SPRIND begins accepting applications for a €125 million competition.

      The Next Frontier AI Challenge, unveiled at EurIPS in December, advises applicants not to attempt catching up to OpenAI, but rather to leapfrog into the next architectural S-curve, with up to €1 billion in subsequent funding available for the three winning labs.

      Today, SPRIND, Germany’s federal agency for breakthrough innovation, has opened applications for its Next Frontier AI Challenge, a €125 million, structured competition over two years aimed at establishing up to three European frontier AI labs from the ground up.

      The application period will continue until June 1, 2026, with jury pitches planned for June 24-25 and the first ten funded teams set to commence in July.

      This challenge was introduced during EurIPS in Copenhagen on December 3, 2025, a European conference officially supported by NeurIPS, the foremost AI research conference worldwide. The challenge clearly highlights Europe’s current standing.

      According to the challenge brief, “Europe’s competitiveness in AI innovation lags significantly behind that of the USA and China.” It warns, “Without training its own models, Europe risks heightening its strategic dependency on these technologies.”

      The aim is not merely to reduce that gap on the current path but to completely bypass it by pursuing what SPRIND describes as the next S-curve: a significant architectural and paradigmatic advancement beyond the current generation of transformer-based systems.

      How does the competition function?

      The €125 million will be allocated across three different stages with a progressive down-selection process. In Stage 1, up to ten teams will receive a maximum of €3 million each over seven months, aiming to deliver initial technological proof points for their frontier hypothesis, including a technical report, a preprint, experimental artifacts, or evidence of a potential new scaling dimension or emerging phenomenon.

      From the initial teams, up to six will progress to Stage 2, with each receiving up to €8 million over eight months, where the criteria will shift to include production-ready engineering processes, validated scaling dimensions, and the initial identification of what SPRIND terms ‘technical secrets,’ proprietary knowledge leading to significant performance enhancements.

      The final three winners will then move to Stage 3, each receiving up to €15.5 million over nine months, with objectives that include developing a working frontier system prototype, user-facing applications in testing, and establishing an investment-grade data room for the next funding round.

      The total potential non-dilutive funding available for each team across all phases is €26.5 million, a substantial seed for a serious AI lab, although a fraction of what leading frontier labs like Anthropic or Mistral have garnered.

      The real incentive is seen in what follows: SPRIND is explicitly structuring the program with a goal of securing a €1 billion scale-up round for each winning lab at the end of the 24-month competition, aiming to convert that capital into the equivalent of a US mega Series A which would transition a ‘serious seed lab’ into a ‘real frontier player.’

      The €1 billion sum is separate from the challenge’s budget and would necessitate that the labs secure it from external investors; SPRIND’s Financing Workstream aims to assist teams in creating investment-grade data rooms to enhance the credibility of their fundraising efforts.

      The architectural gamble

      The challenge is intentionally neutral regarding technology in its submission requirements but clearly states what it does not seek.

      SPRIND’s disqualifying criteria serve as guidance: incremental transformer optimization without entirely new capabilities; recreations or derivatives of existent models like OpenAI, Llama, or Qwen; small efficiency improvements like enhanced quantization or streamlined MoE routing; traditional agent architectures lacking systemic innovation; domain-specific customization without foundational breakthroughs; and primary innovation based on brute-force scaling.

      What SPRIND is specifically searching for is harder to define, which the agency acknowledges. Potential directions include alternative model architectures (state-space models, energy-based transformers, diffusion LLMs, JEPA-style objectives, Titans architectures, or ‘entirely new frameworks’).

      Another focus is agentic systems incorporating fundamentally new orchestration theories as opposed to standard tool-use mechanisms; embodied AI and world models; neuro-symbolic and hybrid methods; scientific foundation models for protein design, materials science, or drug discovery; and novel training methodologies that replace the pre-training plus RLHF stack.

      The S-curve framing is a strategic choice. SPRIND argues that if European teams attempt to mimic the existing generation of frontier labs with European budgets and limits, they are likely to fall short in terms of cost and speed.

      The current S-curve is primarily characterized by extensive transformer and diffusion stacks, and the capital necessary to compete on that scale surpasses the funding capabilities of European public institutions.

      However, moving to a new architectural paradigm, whatever it may be, presents an opportunity where early engagement and accumulated expertise outweigh capital availability. SPRIND is betting that, with the proper support, European teams can establish a position on this curve before US labs dominate it.

      Founded in 2019, SPRIND