Skip to content
TrustList
Funding news

Daisytuner raises €6m seed as it launches the Daisyflow inference engine

Editorial

By TrustList Editorial

Elaia and LEA Ventures co-led the Darmstadt compiler company's seed round, with Amadeus APEX Technology Fund and Seedcamp. Daisyflow runs open-weight models on a customer's own hardware.

About Daisytuner raises €6m seed as it launches the Daisyflow inference engine

Daisytuner raises €6m seed as it launches the Daisyflow inference engine

16 November 2026: Daisytuner, a Darmstadt company that automates the tuning of code for different chips, has raised €6 million in seed funding co-led by Elaia and LEA Ventures, with Amadeus APEX Technology Fund and Seedcamp taking part. It announced the round alongside Daisyflow, an engine for running AI models inside a company's own infrastructure.

Not yet independently verified. The round and the benchmark come from Daisytuner's own post of 16 November; we have not seen a post from either lead investor, and the speed figure against vLLM is the company's own test. We will update this when it can be confirmed, and remove this note.

Tuning code without the specialists

The company's argument is that every new model or chip undoes yesterday's optimisations, and that the work of getting fast code onto new hardware has been done by hand, mostly inside frontier labs and only for the newest chips. Its open-source compiler, docc, turns PyTorch, NumPy and C or C++ code into a dataflow representation and generates native code for CUDA, ROCm and newer architectures. AI agents then search for optimisations continuously and benchmark them on real hardware. Daisytuner says docc proves each optimisation correct before it is used, and stores proven ones in a performance database so they can be carried over to similar code.

Lukas Truemper, the founder and chief executive, started the company from his research on AI-driven compilers at ETH Zurich; Adrian Schmitz, a co-founder, works on its GPU optimisation. The team comes from ETH Zurich, TU Munich, RWTH Aachen, TU Darmstadt and Lawrence Livermore National Laboratory.

What Daisyflow adds

Daisyflow is built on open frameworks and serves language, vision, speech and recommendation models, with batching, routing and APIs compatible with OpenAI's, all inside the customer's infrastructure. Daisytuner's first published result is that Qwen 2.5 runs up to 18% faster than vLLM, the common open-source serving engine, on AMD Instinct GPUs.

There are three ways to buy it. The Daisytuner Platform is hosted and ready to use; Daisytuner Enterprise is the same platform self-hosted with a service level agreement; and Daisyflow adds the full inference stack on top. The company names cloud providers with mixed hardware fleets, enterprises that want control of their data, and chip makers that want their hardware in production quickly as the buyers it has in mind.

Where the money goes

Daisytuner says it will use the funding to bring Daisyflow to more enterprises, grow its team and extend the compiler. A month earlier it became an official sponsor of HiPEAC 2027, the European computing architecture conference where it had previously won an award. Our machine learning software category lists nine products, most of them for building models rather than serving them.

Company profile on TrustList: Daisytuner

Sources

Categories & features

TrustList Weekly

The week in software and IT, in one email

The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.

We will email you to confirm. Unsubscribe with one click in any issue. Privacy policy

You may be interested in