EN
Back to the archive

The archive · AI & Models · Strategic decision · 2025–2026

NobodyWho bets small local AI models beat cloud LLMs; €2M pre-seed from Nordic funds

Copenhagen startup NobodyWho raises €2M to run open-source small language models on laptops and phones, betting Europe wins AI with privacy, cost and energy.

NobodyWho

The betThat business AI needs are met by small open models on users' devices — winning on privacy, cost, energy — so Europe competes without playing the bigger-is-better game.Building

What the business is

NobodyWho builds an open-source engine that runs small language models directly on laptops and phones, letting developers ship on-device AI without managing cloud infrastructure.

Starting capital€2M pre-seed from PSV Tech, The Footprint Firm and Norrsken Evolve (Dec 2025)

How it started

Founded by Danish artist-technologist Cecilie Waagner Falkenstrøm — a two-time Lumen Prize winner whose AI artwork flew to the International Space Station — with CTO Asbjørn Olling, the team spent nearly a decade on local and edge AI before launching NobodyWho in 2025.

What happened

In December 2025 the startup raised €2M pre-seed from Nordic funds PSV Tech, The Footprint Firm and Norrsken Evolve. Its engine supports more than 10,000 open-source models under the EUPL 1.2 licence, claims up to 100x lower training and 500x lower inference energy footprints in early benchmarks, and had over 5,000 developers building on GitHub.

No ending yet — it is still running.

Background

NobodyWho is a Copenhagen open-source startup building an engine that runs small language models (SLMs) directly on laptops and phones, so data never leaves the user's device. Founder and CEO Cecilie Waagner Falkenstrøm is a Danish artist-technologist — a two-time Lumen Prize winner whose interactive AI work has been exhibited at the UN and the Victoria and Albert Museum, and who sent the first AI artwork to the International Space Station. Together with CTO Asbjørn Olling she spent nearly a decade on local and edge AI before launching NobodyWho in 2025.

The bet is contrarian: while OpenAI, Anthropic and Google race to scale ever-larger cloud models, NobodyWho argues most business AI needs are met by smaller models fine-tuned per domain. Its open-source engine supports more than 10,000 open-source models, integrates with developer frameworks, and lets app developers run a local model with a couple of lines of code. Early company benchmarks claim up to 100x lower training and 500x lower inference carbon footprints.

In December 2025 the startup raised €2M pre-seed from Nordic funds PSV Tech, The Footprint Firm and Norrsken Evolve, with more than 5,000 developers already building on the platform via GitHub. It runs an open-core business: core components stay open under the EUPL 1.2 licence, while NobodyWho charges for fine-tuning compute. The founders argue Europe cannot outspend the US or China on frontier models, so it should win on decentralised, privacy-first AI instead.

What has to be true

  • Cloud LLMs force users to send data to third-party servers and pay rising inference bills, which NobodyWho saw as an opening for on-device models.
  • The founders had a decade of edge-AI credibility, including an AI artwork on the ISS, which made an unproven local-inference thesis fundable.
  • Open-sourcing the core under EUPL 1.2 aimed to build an ecosystem of developers rather than a proprietary demo, giving the startup distribution without a sales team.
  • Nordic investors explicitly bet on climate: The Footprint Firm called on-device AI a transformative opportunity in a fast-growing emissions category.

What can be applied

When you cannot win the scale race, change the game: an open, on-device alternative reframes AI competition around privacy, cost and energy — whether fine-tuning revenue can pay for it stays open.

Aftermath

As of December 2025 NobodyWho was past MVP and scaling: over 5,000 developers used its engine on GitHub, Python support had just launched with more frameworks planned, and the pre-seed capital was earmarked for expanding platform support and production-grade adoption. No revenue figures or customer names were disclosed, and headline efficiency claims of up to 100x/500x reductions remained company benchmarks awaiting independent verification.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases