Next Frontier AI: Germany's Bet on Its Own AI Labs
Germany is using public money to close a gap that has long persisted in foundation models: its own labs on par with OpenAI, Anthropic, or DeepSeek. The Next Frontier AI Challenge, run by the federal agency SPRIND, awards 125 million euros across three stages, while SAP and Cohere build their own labs in parallel. This article explains how the process works, what it means for AI sovereignty, and what European companies should take from it.
Germany's federal innovation agency SPRIND launched the Next Frontier AI Challenge on 25 May 2026, a competition with 125 million euros in non-dilutive funding. The goal is up to three European frontier AI labs that develop their own foundation models. The process runs across three stages over 24 months: 10 teams at 3 million euros each, then 6 teams at 8 million euros each, finally 3 teams at 15.5 million euros each. The application deadline closed on 1 June 2026, with pitches on 24 and 25 June. SPRIND's head of challenges, Jano Costard, explicitly calls the sum only the first step, with each finalist lab then expected to raise 250 million to 1 billion euros privately. In parallel, SAP is investing over 1 billion euros in Freiburg-based Prior Labs, and Cohere has acquired Aleph Alpha. On its own, 125 million euros is not enough to match US and Chinese labs, since Europe accounts for under 10 percent of global AI venture capital. The leverage lies in private follow-on capital and new research approaches.
Germany bets on its own frontier AI labs
Germany wants to use public money to build its own top-tier AI labs. The federal innovation agency SPRIND launched the Next Frontier AI Challenge on 25 May 2026, a competition worth 125 million euros . The application deadline already closed on 1 June 2026 at 12:00 CET, pitches take place on 24 and 25 June, and the jury decides afterwards by consensus.
The backdrop is Europe's dependence on US and Chinese providers, sharpened by DeepSeek's V4 model in April 2026. How wide the gap in foundation models really is was assessed by innobu in its piece on China's AI models and their benchmark promises .
Germany is leading this because we have no time to waste in waiting for other actors to get into that space.
Jano Costard, Head of Challenges at SPRINDHow the Next Frontier AI Challenge works
The competition is a multi-stage elimination process over 24 months. Rather than taking equity, SPRIND awards non-dilutive capital, so teams keep control over research direction and intellectual property. After each stage, the jury decides on advancement, a pivot, or elimination.
| Stage | Teams | Funding per team | Duration |
|---|---|---|---|
| Stage 1 | up to 10 | 3M € each | 7 months (from July 2026) |
| Stage 2 | up to 6 | 8M € each | 8 months |
| Stage 3 | up to 3 | 15.5M € each | 9 months (to autumn 2028) |
Eligible applicants were teams headquartered in the EU, EFTA, Israel, or the UK, from startups and established companies to universities and research institutions. Beyond the money, teams receive compute access through bulk-negotiated contracts, support with legal work, hiring, and compliance, as well as access to expert networks.
The 125 million euros are not a one-off grant but a tournament. Whoever reaches stage 3 should be far enough along, with pilots, eval suites, a team, and an investment-grade data room, that private investors finance a large follow-on round.
German and European perspective
The public challenge is only one part of a broader movement in the German AI market. Private players are building their own labs in parallel, and the federal government adds further funding. For the first time in years, a visible European answer to the dominance of the large providers is taking shape.
SPRIND: public challenge
125 million euros in non-dilutive funding for up to three labs. The state sets the frame and shares the risk without taking equity.
SAP: Prior Labs
SAP is investing over 1 billion euros across four years in Freiburg-based Prior Labs, which develops Tabular Foundation Models for structured business data (TabPFN-2.6, over 3 million downloads).
Cohere and Aleph Alpha
Cohere has acquired Aleph Alpha, with a combined valuation of around 20 billion US dollars, dual headquarters in Toronto and Heidelberg, and the Schwarz Group as lead investor.
There is also state money beyond the challenge: the federal government is providing around 805 million euros for a European AI gigafactory, and the research ministry funds AI at universities with additional means. innobu describes the wider regulatory frame in its article on EU tech sovereignty in chips, cloud, and AI , and the role of homegrown models for industry is explored in the piece on sovereign AI as Germany's industrial future .
Reality check: is 125 million euros enough?
On its own, 125 million euros is not enough to match US and Chinese labs, and that is not the goal. Costard himself calls the sum only the first step. The real leverage is private follow-on capital and a deliberately different research approach.
The numbers show the imbalance. According to the Centre for Future Generations, the cost of training top models rises around 3.5 times per year, while Europe's compute infrastructure and capital base lag well behind the US and China. This tension between record investment and structural gap is exactly what innobu describes in its article on the German AI startup paradox .
SPRIND's bet is not the copy: instead of copying today's models, the agency relies on new methods and European strengths such as industrial data, manufacturing know-how, and privacy-friendly AI. Whether that is enough will only be decided once private capital takes over from the public seed funding.
Challenges and risks
The program addresses real gaps but also carries clear risks. A balanced view has to name both, otherwise funding policy quickly becomes symbolic politics.
Talent drain
European teams often scale in the US because capital and compute access are easier to find there. Without competitive salaries and reliable compute contracts, the risk remains that the best minds move on once the funding ends.
Security and espionage
Frontier research is an attractive target for espionage and theft and at the same time a geopolitical lever. The labs need strong protection from the start, otherwise the funded knowledge leaks away unintentionally.
Governance and pace
Consensus-based jury decisions are transparent but can be slow. A program competing with labs that ship on a weekly cadence has to keep its own decision paths lean.
The funding gap after 24 months
The biggest uncertainty sits at the end. Without reliable private follow-on capital, the effort risks breaking off exactly when the labs would need to scale. The 125 million euros carry the start, not the peak.
What companies should do now
For most European companies, the challenge is less about applying than about watching and connecting. Those with structured data, domain knowledge, or compute capacity can explore partnerships early. Four steps help.
-
Track the finalists
Watch which teams advance at the end of June and check whether their specialisation fits your use case. Early proximity to a rising lab can later secure access and terms.
-
Assess your own data
Industrial and manufacturing data are a European advantage that the large US models do not have. Check whether your data holdings could serve as a contribution or a differentiator in a model partnership.
-
Clarify sovereignty requirements
Define where homegrown or European models make sense for data protection or compliance reasons and where established US models suffice. This judgement prevents costly mistakes in both directions.
-
Plan for model diversity
In your own AI architecture, do not bet on a single lab but keep models switchable. That way you benefit from new European providers without becoming dependent on any one of them.
The Next Frontier AI Challenge does not decide Europe's AI future alone, but it is a serious start. For companies, what matters less is whether a German lab wins than their own ability to build model diversity and sovereignty deliberately into their strategy.
Further reading
Frequently asked questions
The Next Frontier AI Challenge is a competition run by Germany's federal innovation agency SPRIND, launched on 25 May 2026. With 125 million euros in non-dilutive funding, it aims to build up to three European frontier AI labs that develop their own foundation models. Selection runs across three stages over 24 months, from July 2026 to autumn 2028.
The process has three stages. In stage 1, up to 10 teams receive 3 million euros each over 7 months. In stage 2, up to 6 teams advance with 8 million euros each over 8 months. In stage 3, up to 3 teams receive 15.5 million euros each over 9 months. After each stage, a jury decides on advancement or elimination.
Not on its own, and that is not the goal. SPRIND's head of challenges, Jano Costard, explicitly calls the sum only the first step. Each finalist lab is then expected to raise 250 million to 1 billion euros privately. Europe today accounts for under 10 percent of global AI venture capital, so the leverage lies in private follow-on capital and new research approaches.
Non-dilutive funding means SPRIND takes no equity in return for the money. Teams keep full control over their research direction and intellectual property. The aim is to free research from the short-term pressure to ship a sellable product and to create room for new methods.
For most companies, the challenge is less about applying than about watching and connecting. Those with structured data, domain knowledge, or compute capacity can explore partnerships early. It makes sense to track the three finalists at the end of June, clarify sovereignty requirements, and plan for model diversity and switchability in your own AI architecture rather than betting on a single lab.