Daisytuner raises €6 million and launches its inference engine Daisyflow
Darmstadt-based DeepTech startup Daisytuner has closed a €6 million Seed round for its technology that runs AI models at full performance on any hardware. The team took part in the IGNITION incubator of hessian.AI | AI Startup Rising and is now part of the Accelerator RISING.
The challenge: compute is locked into single vendors
Compute is one of the scarcest resources in AI, and much of it is tied to a single vendor’s software stack. Running models on other hardware usually means extensive code changes. With every new model or chip, existing optimisations often stop paying off.
The approach: optimisations that add up
Daisytuner automates performance engineering for AI models. At its core is an open-source compiler that turns code from common AI frameworks into native code for a range of chips, including Nvidia and AMD GPUs as well as new accelerators. AI agents test optimisations on real hardware, and the compiler verifies that each one is correct. Verified optimisations go into a performance database, so they can be reused for similar code and each new optimisation comes faster.
Daisyflow is in production
Alongside the funding, Daisytuner has launched Daisyflow, an inference engine that runs inside the customer’s own infrastructure. It serves language, vision, speech and recommendation models via standard APIs. According to the company, Qwen 2.5 runs up to 18% faster with Daisyflow on AMD Instinct GPUs than with the widely used inference framework vLLM
Team and progress
Lukas Trümper, formerly a researcher at ETH Zurich, founded Daisytuner in 2024 together with Adrian Schmitz. The team now has 12 people with backgrounds at ETH Zurich, TU Munich, RWTH Aachen, TU Darmstadt and Lawrence Livermore National Laboratory, and aims to grow to well over 20 by early 2027. Its core patent was granted a few weeks ago. Over the past year, Daisytuner has worked with the chipmakers Tenstorrent and Q.ANT.
From IGNITION to Slush with AI Startup Rising
Daisytuner went through the IGNITION incubator of hessian.AI | AI Startup Rising and is now part of the Accelerator RISING. Together with the RISING batch, the team is at Slush in Helsinki, one of Europe’s leading startup and tech events.
Congratulations to Lukas Trümper, Adrian Schmitz and the whole Daisytuner team!
More about daisytuner: https://daisytuner.com/news/post-7