Skip to content
AI.info

The Pulse

Daisytuner raises €6 million to expand cross-chip AI software

Daisytuner has raised €6 million in a seed round co-led by Elaia and LEA Ventures. The company plans to expand its AI inference platform, compiler technology and team.

Daisytuner raises €6 million to expand cross-chip AI software

Daisytuner has raised €6 million in seed funding to expand software that tunes AI models for different chips, Tech.eu reported on November 16. The German startup’s Daisyflow platform runs models during inference, the process of using a trained model rather than training one.

Elaia and LEA Ventures co-led the round, with participation from Amadeus APEX Technology Fund and Seedcamp. The company plans to use the money to bring Daisyflow to more enterprises, develop its compiler technology and grow its team.

Funding and hiring

Most of the funding is intended for hiring, with the remainder to be spent on chips from different vendors for testing, benchmarking and training the AI agents that improve the compiler, Tech Funding News reported. The company has 12 people and is targeting well over 20 by early 2027, according to the outlet.

The round backs a company based in Darmstadt, Germany. Its platform is in production and can run inside customers’ own infrastructure, rather than requiring them to rely on an external AI cloud provider.

From code to chips

The software combines an AI-powered compiler with Daisyflow, which serves models during inference. Companies can deploy the platform in their own data centres.

The open-source compiler translates code from frameworks such as PyTorch into native code for different chips, including Nvidia and AMD GPUs, according to an announcement published by investor Amadeus Capital Partners. AI agents test optimizations on hardware, and the compiler checks them before use. Verified optimizations are stored in a performance database and reused on similar code.

Daisyflow is built on open frameworks and serves language, vision, speech and recommendation models using APIs already used by teams, the announcement said. The company aims to reduce the manual work involved in tuning models for particular chips.

Benchmark and customers

In a company benchmark, Daisytuner claimed Daisyflow ran Alibaba’s Qwen 2.5 model up to 18% faster than vLLM, an open-source inference engine. The comparison used AMD Instinct GPUs.

Lukas Trümper and Adrian Schmitz founded Daisytuner in 2024. Trümper, the company’s CEO, told Tech Funding News that Daisytuner had spent the past year working with chipmakers Tenstorrent and Q.ANT. The company was very close to finalising its first enterprise deals, the outlet reported.

The team includes researchers from institutions such as ETH Zurich, TU Munich, RWTH Aachen, TU Darmstadt and Lawrence Livermore National Laboratory. Daisytuner says its technology builds on more than a decade of research.

Sources

  1. Tech.eu
  2. Tech.eu
  3. Tech Funding News

Related stories

Explore

More articles