SPRIND, Germany’s federal agency for breakthrough innovation, opened applications on April 30 for a €125 million competition to identify and fund up to ten teams building frontier AI labs from scratch, according to The Next Web. Applications close June 1, 2026, with jury pitches scheduled for June 24-25 and the first funded teams starting work in July.
The competition, called The Next Frontier AI Challenge, was first announced at EurIPS in Copenhagen in December 2025. Its premise is blunt. “Europe’s competitiveness in AI innovation remains far behind that of the USA and China,” the challenge brief states. “Without training its own models, Europe risks deepening its strategic dependence on these technologies.”
Three Stages, Progressive Down-Selection
The €125 million is distributed across three stages. In Stage 1, up to ten teams each receive up to €3 million over seven months to produce technical proof points for a frontier hypothesis: a technical report, preprint, or evidence of a new scaling dimension. Up to six advance to Stage 2, receiving up to €8 million each over eight months, where the bar shifts to production-ready engineering and proprietary technical advantages. Up to three winners enter Stage 3, receiving up to €15.5 million each over nine months to build a working frontier system prototype with user-facing applications in testing and an investment-grade data room.
Maximum non-dilutive funding per team across all stages: €26.5 million.
What SPRIND Will Not Fund
The disqualification criteria are as revealing as the targets. SPRIND explicitly rejects: incremental transformer optimization, reproduction of established models (rebuilding OpenAI, Llama, or Qwen), incremental efficiency gains like better quantization or leaner mixture-of-experts routing, conventional agent architectures without systemic innovation, domain-specific fine-tuning without foundational breakthroughs, and brute-force scaling as the primary thesis.
What it does want: alternative model architectures (state-space models, energy-based transformers, diffusion LLMs, JEPA-style objectives, Titans architectures), agentic systems with fundamentally new orchestration theory, embodied AI and world models, neuro-symbolic approaches, scientific foundation models, and novel training paradigms that replace the pre-train plus RLHF stack.
The Billion-Dollar Endgame
The real prize is what comes after. SPRIND designed the program backwards from a target €1 billion scale-up round for each winning lab at the end of the 24-month competition. That capital comes from external investors — it sits outside the challenge budget entirely. SPRIND’s Financing Workstream is designed to help teams build the data rooms to make that raise credible.
The strategic logic: if European teams try to compete on the current transformer curve with European budgets, they lose on cost and speed by construction. The transition to a new architectural paradigm, whatever that turns out to be, represents a window where early expertise matters more than capital depth. SPRIND is betting European teams can claim that window before US labs dominate it.
Scale in Context
At €26.5 million per winning team, the funding is a meaningful seed for a serious AI lab but a fraction of what frontier players have raised. Anthropic is in talks for a round that would value it at $900 billion. Mistral raised hundreds of millions in a single round. The gap between SPRIND’s non-dilutive capital and the multi-billion-dollar war chests of US labs underscores the bet: this only works if the next paradigm genuinely resets the competitive field.