On August 4, 2026, Estonia's Ministry of Justice and Digital Affairs confirmed it will commit up to €20 million to purchase artificial intelligence computing services between 2028 and 2032, joining 17 other EU member states in a joint procurement agreement under the EuroHPC Joint Undertaking's new "AI Gigafactories" call (Justiits- ja Digiministeerium; EuroHPC JU). The money buys Estonia a guaranteed claim on compute time from whichever consortium wins the right to operate a facility with an Estonian branch — not ownership of a data centre, but a five-year purchase commitment.
The parent programme dwarfs Estonia's slice of it. EuroHPC JU opened the call on July 30, 2026, offering up to €10 billion in combined EU and national public funding to build as many as seven AI Gigafactories across the bloc, expected to unlock at least €20 billion more in private investment. Bids close November 12, 2026; contracts are expected in early 2027, with the first Gigafactories operational within roughly 18 months of signing — likely 2028, precisely when Estonia's purchase window begins.
From host ambitions to a modest buy-in
This is Estonia's third AI-infrastructure move in a year, and each has been smaller and safer than the last. In August 2025, the government backed a Baltic-Poland "AI GigaFactory" bid promising over 100,000 processors, with Estonia's own contribution pencilled at roughly €60 million. By December 2025, Tallinn had shifted to a Nordic consortium led by Nokia alongside Finland and Latvia, with Minister Liisa-Ly Pakosta arguing that "this isn't just about building infrastructure, but creating a platform for future AI services, including the necessary skills" (Invest in Estonia). Neither bid produced a signed facility.
What Estonia signed onto this August is the more conservative half of that bargain: buying capacity as a customer inside a continent-wide EU procurement, rather than betting tens of millions on winning a site. That is a sensible narrowing of ambition, not an abandonment of it — Tallinn keeps a claim on future Gigafactory output without carrying construction risk, and EuroHPC's rules explicitly allow cross-border, distributed facilities, so the winning consortium need not build in Estonia at all.
The sovereignty case, stated fairly
The strongest argument for public money doing this is concrete, not abstract. A government report released December 4, 2025, "Tulevikukindla arvutusvõimekuse tagamine" ("Ensuring future-proof computing capacity"), estimated Estonia will need at least 2,105 GPUs by 2030 to keep training and running AI models for science and public services domestically (Justiits- ja Digiministeerium). Pakosta's framing in that report deserves to be taken seriously: relying entirely on foreign data centres means, in her words, "surrendering a foundational element of Estonia's digital state." A country that already routes some 2.2 billion transactions a year through X-Road across tens of thousands of organisations has a real stake in not putting the next layer of that stack — model training, health-data processing, crisis coordination — behind a foreign vendor's terms of service. Ott Velsberg, Estonia's head of AI and data affairs, made the operational version of the same point: without local capacity, the state cannot securely train models on sensitive health, security, or social data. That is a genuine infrastructure argument, echoing the same logic that justified building X-Road on state-controlled rails in the first place — not merely industrial-policy boosterism.
Where the case runs thin
But €20 million is a rounding error against a 2,105-GPU target, and the Gigafactories programme does not resolve the dependency the sovereignty argument is meant to address. Every large EU-backed AI cluster announced so far — Mistral's, Deutsche Telekom's Munich buildout, the Nscale/Microsoft project at Sines — runs on Nvidia hardware, and the Centre for European Policy Studies has warned this creates an "Nvidia dependency trap": compute that is physically European but technologically dependent on a single American supplier (Tech Policy Press). Europe still produces under 10% of the world's advanced chips, and the EU's own auditors project the bloc will miss its 2030 chip-production target by a wide margin. Building a data centre inside any member state's borders doesn't change who controls the accelerators or the software stack running on them — physical sovereignty is not the same as technological sovereignty, and officials should stop implying the two are interchangeable.
That gap is a reason to be precise about what a €20 million buy-in actually accomplishes: near-term, contracted access to compute for a country whose researchers and agencies are genuinely capacity-constrained, plus a stake in EU-level bargaining leverage over pricing that a nation of 1.4 million people could never negotiate alone. It is not a sovereignty guarantee. The fairer test of this policy is narrower than "does Europe control AI": does Estonia's digital state get cheaper, more reliable access to the GPUs its own government says it needs by 2030? On that test, a modest, five-year purchase commitment folded into a much larger EU procurement is a proportionate way to hedge — better than a solo national supercomputer bet, and more honest than the two host-country pledges that came before it and went nowhere.