A German deep tech startup has secured €6 million in seed funding to crack one of artificial intelligence’s most stubborn bottlenecks: getting AI models to run at full speed on any chip, not just Nvidia’s.
Darmstadt-based Daisytuner announced the round alongside the launch of Daisyflow, a production inference engine that adapts AI workloads to whatever silicon sits underneath. Elaia and LEA Ventures co-led the investment, with Seedcamp and Abuiltus APEX Technology Fund also participating.
The startup’s pitch lands at an inflection point. Hyperscalers and AI companies have started shifting budgets away from raw capacity expansion toward wringing more efficiency out of existing infrastructure. Performance engineering, long a slow, manual craft practiced by a handful of frontier labs, now represents a competitive edge.
Daisytuner automates that work through a self-learning compiler called DOCC. It converts code from frameworks like PyTorch into native instructions for Nvidia GPUs, AMD accelerators, and newer architectures. AI agents continuously test optimizations on physical hardware, store verified results in a performance database, and reuse those wins on similar workloads.
“Compute is the scarcest resource in AI, and most of it currently sits behind a single vendor’s software stack,” said Sébastien Lefebvre, partner at Elaia. “Daisytuner lets the same code run at full performance on the silicon you choose, without a rewrite.”
Founded in 2024 by Lukas Trümper and Adrian Schmitz, the company draws on more than a decade of research from ETH Zurich, TU Munich, RWTH Aachen, TU Darmstadt, and Lawrence Livermore National Laboratory.
Meanwhile, Daisyflow already demonstrates tangible gains. The company reported Qwen 2.5 running up to 18% faster than vLLM on AMD Instinct GPUs. The platform offers APIs for language, vision, speech, and recommendation models and deploys inside a customer’s own infrastructure.
As OpenAI and Anthropic develop custom silicon, the race to extract performance from heterogeneous hardware intensifies. Daisytuner plans to expand its engineering team, scale Daisyflow’s enterprise reach, and broaden compiler support for emerging chip designs.














