Details

Next Frontier AI is run by SPRIND – the German Federal Agency for Breakthrough Innovation. Think of SPRIND as a state-backed deep-tech fund and company builder: we deploy public money with a venture mindset, take real technical risk and are explicitly mandated to try things that can fail.

For Next Frontier AI, SPRIND coordinates a wider European coalition: industry partners, compute providers, investors and a heavyweight jury of AI builders. We kick-off Next Frontier AI by running a SPRIND Challenge – a competitive DARPA-style, outcome-driven program – but scaled up in budget, ambition and international scope. With short proposals, no bullshit bureaucracy and funding within a few weeks, SPRIND Challenges are engineered to boost top innovators, not slow them down.

You'll find more about SPRIND at sprind.org/en – but the short version is: we are here to back weirdly ambitious things that normal instruments can't touch.

By ~2032 at the latest, create three European frontier labs that can stand at eye level with today's global leaders – and do it by leapfrogging to the next S-curve, not just chasing the current one.

The current S-curve of AI is dominated by massive transformer and diffusion stacks:

• insane data/compute/energy appetite,
• scaling laws that favour whoever can burn the most H100s,
• and architectures and value systems optimised for a US/China context.

If you try to "catch up" on that curve with European budgets and constraints, you lose on cost and speed by construction.

So Next Frontier AI explicitly bets on the next S-curve – call it "electric-drive AI" vs today's "combustion-engine AI":

• more sample- and energy-efficient models and training regimes,
• architectures that play to Europe's strengths, e.g. in industrial data
• systems that are controllable, auditable and easier to align with European values from the start.

We do not hard-code a technical thesis. You can optimize and scale SOTA architectures, build agentic systems, work on neuro-symbolic stacks, embodied AI, world models, systems-level breakthroughs in data/infra – as long as you can argue a credible path to frontier-grade capability and a lab that is European at its core.

The product of the first 24 months is "labs with pilots" – seed-stage frontier lab seedlings with real teams, infra, artefacts and a prepared bridge to €1B per winner. The initiative then continues beyond the Challenge to help them grow into full-blown frontier labs.

€125M non-dilutive funding for 10 teams following three stages:

1. Proposal + pitch; the jury selects up to 10 teams.
   • Join our waitlist now: you will receive infos via e-mail on how to apply
   • Form a team and a a European legal entity
   • Apply until May 31st 2026
   • Pitch in June 2026
2. Funded build phase (July 2026 – Autumn 2028)
   • Those 10 teams get serious funding, compute and support, not stipend-level money.
   • You move through staged gates with down-selection until up to 3 winners remain.
   • At each gate, the jury looks at research agenda, metrics, infra, MLOps, evals, safety, early pilots.
   
   You have questions? Or you worry that this plan doesn't fit your frontier lab? Reach out and let us know! Contact us at nfai@sprind.org
3. Outcome of the Challenge
   Not a single "hero checkpoint", but 3 venture-ready labs with pilots, eval suites, infra, teams, and investment-grade data rooms.
4. Principles (as a participant you'll feel this):
   • Competition meets collaboration: you run in parallel with other labs, see what they're doing, sometimes share insights – and still fight to outrun them.
   • Freedom over bureaucracy: we cut paperwork where we can. Funding follows progress, not PowerPoint.
   • Results & reports: if your system learns faster, scales better, or moves real-world metrics, you advance.

Yes. We know your competitors won't wait – and neither should you. That's why we are launching an early-bird track ahead of the Challenge that starts on July 1st 2026: the "Frontier AI" Funke (New Frontier Concepts), which starts on February 1st.

Apply from December 15 until January 15 and, if selected by the jury, get funded by February 1st: up to €500k

Apply from January 15 until February 15 and, if selected by the jury, get funded by March 1st: up to €400k

You apply with a sharp frontier thesis and a compact concept sketch, over 2–3 months you turn that into a serious system design plus a mini-PoC with any meaningful "sign-of-life" on reduced scale.

Our funding covers people and compute so you can run real experiments, not just write PDFs.

The Funke is optional and decoupled from the main Challenge: joining it will neither be an advantage nor a disadvantage when the jury selects up to 10 teams for the 24-month Next Frontier AI Challenge starting July 2026.

What you do get is a head start – tighter hypotheses, first artefacts, and early exposure to how we work – so that by the time applications close in May 2026, you're already operating like a small lab instead of just having a slide deck.

Send an email to nfai@sprind.org to learn more.

The famous "up to €1bn" is not a random PR number and not part of the Challenge budget. It's the order-of-magnitude scale-up capital each winning lab should be able to raise after the 24-month competition – the equivalent of a US mega Series A that gets you from "serious seed lab" to "real frontier player".

Everything in Next Frontier AI is engineered backwards from that moment.

What the Next Frontier AI initiative actually does:

From day one, there is a dedicated Financing Workstream, not a last-minute "oh, now go fundraise" push:

• We work with you to build investment-grade data rooms – metrics, infra, security, safety, governance, compliance, IP – the stuff serious investors and corporates will grill you on.
• We design the labs to be investable – sane cap tables, governance, legal setup, risk management – so big tickets from VCs, corporates and public instruments are structurally possible.
• We rally a coalition of capital around the Challenge – European and global VCs, corporate venture arms, and infra providers, public banks and funds (EIB, national programs, etc.).
• The goal is that by 2028, these players don't hear about you for the first time – they've been following you through the Challenge.
• We run structured investor & partner processes – dedicated diligence days, 1:1s, joint workshops with potential anchor customers and compute providers – so that by the end of the Challenge, each winner has a concrete path to a ~€1bn round (term sheets / MOUs / structured processes), not just "good vibes".

What this means for you as a team

• The Challenge money (€125m total) pays for mid-scale training, infra and pilots and gets you to "frontier-lab seedling with receipts".
• The initiative then aims to convert that into a €1bn round per winner by making you legible, de-risked and highly visible to a pre-aligned investor & partner crowd.

We cannot promise "you will receive exactly €1,000,000,000 on date X". But the mission of Next Frontier AI is explicitly to close the missing link – company building plus serious follow-on scale-up financing. If by the end of the Challenge our winners don't have a very high-probability path to that kind of round, we haven't done our job.

We're optimising for teams, not lone geniuses or random consortia.

What kind of teams?

• Deep technical backbone: You should have people who have actually trained serious models, run infra, shipped systems, or done adjacent hard things (robotics, large-scale scientific computing, STEM, etc.).
• Builder mindset: This is for people who are willing to build a company or lab, not just publish a paper. Product sense, ops, and some tolerance for chaos are a plus.
• Diverse skills Ideal teams mix:
  • research (ML, optimisation, theory),
  • engineering (systems, MLOps, infra, security),
  • data (pipelines, curation, governance),
  • and from Stage 3 of the Challenge on at least one person who cares about business / users / customers / GTM

Already formed teams / companies?

Yes, absolutely. Intact teams and existing startups can apply as a unit – that often accelerates execution.

Don't have a team, yet?

• We will help you to find co-founders and partners
• Please visit our on-site roadshow in several European cities and our virtual meet-ups from February through April 2026

Team size & composition over time?

• At application: a committed core (usually 3–7 people) with a clear technical thesis and early roadmap.
• During the Challenge: scaling into a multi-disciplinary lab. Think: a few dozen people over time – researchers, engineers, MLOps, data, product, ops. Exact numbers are up to you; we care about functionality, not org charts.
• If in doubt: if you can credibly argue "we can build and run a frontier-grade system given enough budget and time", you're in the right ballpark.

• Exact rules will be in the formal call, but you should assume:
  • you'll need a European legal entity (UG, GmbH, SAS, Oy, S.r.l., etc.) by the time funding flows,
  • and the lab must be effectively headquartered in Europe.
• We'll provide templates and support on cap table, governance and contracts, so lack of prior startup experience is not a blocker.

Compute is part of the Challenge budget. The plan is: we try to negotiate bulk deals with (preferably European) HPC / cloud providers. If that's not possible, we will fund you directly.

We prioritise "buy European – where viable", but we are not dogmatic: if the best option for your setup is a hyperscaler, that can be okay – the goal is to make the labs win, not an ideology.

• Yes. That's a core design feature, not an afterthought.
• Next Frontier AI and partners provide company-building support:
  • templates for legal structures, equity, IP,
  • help with hiring, HR and finance basics,
  • governance, boards, non-executive advisors,
  • GTM coaching, safety & compliance playbooks,
  • support in recruiting MLOps, security, data, product, etc.
• We don't expect every PI or staff engineer to suddenly become a perfect CEO. We care that the lab becomes investable and operationally excellent; we'll help you get there.

In principle: yes, as long as it doesn't break state-aid rules, double-fund the same cost block, or create impossible governance.

You can have existing grants, VC money or corporate collaborations; we'll just need transparency so contracts don't conflict.

For follow-on, we explicitly want you to combine: venture capital, strategic corporates, public instruments (EIB, national programs), and other non-dilutive sources.

The exact rules and combinations will be spelled out in the official call and contracts.

The whole endeavor is built around iterative, milestone-driven progress:

• you'll run many experiments rather than a single moonshot,
• milestones tighten over time: agenda + teams → first scaled runs + evals → pilots → data room + financing,
• funding follows progress; we keep room for pivots as long as your thesis remains frontier-relevant and technically credible.

If you're used to "ship, measure, iterate" rather than "five-year grant and a PDF", you'll be at home.

After 6 months, we expect you to have validated your core technological hypotheses at small scale. The jury needs to find credible evidence that your system will be able to scale over the next phase of the challenge.

All information on this site is indicative and subject to change. The final conditions of participation – including eligibility, selection criteria, funding amounts, timelines and any financing mechanisms – will be defined in the official call for applications and the contracts between SPRIND and participating teams.

While Next Frontier AI is designed with a target of enabling successful teams to raise up to €1bn in follow-on funding, this cannot be guaranteed. SPRIND and its partners will use reasonable efforts to prepare and support such financing rounds but do not assume any obligation to provide or secure specific investment amounts.

SPRIND reserves the right to modify, postpone or cancel the Challenge or parts of the initiative if funding, regulatory approvals or other material conditions cannot be secured. In such cases, already executed contracts and applicable law will determine the handling of any outstanding obligations.