Daisytuner Secures €6M Seed to Break NVIDIA's Software Lock on AI Compute

As the AI industry struggles with severe compute shortages and vendor lock-in, German deep-tech startup Daisytuner has raised €6 million to let developers run models at peak performance across any chip architecture without rewriting code.
The explosion of generative AI has exposed a critical bottleneck: the industry's reliance on a single vendor's software stack. While alternative silicon from AMD and emerging chipmakers offers promising raw power, adapting complex AI models to run efficiently outside of NVIDIA's CUDA ecosystem often requires massive manual engineering effort. Darmstadt-based Daisytuner aims to dismantle this barrier, securing €6 million in Seed funding co-led by Elaia and LEA Ventures to automate AI model performance engineering across diverse hardware.
Alongside the funding, the startup launched Daisyflow, a high-performance inference engine built on its open-source DOCC compiler. The platform translates standard frameworks like PyTorch and C/C++ into a universal intermediate representation, automatically generating optimized native code for NVIDIA, AMD, and novel architectures. Instead of relying on human engineers to manually tune kernels for each new chip, Daisytuner deploys AI agents to search for optimizations, benchmark them on real hardware, and store proven recipes in a performance database.
This approach fundamentally shifts the economics of AI deployment by decoupling software from underlying silicon. Early benchmarks show models like Qwen 2.5 running up to 18% faster on AMD Instinct GPUs compared to existing serving engines like vLLM. By allowing enterprises to transition workloads to the most cost-effective or available hardware, Daisytuner provides a critical abstraction layer that could accelerate the adoption of alternative AI accelerators and loosen the grip of legacy software monopolies.
INTELLIGENCE BRIEF
WHY IT MATTERS
Daisytuner addresses the "CUDA moat," the primary reason alternative AI chips struggle to gain market share despite competitive hardware specs. By automating the translation and optimization of AI models for any silicon, the startup enables true hardware portability for enterprise AI workloads.
WHO IS INVOLVED
MARKET IMPACT
As data centers face power constraints and chip shortages, hardware-agnostic inference engines will become essential infrastructure, allowing companies to dynamically route AI workloads to the most efficient available processors.
This story was drafted with AI assistance and reviewed by TurkSpark editors before publication. Facts, figures, and names may be inaccurate — verify important details independently.


