IPCEI AI: Europe's catch-up race starts in bureaucracy's waiting room

On 16 September 2026 the European Commission welcomed an initiative by 19 member states to set up the first Important Project of Common European Interest in artificial intelligence, called IPCEI AI. Coordinated by Germany, the initiative aims to cover the entire AI technology stack. Its stated goal, as reported by the Commission's Representation in Italy, is to "develop a range of decentralised AI services accessible to all". The announcement opens the pre-notification phase: from September, the 11 states that will grant state aid will begin submitting projects, after which the Commission will assess their eligibility and compatibility.

The project rests on national tracks that are still uneven. In Germany, the federal IPCEI-KI guideline sets clear limits: a maximum of €25 million per company and project (a base rate of 25% of eligible costs, up to 60% for companies), projects lasting no more than 36 months, and a technology readiness level between TRL 5 and 8. In Italy, the Ministry of Enterprises and Made in Italy closed its call for expressions of interest early this year, setting ambitious goals such as open foundation models and open-source solutions for strategic sectors like energy, defence and aerospace. The overall figures, however, are still missing: the announcement gives no total budget and neither the number nor the names of participating companies, and it has not been made public which 11 states will grant state aid to direct participants, so it is not confirmed that Italy is among them.

The historical timelines of these instruments are enough to cool any easy enthusiasm about speed. Although the matchmaking phase officially began in Berlin on 10 March 2026, and the Commission followed the preparation through the IPCEI Design Support Hub (set up in April 2025), analysts at the law firm Covington & Burling point out that an IPCEI procedure usually takes more than two years to complete, with over 18 months needed for the state aid authorisation alone. Getting from the negotiating table to real computing infrastructure will therefore take a good deal of industrial patience.

Building a European alternative in AI is a declared objective, but without a known budget or participants it is too early to judge its scope. And with procedures that usually run past two years, it remains to be seen whether the technology that comes out of it will still be cutting-edge. — Olya