answer: push the deadlines

The European Commission's actual answer to the compliance chorus is the Digital Omnibus on AI — Regulation (EU) 2026/1744 — adopted on July 8, 2026, published on July 24, and in force from July 27. Its centrepiece is delay: high-risk obligations under Annex III, covering hiring, credit scoring, education and biometrics, slide from August 2, 2026 to December 2, 2027. Annex I obligations, for AI embedded in regulated products, move to August 2, 2028. Public authorities get until August 2, 2030. Small-business relief has been extended to "small mid-caps" with fewer than 750 staff.

But the fine print bites back. Article 50's transparency rules were not delayed — they apply from August 2, 2026, and existing generative systems get only a grace period to December 2, 2026 for machine-readable marking. And from December 2, 2026, new bans take effect on non-consensual intimate imagery and CSAM. Brussels gave the industry 18 months of breathing room on the hardest obligations, then kept the transparency provisions on the original schedule. Whether that is a compromise or a trap depends on which deadline your product hits first.

Brussels has also been visibly active on the money side: earlier this month, the EU's RAISE initiative drew DeepMind, Nvidia and Mistral pledges for AI-for-science — a genuine signal that the Commission understands it must be seen building, not just banning.

What Brussels has not done, as of this writing, is respond to Randery at all. We checked for a Commission statement on her remarks and found none. Europe's AI rulebook is being litigated in public by American cloud vendors, and the regulator is not in the room.

None of this exasperation is new. Back in January 2025, after Donald Trump's $500 billion Stargate announcement, European CEOs were already calling for AI-infrastructure investment, according to PYMNTS' relay of CNBC's reporting — with IBM's Arvind Krishna calling the AI Act "a sledgehammer" on the same panel. The frustration predates Randery by nearly two years; she is just the first person to put a 40-cents number on it.

The Berlaymont in Brussels, seat of the European Commission — the regulator behind the AI Act at the centre of Randery's warning. Photo: tpholland, CC BY 2.0, via Wikimedia Commons.

The money case is real — and so is the counter-case

Strip away the vendor advocacy and Randery's funding argument survives contact with the data. Dealroom's "State of AI in Europe — The Invisible Giant," published in March 2026 with Prosus, recorded a record $21.8 billion of venture funding into European AI startups in 2025, up 58 per cent from $13.9 billion in 2024 and roughly ten times the $2.3 billion of 2016. But the late-stage picture is colossally lopsided: Europe put $12 billion into late-stage AI rounds in 2025 against $141 billion in the US — a twelve-to-one gap — even as the two regions sat near parity at early stage, $4 billion versus $5 billion. The US has 56 AI decacorns to Europe's nine, and 33 notable AI models to Europe's three. Europe's startups can be born; they cannot be grown.

Dealroom's June 2026 Transatlantic Deep Tech Report sharpens the point: US deep tech funding hit $141 billion in 2025, nine times Europe's $15.5 billion. The gap is two-to-one at early stages and sixteen-to-one at later stages. Europe actually closes more deep tech rounds than the US — but American dollars run nine times higher, and the Bay Area alone, with 16 per cent of deep tech startups, takes more than half the capital.

And the capital that does flow is foreign. The European Deep Tech Report 2026, from Lakestar, Walden Catalyst and Hello Tomorrow, found that 70 per cent of late-stage funding for European deep tech comes from outside Europe, mostly the US; that only 54 per cent of funding for European rounds above $15 million comes from within Europe, versus 80 per cent in the US; and that more than 80 per cent of European deep tech exits are acquisitions, with American buyers taking the majority. The European Investment Fund estimates a €70 billion funding gap between Europe and the US at the growth stage; its answer, the €15 billion European Tech Champions Initiative 2 fund-of-funds — targeting around 100 VC funds, with a first close expected this past summer — aims to unlock up to €80 billion in scaleup investment. Europe's solution to American capital is, in effect, a state-backed attempt to manufacture more of it.

The counter-case has a name: Mistral. In September 2026 the French lab closed a €3 billion Series D led by Samsung at a €21 billion-plus valuation — on top of a €1.7 billion Series C led by ASML in September 2025 and €722 million in debt financing in March 2026 for its first large-scale data centre near Paris, with a16z, Nvidia, Salesforce Ventures, ASML and Bpifrance in the mix. Europe's champions can raise at home. And Pigment's $250 million raise at a $2.3 billion valuation shows French software scaling on European terms. The honest reading is not that funding doesn't exist — it is that late-stage capital is foreign, exits are American, and it takes a sovereign champion to break the pattern. That is a subtler problem than Randery's survey suggests, and a harder one.

Huang's Europe

Which is what makes Jensen Huang's counter-diagnosis worth taking seriously, even at vendor-discount rates. On October 21, Huang told GTC Berlin that "every company in Europe will become an AI company," arguing that no region has dedicated itself to "making mechatronics, electronics and mechanical systems at the level of precision and craft as the European region." His pitch: Europe's edge is physical AI — the robots and industrial systems the continent already builds better than anyone. "To be part of this AI economy, you don't want to just be consuming the AI; you want to be producing the AI. We could do all of that here in Europe," he said.

He made the case with props. Huang's Berlin stage was flanked by duck-shaped robots — trained by French startup Pollen Robotics with Hugging Face — doing somersaults for the audience, a spectacle no Brussels committee room could ever produce. Germany's Agile Robots, meanwhile, announced plans for 1,000 to 2,000 petite humanoids next year built with Nvidia, with production starting late this year, according to chief executive Zhaopeng Chen speaking to Reuters. Nvidia has committed nearly $10 billion to Europe's tech ecosystem. It was the same week Huang warned Berlin that Europe gets "no second chance" on AI — carrot and stick, days apart.

Two vendors, two diagnoses

Here is the judgment this story requires. Randery runs AWS's business across Europe, the Middle East and Africa; she has run it since 2020, was awarded a CBE in last year's King's Birthday Honours, and has spent years arguing — as she did to SiliconANGLE in 2024 — that AI could unlock trillions of euros of European economic value. Her remedy for the exodus is instructive: "I think inferencing is where European businesses need to really go invest." Inference runs on cloud infrastructure. She sells cloud infrastructure. Huang's remedy is equally convenient: Europe should produce AI — on Nvidia's chips. Both are right about the money. Both are vendors.

But vendor math and real math are different things, and the real math — Dealroom's twelve-to-one late-stage gap, Atomico's relocated-value figures, the EIF's €70 billion hole — checks out independently of either executive's sales deck. Europe's problem was never only regulation; it is that the growth-stage money lives in California, the buyers are American, and the compliance bill arrives in Brussels. Randery's survey numbers may be opaque, but the direction of travel they describe is the one every independent dataset confirms: Europe builds AI companies and loses them.

That is also what makes Huang's promise the more interesting bet. If "every company in Europe will become an AI company," the question is where those companies will be headquartered when they grow up. An exodus and an AI-everywhere Europe are not opposites — they are the same future, with the profits booked elsewhere. The two diagnoses agree on the disease. They just sell different cures.

"We need to go faster in Europe," Randery said. "It isn't about progress versus safety. It's an 'and,' not an 'or.'"

On that much, at least, both vendors agree.

Sources: Reuters (reporting by Mehnaz Yasmin and Deborah Mary Sophia); SRN News; PYMNTS; Channel NewsAsia; Dealroom; Sifted; Atomico; the Economic Times; EU-Startups; itanlabs; McKinsey; SiliconANGLE. Quotes are from the Reuters Global Markets Forum interview of October 21, 2026, and the Reuters wire from GTC Berlin (Toby Sterling and Hakan Ersen) of the same day, unless otherwise attributed. Analysis is the author's own.

Daniel Rao covers AI, autonomous agents and the business of technology for TechPulse USA.

 

Update, Oct. 22: the British front just escalated

The warning just got a British sequel. The Sovereign AI Leadership (SAIL) taskforce — Nvidia, Microsoft and Cisco — told Labour ministers Britain has roughly 12 months to win public consent for new datacentres or face the US-style organised opposition that has already halted fully permitted projects across the Atlantic. Its pitch: a formal community-benefits "compact" giving councils and mayors recommended terms, with an indicative contribution scaled to each site's size — money or facilities — and letting host councils keep more tax revenue. Microsoft, which has pledged $30B (£22B) for UK AI infrastructure (Acton built, Newport under construction), is engaging local and national leaders directly. The pushback is already organised: PauseAI UK says payments would look like an undignified attempt to buy support, and Action to Protect Rural Scotland counts 15 active community campaigns. The government points back at the local planning system and notes councils in AI growth zones will retain additional business rates. Randery's exodus warning and SAIL's 12-month clock are the same diagnosis from two directions: Europe's AI buildout is losing the consent of the governed. Nvidia, Microsoft: 12 Months to Stop UK Datacentre Protests

European startup exodus: your questions, answered

Who is Tanuja Randery?

Tanuja Randery is AWS's EMEA chief — the named source behind the 4-in-10 warning, speaking at Reuters' Global Markets Forum on October 21.

What did the 4-in-10 survey actually say?

The unpublished survey found 40 cents of every euro of AI investment now goes on compliance, and 4 in 10 European startups might leave Europe to scale. The methodology has never been published — treat it as a claim, not a census.

What is Huang's counter-case?

The same day Randery warned of an exodus, Nvidia's Jensen Huang told Berlin that "every company in Europe will become an AI company." Two vendors, two diagnoses: Randery says regulation is driving startups out; Huang says the buildout is inevitable anyway.

What are AI growth zones?

Designated UK areas for fast-tracked AI data-centre development, where host councils retain additional business rates generated by the facilities — the government's answer to the community-benefits question.