Daisytuner raises €6 million and launches Daisyflow, an AI inference engine for any chip

German startup Daisytuner announced on Monday, November 16, 2026, a €6 million seed round and the launch of Daisyflow, its production-ready AI inference engine. The round was co-led by Elaia and LEA Ventures, with participation from the Amadeus APEX Technology Fund and Seedcamp.

Daisyflow is a high-performance inference stack that runs AI models inside the customer's own infrastructure — on-premises servers, data centers or private cloud — and adapts execution to whatever chip is underneath, from Nvidia and AMD GPUs to newer accelerators such as Intel and Tenstorrent. It serves language, vision, speech and recommendation models through standard APIs.

Open-source compiler, up to 18% faster than vLLM

At the center of the technology is DOCC, Daisytuner's open-source compiler, which turns code from frameworks such as PyTorch into native code for each chip, with no rewrites and no custom kernels. AI agents test optimizations on real hardware, the compiler verifies that each one is correct, and validated optimizations enter a performance database that is reused on similar workloads.

According to the company, early results show Qwen 2.5 running up to 18% faster than vLLM, the industry standard, on AMD Instinct GPUs. The figure comes from Daisytuner itself, from its production tests — it stands as a company statement, not an independent benchmark.

Founded in 2024 in Darmstadt by Lukas Trümper and Adrian Schmitz, Daisytuner brings together a team with more than a decade of research at labs such as ETH Zurich, TU Munich, RWTH Aachen, TU Darmstadt and Lawrence Livermore. The money will go to growing the team — 12 people today, with a target of more than 20 in early 2027 — buying chips from multiple vendors for testing, and bringing Daisyflow to more enterprises and European hyperscalers.

The announcement comes months after Qualcomm closed its acquisition of Modular — a company in the same space, building software to run models across different chips — for $3.9 billion in June 2026. Daisytuner names Modular as its biggest competitor and bets that the AI bottleneck is in the software layer, not the hardware.

Image: Daisytuner logo. Credit: Daisytuner.

Sources

Transparency note: this article was produced with the help of artificial intelligence, based on public posts and sources on the internet, and reviewed editorially. Check the original sources to confirm the information.

By Muse