SPRIND has launched applications for a €125M competition.

      The Next Frontier AI Challenge, introduced at EurIPS in December, advises candidates against attempting to catch up with OpenAI. Instead, it encourages them to leapfrog to the next architectural S-curve, offering up to €1 billion in additional funding for the three winning labs.

      SPRIND, Germany’s federal agency for breakthrough innovation, has begun accepting applications for the Next Frontier AI Challenge, a structured competition lasting 24 months with a budget of €125 million aimed at establishing up to three European frontier AI labs from the ground up.

      The application period is open until June 1, 2026, with jury presentations planned for June 24-25, and the first ten funded teams slated to begin work in July. The challenge was first unveiled at EurIPS in Copenhagen on December 3, 2025, at a European conference recognized by NeurIPS, the leading AI research conference worldwide. Its message clearly highlights Europe’s current standing:

      “Europe’s competitiveness in AI innovation is significantly lagging behind the USA and China,” the challenge brief notes. “If Europe doesn’t develop its own models, it risks increasing its strategic reliance on these technologies.”

      The aim is not just to close the existing gap but to completely bypass it by focusing on what SPRIND classifies as the next S-curve—the next significant architectural and paradigmatic advance beyond the current transformer-based systems.

      How does the competition operate? The €125 million is allocated across three stages with progressive selection narrowing down applicants. In Stage 1, up to ten teams may receive up to €3 million over seven months, expected to produce initial technological proofs of concept for their frontier hypothesis, such as a technical report, preprint, experimental artifacts, or indications of a prospective new scaling dimension or emergent phenomena.

      Up to six teams will advance to Stage 2, where they can receive up to €8 million each over eight months; at this stage, they must focus on production-ready engineering methods, validated scaling dimensions, and identifying what SPRIND terms ‘technical secrets’—proprietary insights that bring notable performance benefits.

      Finally, up to three winners will progress to Stage 3, obtaining up to €15.5 million each over the next nine months, with the objective of creating a functioning frontier system prototype, launching user-facing applications in testing, and preparing an investment-grade data room for subsequent capital raising.

      The maximum non-dilutive funding available for each team throughout all three stages is €26.5 million, a significant initial investment for a serious AI lab but still a small fraction compared to the amounts raised by frontier labs like Anthropic or Mistral.

      The real incentive is what comes next: SPRIND has explicitly designed the program with a view towards achieving a €1 billion scale-up round for each winning lab at the end of the competition, equating that funding to a US mega Series A that transitions a ‘serious seed lab’ into a ‘true frontier player.’

      The €1 billion target is separate from the challenge budget and would necessitate the labs to secure it from outside investors; the Financing Workstream with SPRIND aims to assist teams in crafting investment-ready data rooms to enhance their fundraising efforts.

      The challenge adopts a technology-agnostic stance regarding submission criteria while also clearly defining what it is not seeking.

      SPRIND identifies disqualifying categories that provide guidance: incremental transformer optimizations that do not introduce fundamentally new capabilities; reproductions or variations of existing models such as recreating OpenAI, Llama, or Qwen; minor efficiency improvements involving quantization or leaner MoE routing; conventional agent architectures that lack systemic innovation; domain-specific fine-tuning without foundational breakthroughs; and relying on sheer scaling as the primary innovation thesis.

      What the challenge does seek is harder to precisely define, which SPRIND openly acknowledges. Suggested directions include alternative model architectures (such as state-space models, energy-based transformers, diffusion LLMs, JEPA-style objectives, Titans architectures, or entirely novel frameworks); agentic systems that feature fundamentally new orchestration theories instead of traditional tool-use wrappers; embodied AI and world models; neuro-symbolic and hybrid approaches; scientific foundation models tailored for protein design, material science, or drug discovery; and innovative training paradigms that replace the conventional pre-train plus reinforcement learning from human feedback (RLHF) model.

      The S-curve framing is a strategic decision. SPRIND argues in the challenge materials and through commentary from its leadership that if European teams attempt to replicate the existing generation of frontier labs under European budgets and limitations, they are likely to lose out in terms of both cost and speed.

      The contemporary S-curve is largely dominated by extensive transformer and diffusion structures, and the financial resources necessary to compete on that curve at scale exceed the capacities of European public funding.

      However, the shift to a new architectural framework, regardless of its ultimate form, signifies a moment where early entry and accumulated expertise may