Skip to content

Daisytuner Secures €6M To Speed Up AI On In-House Chips

Elaia and LEA Ventures have led a €6M seed round in Daisytuner, a German startup whose software makes AI models run faster on chips companies already own.

Daisytuner Secures €6M To Speed Up AI On In-House Chips
Image courtesy: Tech Funding News

German software startup Daisytuner has secured €6M in a seed round to help companies run AI models faster on chips they already own. Paris-based venture firm Elaia and Germany’s LEA Ventures led the round, with Amadeus APEX Technology Fund and London’s Seedcamp also investing.

Alongside the funding, the company is releasing Daisyflow, a system businesses install in their own data centres to run AI models, instead of renting capacity from an AI cloud provider. Daisytuner plans to use the money to win more enterprise customers, develop its compiler technology and grow its team.

Faster Answers From The Same Hardware

Daisyflow focuses on inference, the stage where a trained AI model is put to work answering questions or processing data. That is often where much of the day-to-day cost of AI sits, because a model runs every time someone uses it.

Behind Daisyflow is Daisytuner’s compiler, software that translates code into instructions a particular chip can carry out. The company uses AI to tune that process automatically for each type of hardware, rather than relying on engineers to hand-optimise code for every chip.

In the company’s own test, Daisyflow ran Alibaba’s Qwen 2.5 model up to 18% faster than vLLM, a widely used open-source system for running AI models. The test used AMD Instinct graphics processors. The figure has not been independently verified.

Daisytuner says its software already works with AMD and Nvidia graphics processors, accelerators from chipmaker Tenstorrent and a light-based processor from German startup Q.ANT. Companies can run its software on their own servers or use a hosted version.

From Research Labs To A Seed Round

Lukas Trümper and Adrian Schmitz founded Daisytuner, which has an office in Darmstadt, Germany. Its team includes researchers from ETH Zurich, TU Munich, RWTH Aachen, TU Darmstadt and Lawrence Livermore National Laboratory in the US.

The company started out with a tool that measures how code performs on real hardware. In 2025, it raised €1M in pre-seed funding led by LEA Partners, whose venture team has since spun out as LEA Ventures, with backing from angel investors including former footballer Mario Götze.

When the company launched a test version of its platform after that round, Trümper said popular AI coding assistants miss what happens when software runs on a physical chip. “Tools like GitHub Copilot and ChatGPT don’t understand what happens when code hits real hardware,” he said.

In December 2024, Daisytuner and ETH Zurich’s SPCL research lab shared a technology transfer award from HiPEAC, a European research network on computer systems and compilers.

A Crowded Field For AI Efficiency

Daisytuner is one of several young companies trying to get more out of AI hardware, each tackling a different part of the problem. US-based Clockwork, for example, builds software that keeps large AI training runs going when hardware fails.

The seed round takes Daisytuner from a small research-led team with a code-testing tool to a company selling an AI inference system to businesses. Its pitch rests on its compiler, which tunes AI models for each type of chip, so companies can run them on their own servers instead of paying an AI cloud provider.

The startup now has €6M from backers in France, Germany and the UK, and software that already supports chips from AMD, Nvidia, Tenstorrent and Q.ANT. Its speed figures come from its own testing so far, and it has yet to name the enterprise customers it plans to win with the new money.

Add Morning Tick on Google