ETH Zurich launches AI Foundry, a programme for frontier AI spin-offs

October 2, 2026, 08:30Source: ETH AI Center

ETH Zurich has launched AI Foundry, its first programme dedicated to AI spin-offs. Run by the ETH AI Center, it gives founders access to the Alps supercomputer, support from the engineers behind the Apertus language model and guidance on protecting intellectual property.

Key points

ETH Zurich has launched AI Foundry, its first programme dedicated to AI spin-offs, run by the ETH AI Center.
The six- to twelve-month programme offers access to the Alps supercomputer (more than 10,000 GH200 GPUs) and Euler, with an initial 10,000 GPU hours per team.
Teams work with the engineers who pre-trained the open Swiss LLM Apertus and get patenting guidance, aiming for at least one patent filing per team.
Terms: typically 2% equity plus a 2% fee on capital raised, declining to zero over four years.
Applications opened on 2 October 2026 and are accepted on a rolling basis.

What AI Foundry offers

AI Foundry is a six- to twelve-month programme for frontier and deep-tech AI ventures. According to ETH Zurich, it targets ambitious founders who want to build the next frontier labs, work on novel architectures and large-scale AI, or pursue IP-heavy fields such as physical AI, AI for science and industry-specific solutions.

The programme rests on three pillars:

Compute: access to the CSCS Alps system, which has more than 10,000 GH200 GPUs, and to ETH Zurich's Euler cluster.
Pre-training know-how: hands-on support from the engineers who pre-trained Apertus, the fully open Swiss large language model.
IP protection: guidance on patenting, with the goal of at least one patent filing per team.

ETH says it is one of the very few programmes anywhere, and the first in Europe, to put a frontier-model pre-training team at the disposal of start-ups.

Compute and engineering expertise

The programme is built around two problems the university says AI start-ups face: training and running modern models takes large amounts of compute, and turning a research result into a reliable product takes engineering skill beyond a working prototype. General start-up programmes can rarely provide either, according to the announcement.

Teams use capacity at the Swiss National Supercomputing Centre (CSCS) as they grow, instead of buying or reserving hardware in advance. Each team receives an initial 10,000 GPU hours, and the programme helps secure additional compute grants for projects that need more. Further benefits, including partnerships for inference compute, will be announced in the coming months.

The Apertus engineers help with AI systems design, scaling training and inference, and building infrastructure to move from research prototype to robust product. Teams can also use open-source assets of the Swiss National AI Institute (SNAI).

Melanie Gabriel, Co-Director and COO of the ETH AI Center, said: "Too many of Europe's best AI founders conclude that serious compute and frontier-grade engineering are only available in Silicon Valley." She added that the decision about where to build a company is made early, and that the aim is for the answer to be Zurich.

IP, venture support and terms

For deep-tech start-ups, ETH notes that a defensible IP position is often what convinces investors. Teams receive guidance on combining protection with openness, for example by patenting core methods while open-sourcing models or tools to build a community.

Each team gets a support plan matched to its stage and needs, delivered by ETH AI Center staff and experienced entrepreneurs from ETH Zurich's Entrepreneur in Residence programme. Depending on the project, this can include coaching, venture development, access to investors and fundraising support. Teams also join the ETH AI Center community, including events such as the AI+X Summit.

On terms, ETH Zurich applies its standard spin-off terms, typically 2% equity. In addition, the programme receives a 2% fee on capital raised by its companies, declining to zero over four years. Executive Director Alex Ilic called this the "2+2 model" and said it aligns ETH and founders around raising big initial rounds. He said what flows back funds the next generation of AI founders.

Who can apply and where it fits

The programme is aimed at AI start-up teams from ETH with a connection to the faculty, research, or activities of the ETH AI Center. Applications open on 2 October 2026 and are accepted on a rolling basis.

AI Foundry joins ETH Zurich's other entrepreneurship programmes. The Pioneer Fellowship supports researchers who want to turn ETH research results into a spin-off, and UPortunity, launched with UBS, supports recent graduates. According to ETH, AI Foundry adds dedicated support for ventures in artificial intelligence.

For ETH-linked founders building AI companies, the programme combines compute, model-training expertise and IP guidance in one offer, in exchange for the equity and fee terms described above. Further benefits, including inference compute partners, are still to be announced.