When the Commission Picks Five Winners for AI
On 30 July 2026 the European Commission opened a call for tenders — titled "EU launches AI Gigafactories call to boost Europe's computing capacity and unlock more than €30 billion in investment" — to build up to five "AI Gigafactories" financed by the new InvestAI Facility, a €20 billion European fund for public-private partnerships in frontier-AI compute. The Commission's AI strategy page describes the Gigafactories as "large-scale facilities dedicated to the development and training of next-generation AI models containing trillions of parameters," bringing together "over 100,000 advanced AI processors."
The press conference was careful. The Commission speaks of "AI Gigafactories" rather than "industrial policy," "catalytic public investment" rather than "subsidies," and "unlocking private capital" rather than "picking winners." The critique is straightforward: a Brussels committee is about to pick up to five sites out of up to twenty-seven member-state candidates, on the basis of criteria none of which are exchange prices at the relevant margin — and the Commission's own strategy page already documents a parallel programme of 19 AI Factories and 13 Antennas across 16 Member States, on top of which the five Gigafactories will be layered.
What the Story Claims
The dominant narrative is generous. Europe is "behind" the United States and China on frontier AI compute. The bloc risks losing the next decade of industrial competitiveness. The Commission is "stepping up" with patient, catalytic investment that finance alone will not provide. The five Gigafactories are framed as Europe's equivalent of the United States' CHIPS Act or China's national compute buildout — pooled at the union level so no single member state carries the full political risk.
The framing is missing the Austrian question. The Commission is choosing which sites will receive a competitive grant, on the basis of an evaluation methodology the Commission maintains. Each grant is an administered price — the price the Commission is willing to pay for AI infrastructure sited in a particular place. It is not, and cannot be, a market price.
The Austrian Diagnosis
Ludwig von Mises, whose 1920 essay "Economic Calculation in the Socialist Commonwealth" identified why centrally-planned systems cannot rationally allocate capital goods without market prices for them, would have read the call first. The sites are bundles of location-specific capital — power-purchase agreements, grid connection rights, water-cooling, talent pipelines. Each input is priced in a different market the Commission does not clear. The Commission's choice is the Commission's calculation of which bundles will deliver the most "frontier compute" per euro.
That calculation is wrong by construction. The sites that win will be the sites whose member states wrote the most competitive proposal on the criteria the Commission weighted most heavily. The criteria will be visible. The actual marginal cost of an extra gigawatt-hour at the chosen site, the marginal hour of senior AI-researcher time, the marginal unit of grid capacity freed — these will be invisible. They are facts held by operators, regulators and engineers on the ground, and no proposal can transmit them in a way the scoring framework can value. The Commission's winners are the winners of a beauty contest whose judges cannot see the actual beauty.
Friedrich Hayek, whose 1945 essay "Use of Knowledge in Society" argued that the knowledge relevant to economic calculation is dispersed across millions of actors, would have sharpened the cut. The Commission cannot know whether the winning site is genuinely the lowest-cost place to put a frontier AI buildout, or whether another losing site would have carried a lower marginal watt, a lower marginal researcher, a faster marginal grid connection. Each of these is a different ranking, and they all share one feature: they depend on facts the Commission cannot enumerate.
Frédéric Bastiat, whose 1850 essay "That Which Is Seen, and That Which Is Not Seen" gave economics its sharpest distinction between the visible effect of a policy and the cost it shifts onto someone else, would have closed the loop. The seen is the Gigafactories, the press conferences, the member-state proposals. The unseen is the productive AI investment the selection distorts: the member states that lose will slow-walk local AI projects that depended on Brussels legitimising their location, the AI workloads priced against the chosen sites rather than the marginal cost of compute everywhere on the continent.
The 1982 Parallel
The structural precedent is the European response to Japan's Fifth Generation Computing Programme (1982–1992). When Japan's Ministry of International Trade and Industry committed ¥100bn to a state-led push for symbolic and parallel inference machines, Europe responded with ESPRIT (European Strategic Programme for Research in Information Technology, 1984) — the EU's framework that ran in successive multi-year phases and channelled billions of euros in cumulative funding into R&D consortia. The parallel is direct: a strategic national AI programme in one bloc, met by a Europe-wide consortium-and-grant framework in another, both justified by the gap they would close to the leading bloc.
The result was structural. ESPRIT funded mainframe-class knowledge engineering; the desktop computing revolution happened elsewhere; the personal computer era was won by US actors (and to a lesser extent Taiwanese and Korean ones) that did not compete for ESPRIT grants. The lesson is not that European coordination is futile. It is that a committee cannot choose winners for a market that has not yet built them. The Commission's 2026 AI Gigafactories sit in the same posture: the chosen sites will be locked into 2026's stack, while the frontier model that matters in 2028 may run on a different architecture.
Why This Matters for Sound Money
Part 1 of Rails to Freedom identifies the calculation problem as the structural constraint on any authority that substitutes an administered price for a market price. The AI Gigafactories programme is the textbook example: an envelope set by the Commission against a methodology maintained by the Commission, applied to sites whose marginal cost the Commission cannot see. The winners will be right by construction only if Europe's actual frontier-AI demand clears at the methodology's forecast — and the methodology's forecast is the committee's forecast, set before the market has built the demand.
Part 4 of the book identifies the on-chain monetary primitive that does not require a committee to defend its unit of account. The contrast is structural. EigenLayer's restaking on Ethereum mainnet is a market-priced budget for decentralised verification and computation: ETH stakers opt into Actively Validated Services — bridging, data availability, oracle work and decentralised AI inference — and the rewards each service pays emerge from supply and demand for that workload, defended by slashing enforced by code rather than by Commission scoring.
What Markets Are Already Doing
The on-margin alternative to Brussels picking Gigafactory sites is the permissionless re-staking market on Ethereum mainnet. EigenLayer lets any ETH staker delegate their staked ETH to Actively Validated Services; each AVS sets its own reward schedule in ETH, priced against the workload's marginal value at every block, defended by slashing that fires automatically if the AVS misbehaves. The result is a market-priced budget for the same services the Gigafactories are meant to subsidise, with no scoring framework and no proposal.
Hayek would have seen the contrast directly. The Commission prices the marginal value of its chosen sites once, against a methodology that cannot aggregate the dispersed knowledge of Europe's grid operators, power producers and AI researchers. EigenLayer prices the marginal value of each AVS every block, against an order book of stakers and AVS operators who each know their own cost of capital, and disburses against a workload the chain enforces. The Commission's price is right by definition only if the methodology is right. EigenLayer's price is right by construction because arbitragers are paid to make it so.
Looking Ahead
The Commission will run the AI Gigafactories call as a formal procurement over the months ahead, evaluate proposals through early 2027, and announce its winners by spring. The press conference will read as measured, the EU logo will go on the door, and the chosen sites will frame themselves as Europe's picked locations for frontier AI. The unseen — the productive AI investment the selection distorts, the losing member states that defer local compute, the capital redirected into co-investment obligations — will appear over the next decade as slower pan-European compute diffusion. EigenLayer on Ethereum mainnet will keep pricing the marginal value of each AVS at every block.