NextFin News - Europe has spent the last three years telling the world how artificial intelligence must be tamed. Now it is writing the checks. In the space of a single week this summer, the European Union moved from enforcer to financier: the bloc's AI Act began applying on 2 August 2026, and five days earlier the European Commission opened a tender for up to seven AI "gigafactories" backed by as much as €10 billion in public money, aimed at unlocking at least €20 billion of private investment. The shift marks the end of Europe's regulation-first approach to artificial intelligence and the start of something more consequential — a continent-wide subsidy race to build sovereign AI infrastructure and domestic champions capable of competing with the United States and China.
The numbers are large enough to matter, and specific enough to track. On 30 July 2026, the Commission launched the AI Gigafactories call through the European High Performance Computing Joint Undertaking (EuroHPC JU). The first lot can support up to four projects, each eligible for up to €100 million of EU funding in phase one and up to an additional €400 million in phase two; the second lot backs up to three projects at up to €200 million plus €800 million. Selected facilities must begin operations within 18 months of contract signature, and the tender closes on 12 November 2026, with awards expected in early 2027.
Eighteen member states — from France and Germany to Estonia and Portugal — have already signed a joint procurement agreement with EuroHPC, committing to buy compute time from the winners. That is not a grant in search of a project; it is an anchor-customer commitment written before the factories exist.
The gigafactories sit inside a wider architecture of public money. The InvestAI initiative, announced at the Paris AI Action Summit in February 2025, aims to mobilise around €200 billion of AI investment across the EU by roughly 2030 — built from a €150 billion private-sector pledge under the EU AI Champions Initiative and €50 billion of EU funds. On 16 January 2026, the Council of the EU amended the EuroHPC regulation to put AI gigafactories explicitly inside the body's mandate, with the EU contributing 17 per cent of capital expenditure.
National governments are moving in parallel, and faster. In France, President Emmanuel Macron announced more than €109 billion in private AI and data-centre investment at the February 2025 Paris summit, including a €10 billion decarbonised supercomputer partnership with UK-based Fluidstack, and Bpifrance deployed €10 billion in March 2025 to develop the AI ecosystem. Germany's government privately backed the April 2026 merger of Canada's Cohere with Heidelberg-based Aleph Alpha — a roughly $20 billion combined entity anchored by €500 million of structured financing (about $600 million) from the Schwarz Group — explicitly to forge a single European sovereign AI champion. The United Kingdom, outside the EU, has created a Sovereign AI programme worth $675 million (£500 million), with grants of $1.3 million to $12 million for early-stage firms and a planned Strategic Assets Programme of up to $215 million.
The common thread is not industrial policy as usual. It is the state acting as customer, creditor, and infrastructure owner at once.
From Rule-Maker to Risk-Taker: Why the Pivot Happened Now
The timing is not accidental. For most of the 2020s, Europe's comparative advantage in AI was its regulatory capacity: the AI Act gave Brussels the power to set global standards. That strategy produced real influence but no companies. By 2025, the gap had become embarrassing. The United States had OpenAI, Anthropic, Google DeepMind and a Stargate-style infrastructure programme reported at $500 billion. China had state-directed compute and models. Europe had Mistral — and a rulebook.
The pivot from regulation to subsidy is therefore a correction to a failed specialisation. Europe discovered that rules travel faster than capital. The AI Act, whose core obligations began applying on 2 August 2026 under the Commission's AI Office, created a compliance moat that European firms understand better than anyone — but a moat only matters if there is a castle behind it. Subsidies are the attempt to build the castle.
There is also a procurement logic. Across 2025 and 2026, the French Ministry of the Armed Forces, the German Bundeswehr and multiple EU institutions began requiring sovereign large-language-model options in AI procurement. Once the state is the buyer of last resort for sensitive AI workloads, subsidising the supplier is no longer industrial policy; it is supply-chain security. The gigafactories' requirement that technology follow EU standards on data protection, safety and security is the procurement condition written into concrete.
The Mechanism: How Subsidies Become Capability
The transmission channel runs through three distinct instruments, each attacking a different market failure.
First, the infrastructure gap. Frontier AI is capital-intensive in a way that European venture markets have historically refused to fund. Mistral AI's answer illustrates the model: in March 2026 it secured $830 million of debt financing from a consortium including BNP Paribas, Crédit Agricole CIB, HSBC and MUFG to purchase 13,800 Nvidia GB300 GPUs for a data centre near Paris, expected to open in the second quarter of 2026. The EU gigafactories generalise that logic — public co-funding de-risks the fixed-cost base so private capital will fund the rest. The Commission's own math is a 1:2 leverage ratio: €10 billion public to unlock €20 billion private.
Second, the demand guarantee. The joint procurement agreement signed by 18 member states is the under-appreciated instrument here. A gigafactory with a guaranteed public buyer faces a fundamentally different cost of capital than one betting on commercial uptake. This is the same mechanism the US defence sector has used for decades: commit to purchase, and the private sector will build.
Third, the regulatory moat. The AI Act does not only constrain European firms; it raises rivals' costs for any non-EU provider selling into the bloc. Sovereign providers — Mistral, Aleph Alpha, and their successors — can market "compliance by design" as a product feature. The second-order effect is that European AI becomes, in part, a jurisdictional service: the value is not only the model but the legal safety of running it inside Europe's rule set.
Cyclical or Structural: This Is a Regime Shift, Not a Spending Wave
The critical question is whether Europe's subsidy wave is cyclical — a burst of spending that will recede — or structural. The evidence points to structural, for three reasons.
The first is permanence of the instruments. The EuroHPC regulation was amended by the Council in January 2026 to embed gigafactories in the body's mandate; the AI Act's enforcement machinery is now operational; the IPCEI on AI has pre-selected roughly 150 projects across 18 member states, with a matchmaking phase launched in Berlin on 10 March 2026. These are not budget-line items that expire with an election cycle; they are institutional fixtures.
The second is the procurement lock-in. Once defence ministries and EU institutions architect their systems around sovereign models, switching costs become enormous. A government that has classified workflows running on a domestic LLM does not migrate back to a US API for reasons of budget convenience. The demand is sticky.
The third is the competitive dynamic that will not self-correct. The US lead in frontier AI is reinforced by scale: more compute produces better models, which attract more capital, which buys more compute. Europe cannot converge on that loop through market forces alone — the loop is already closed around American champions. Only sustained public intervention can break in. That is the definition of a structural shift: a gap that market forces widen rather than close.
The cyclical component exists — individual subsidy programmes can be cut, and the €200 billion InvestAI target depends on private pledges that may not all materialise — but the direction of travel is set. Europe has decided that AI is a strategic industry, and strategic industries in Europe get state aid.
The Counter-Thesis: Too Little, Too Slow, Too Fragmented
The strongest argument against this strategy is not that subsidies are wrong, but that they are mismatched to the speed of the industry. Three objections carry real weight.
Scale is the first. Even if every euro of the €200 billion InvestAI target is deployed, it is one order of magnitude smaller than the reported $500 billion US Stargate programme — and that US figure has reportedly faced turbulence and scaling back. But even a scaled-back American programme dwarfs Europe's public anchor. A spokesperson for industry association DigitalEurope put the imbalance bluntly:
With €150 billion expected from private sources and only €50 billion from public funds, none of it fresh, Europe needs to be more ambitious.
Speed is the second, and it is the more damaging objection. IPCEI state-aid approval typically takes more than two years, including roughly 18 months for the Commission's assessment. AI capability doubles on a much shorter clock. A gigafactory tender that closes in November 2026 and awards in early 2027, with operations beginning up to 18 months after signature, delivers capacity in 2028 or later. In AI time, that is a different era. The risk is that Europe subsidises the previous generation of hardware.
Fragmentation is the third. National subsidy races can undermine the single market they are meant to strengthen: French models on French clouds, German champions on German infrastructure, a UK fund for British companies. If each country builds its own sovereign stack, Europe ends up with five small sovereign stacks instead of one large competitive one. The joint procurement agreement is the antidote, but it covers compute access, not the subsidy decisions themselves.
These objections are serious but not fatal. They argue for better design — faster approval tracks, pooled rather than national subsidies, procurement commitments that span borders — not for abandoning the effort. The counter-thesis would only prevail if Europe's programmes consistently failed to deliver built capacity.
The Falsifying Signal
The thesis that Europe's subsidy pivot is a structural shift worth following rests on one observable test: capacity delivered, not money announced. The specific falsifying signal is this — if by the end of 2027 fewer than three of the seven planned gigafactories are under construction, and Europe still has no frontier model in the global top 10 outside Mistral, then the subsidy wave is rhetoric rather than regime change. A softer early signal is the tender itself: if the 12 November 2026 call closes with fewer than four qualified consortia for the first lot, the anchor-demand story is weaker than it looks.
Who Wins, Who Loses, and What to Watch
The immediate beneficiaries are clear. European cloud and data-centre operators gain a public-order book; sovereign AI labs gain both capital and a protected customer base; and the equipment suppliers — the chipmakers selling the processors that the gigafactories must deploy at two to four times Europe's current most powerful AI factory — gain a guaranteed buyer. The exposed are the US hyperscalers selling into Europe's public sector, where sovereign-preference procurement will increasingly shut them out of sensitive workloads, and the smaller European AI firms that cannot meet the scale thresholds the gigafactory tiers require.
The time-horizon split matters. In the short term — through 2027 — expect announcements to outpace concrete: tenders close, awards are made, and markets price the sentiment. In the medium term — 2027 to 2030 — the test is whether the €10 billion gigafactory commitment actually leverages the promised €20 billion of private capital and whether the 17 per cent EU co-funding rule attracts projects rather than deterring them with complexity. In the long term, the question is whether Europe can convert subsidised infrastructure into subsidised capability: compute does not produce models; talent does. Europe's chronic shortage of AI research talent remains the binding constraint no subsidy has yet addressed.
Three scenarios frame the path. The base case: four to five gigafactories break ground by 2028, Europe consolidates around two or three sovereign model providers, and the bloc becomes a regulated, sovereign niche rather than a frontier leader — a meaningful outcome, but not a winning one. The upside case: the joint procurement mechanism works, private capital follows at better than 1:2 leverage, and a European champion emerges with genuine frontier capability by 2029. The downside case: national fragmentation and approval delays mean the money is committed but the capacity arrives a generation late, and Europe has paid for the infrastructure of the previous AI era.
Europe has correctly diagnosed that it cannot regulate its way to AI sovereignty, and it has finally reached for the only tool that works — the state as anchor customer. But writing the cheque is the easy part. The hard part is spending it fast enough that the money buys capability rather than nostalgia.
Explore more exclusive insights at nextfin.ai.
