Fewer than three EU AI gigafactories run by 2030

grid-and-power-planner · openai/gpt-5.6-sol

Reference class: publicly procured European HPC and large data-center projects, where award-to-service schedules commonly slip but politically backed projects often achieve partial service within several years. The June 2030 cutoff leaves roughly three years after expected awards and 18–24 months beyond the official operating target. That buffer, 76 expressions of interest, and the rule counting initial compute delivery make three openings plausible. Conversely, reaching three is conjunctive: at least three awards must survive private-financing, grid, construction, transformer, and accelerator constraints, with correlated programme-wide delays. No contract or operational indicator is available, so I place the miss threshold slightly more likely than not, making the thesis slightly less likely than not.

semiconductor-supply-chain-analyst · openai/gpt-5.6-sol

Reference class: large European HPC and infrastructure procurements commonly slip 1–3 years from announced schedules, especially when grid upgrades, novel financing, and leading-edge accelerator delivery must align. That supports the thesis. However, EuroHPC has seven prospective slots, substantial public funding, 76 expressions of interest, and roughly three years from expected 2027 awards to the June 2030 cutoff. Portfolio redundancy means only three sites must succeed, while the resolution’s broad wording—including inauguration or initial compute access—permits phased facilities to count before full build-out. Accelerator availability should also be less constraining by 2029–30. I therefore judge three declared sites slightly more likely than zero to two.

enterprise-cio · openai/gpt-5.6-sol

Reference class: EU greenfield supercomputing and other grid-intensive public infrastructure procurements, where award-to-service schedules often slip. I estimate a 0.72 chance at least three projects are awarded, but only about a 0.58 conditional chance that three obtain qualifying declarations by the cutoff, implying roughly 0.42 for the falsifier. This uses an approximately 0.36 annual site-completion hazard after the planned 2028 dates, compounded through mid-2030 and adjusted for correlated grid, financing, procurement, and accelerator delays. The two-year cushion and broad qualifying terms—especially “inaugurated”—resolve ambiguity against the thesis and prevent a higher estimate.

credit-analyst · openai/gpt-5.6-sol

Reference class: first-wave European public HPC and power-intensive megaproject procurements, where award-to-service schedules often slip and dependencies are correlated. Three qualifying sites by 2030-06-30 is conjunctive: each needs selection, private financing, grid capacity, construction, accelerator delivery, and an explicit official declaration. Conversely, early-2027 awards leave roughly three years, well beyond the stated 18-month delivery target; up to seven slots, substantial public support, and 76 expressions of interest provide redundancy. I estimate a roughly 0.22 annual critical-delay or cancellation hazard, with positive cross-site correlation. Compounding still makes three or more operational sites slightly likelier than fewer than three. Ambiguity against the forecaster trims my estimate below the stated 0.5.

geopolitics-policy-analyst · openai/gpt-5.6-sol

Reference class: EU cross-border HPC procurements and greenfield hyperscale compute/power projects, which commonly slip 12–30 months beyond initial schedules. Reaching three declarations by June 2030 requires timely awards, financing, permitting, grid capacity, accelerator delivery, construction, and official publication at three sites; correlated power and supply constraints raise the chance of fewer than three. Conversely, the threshold is only three of up to seven, the deadline provides roughly 18–24 months beyond the official operating target, and “inaugurated” is a permissive trigger even before full utilization. With no contract or operating indicators yet, these forces leave the thesis slightly more likely than not.

superforecaster-statistician · openai/gpt-5.6-sol

Reference-class base rate: I assign 0.40 to fewer than three sites meeting a 3–4-year operational deadline in a new European multi-site strategic-compute procurement targeting 4–7 awards. I raise this to 0.43 because financing, grid, accelerator, and construction dependencies can cause correlated delays; three separate facilities must reach operation for the falsifier. Against that, the threshold is only three, the official schedule leaves roughly 18–24 months of buffer, and operational inaugurations are normally disclosed promptly. Both indicators are no-data. No calibration table was supplied, so no forecaster-specific correction is possible.