Embedl and the EIC Accelerator: Shrinking AI for Small Chips

Embedl, a 2018 spin-out of Chalmers University of Technology in Gothenburg led by CEO and co-founder Hans Salomonsson, sells software that automatically compresses and optimises deep learning models for specific embedded hardware. Its EIC application promised that work "currently taking months, years or abandoned R&D, can now be done in weeks", by pruning, compressing or searching for the best model for each chip while meeting energy and cost targets, according to CORDIS.

The European Innovation Council granted €2,499,999 towards a €3.64 million project from April 2023 to September 2025. The final report says Embedl extended its Model Optimization SDK with neural architecture search and advanced quantisation, raised supported hardware platforms "from four to ten", ran market pilots with "very reputable customers in the automotive and defense space", launched a SaaS platform called Embedl Hub with a paid tier planned for 2026, and filed six patents. It raised a €5.5 million round in June 2025. Embedl's revenue was SEK 26.5 million in 2024 and SEK 22.1 million in 2025, with net losses of SEK 6.6 million and SEK 14.2 million, according to accounts listed on hitta.se.

Where Embedl Stands Now

In July 2026 Embedl joined AMD's Open Robotics Ecosystem after optimising a vision-language-action model to reason in under 100 milliseconds on an AMD embedded chip, a figure AMD now uses on its product pages, according to Embedl. It was named to CB Insights' list of the 100 most innovative AI start-ups for 2025. Its staff grew from 19 to 25 in 2025 while revenue fell by about 17%, and total assets rose to SEK 68.5 million after the funding round.

The Technology: Making Big Models Fit

AI models for cars, drones and robots must run on small, power-limited chips. Optimising them by hand requires scarce experts and has to be redone whenever the model or hardware changes, the report says. Embedl automates this with pruning, quantisation, knowledge distillation and neural architecture search, and says it can cut energy use by up to 83% and make inference up to 18 times faster.

What the EIC Project Promised, and What CORDIS Shows

The project aimed to scale the business: better infrastructure, a stronger product, new products and global reach. The report documents product and platform advances, pilots, a new SaaS offering and follow-on funding. The commercial result is less clear: revenue rose in 2024 but fell in 2025, the year the project ended. CORDIS lists the project as closed.

1. Stronger SDKDeliveredNAS, quantisation, 10 platforms.
2. New productsDeliveredEmbedl Hub.
3. Follow-on fundingDelivered€5.5M, 2025.
4. Business scale-upMixedRevenue down in 2025.

The Money

The June 2025 round was co-led by Fairpoint Capital, SEB Greentech VC and Spintop Ventures, with Chalmers Ventures, STOAF, Almi Invest, board members and employees. The EIC award was a grant only. The report says the round "will secure the company's financial stability after the project concludes"; the 2025 net loss of SEK 14.2 million shows how much of that cushion is being used.

What Cannot Be Checked

Why Europe Should Care

Running AI on devices rather than in data centres can save energy, Embedl says, and its customers include automotive and defence firms. It is a European tool supplier in that field, now working with a US chipmaker's ecosystem. The EIC grant helped it build the product; turning it into a growing business is still under way.

The Verdict

Embedl and the EIC have not been asked for comment for this article. It is based entirely on public records and published statements, linked throughout.