
Daisytuner raises €6 million and launches Daisyflow for private AI inference
The AMW Read
The seed round and launch add an inference optimization provider to a competitive infrastructure market, while broader performance remains unproven.
Daisytuner raises €6 million and launches Daisyflow for private AI inference
Daisytuner has raised €6 million in seed funding and launched Daisyflow, an inference engine for running AI models on customer infrastructure. Elaia and LEA Ventures co-led the round, with participation from Amadeus APEX Technology Fund and Seedcamp. The company plans to expand enterprise adoption, grow its team, and develop its compiler technology.
Daisyflow enters the AI infrastructure market with an approach to serving models across different hardware. It builds on docc, Daisytuner's open-source compiler, which translates supported code into a dataflow representation and generates code for accelerator platforms. The company says its optimization process tests candidate changes on hardware, verifies correctness, and stores successful configurations for reuse. Daisytuner reports that Qwen 2.5 ran up to 18% faster than vLLM on AMD Instinct GPUs in initial tests. That result does not establish performance across other models or hardware.
For builders evaluating private deployment, the key question is whether the measured benefit holds for their own models, accelerators, and workloads. Daisyflow supports deployment in customer data centers or private clouds and keeps requests, data, and outputs within customer infrastructure. Daisytuner also helps install, port, and tune the system. Investors should watch whether the company can extend its early AMD results across the model types and accelerator vendors it aims to support.