Finnect AI Insights

AI Business · · 5 min read

How Daisytuner Aims to Break AI's Hardware Lock-in

We'll explore how a new deep-tech startup plans to free AI models from specific hardware dependencies, making them faster and more adaptable.

BalticServers data center

On November 16, 2026, a Darmstadt-based DeepTech startup named Daisytuner announced it had secured €6 million in Seed funding.

This wasn't just another funding round; it was a bet on a future where artificial intelligence isn't held hostage by specific hardware.

Imagine building a powerful AI model, only to discover it performs best on one company's expensive chips, forcing you into a corner.

Daisytuner and its new inference engine, Daisyflow, aim to change that, promising to make AI models run faster on any chip, from NVIDIA to AMD and beyond.

That's the headline, but the real story here isn't just about a startup raising capital. It's about a fundamental challenge facing the entire AI industry: the increasing dependence on a narrow range of hardware and software stacks.

As Sébastien Lefebvre, Partner at co-leading investor Elaia, put it, "Compute is the scarcest resource in AI, and most of it currently sits behind a single vendor's software stack." Daisytuner believes it has a solution to this critical bottleneck. That raises three questions.

Question 1

What Exactly is AI's "Hardware Lock-in" Problem That Daisytuner is Addressing?

Today, when you develop an powerful AI model, especially a large, complex one, its performance is often tightly coupled to the underlying hardware. For context, NVIDIA's GPUs have become the de facto standard for AI training and inference due to their powerful architecture and a robust software ecosystem, like CUDA. This dominance means that many AI frameworks and tools are primarily optimised for NVIDIA chips. If you want to squeeze out every bit of performance, you often end up buying NVIDIA hardware and tailoring your software to it.

This creates a significant challenge for businesses. Imagine a scenario where a logistics company in India wants to use AI for route optimisation. They might have existing server infrastructure with AMD GPUs, or they might prefer to diversify their hardware suppliers to avoid dependency risks. However, if their chosen AI model is heavily reliant on NVIDIA's specific software stack, they face a dilemma.

The problem is clear: it creates a "single vendor's software stack" dependency, as investor Sébastien Lefebvre highlighted. It's like having a fleet of delivery trucks (your AI models) that run exceptionally well, but only if you use a specific brand of fuel and service them at that brand's exclusive service station.

Other fuel brands exist, other mechanics are available, but your trucks aren't built to leverage them efficiently. This limits your choices, potentially increases costs, and can slow innovation by tying you to one path. It also makes it harder to scale efficiently if you can't easily integrate new hardware from different vendors.

Question 2

How Does Daisytuner's Daisyflow and DOCC Technology Work to Solve This Challenge?

Daisytuner's core mission is to automate what's called "AI model performance engineering" across any chip. They've launched Daisyflow, their high-performance inference engine, which is designed to run AI models efficiently on any hardware, directly within a customer's own infrastructure. This means the AI model can adapt its execution dynamically to the specific hardware it finds itself on.

The secret sauce lies in two components: a self-learning compiler and a patented optimisation platform. Their open-source compiler, named DOCC (Daisytuner Open-Source Compiler), is crucial. It translates code from popular AI frameworks like PyTorch, NumPy, and even C/C++ into native code that can run directly on various target chips. This includes GPUs from NVIDIA and AMD, as well as newer, emerging architectures that are gaining traction in the market.

Think of it like this: your AI model speaks a language (PyTorch). Instead of needing a human translator (a developer rewriting code) for each specific regional dialect (NVIDIA chip, AMD chip, etc.), DOCC acts as a universal, highly efficient translator. It doesn't just translate; it optimises the translation for that specific dialect, ensuring the message (the AI model's computations) is delivered as fast and clearly as possible.

Moreover, AI agents continuously test these translations on real hardware. Every verified optimal result is stored in a performance database for reuse on similar code. This means the system learns and improves over time, getting smarter with each deployment.

The results are already promising: Daisytuner reported that the Qwen 2.5 model ran up to 18% faster on AMD Instinct GPUs using their tech compared to a standard solution like vLLM. As Lefebvre noted, this means "the same code run at full performance on the silicon you choose, without a rewrite." This is a significant step towards true hardware portability for AI.

Question 3

What Does Hardware-Agnostic AI Mean for Businesses Deploying Models in India?

For Indian businesses, embracing hardware-agnostic AI could be a significant step forward, offering both strategic and operational advantages. First, it offers genuine flexibility and reduces vendor lock-in. Instead of being compelled to invest solely in one brand of expensive GPUs, companies can explore a wider range of hardware options. This potentially allows them to leverage more cost-effective solutions from AMD or other emerging chip manufacturers, or even make better use of their existing, heterogeneous infrastructure. This is crucial in a market where budget optimisation and capital expenditure efficiency are often key considerations for business leaders across sectors like manufacturing, retail, and healthcare.

Second, it accelerates deployment and simplifies scaling. If your AI model can run efficiently on existing infrastructure, or on a mix of different chips without extensive code rewrites, you can get your solutions to market faster. Imagine a large retail chain needing to deploy AI for inventory management across hundreds of stores, each with varying local server capabilities. Daisytuner's approach allows for a "write once, run anywhere, fast" paradigm, simplifying operations and reducing the need for specialised engineering teams for each hardware type. This efficiency can translate directly into competitive advantage and faster return on AI investments.

Finally, it fosters innovation. When businesses are not constrained by specific hardware, they are freer to experiment with new AI models and deployment scenarios. This could lead to a surge in AI adoption in areas where cost or technical complexity were previously barriers. For example, government services or smaller logistics firms could leverage AI more readily if they can use off-the-shelf hardware and still achieve high performance. This kind of flexibility is essential for India's growing digital economy.

The Bigger Picture

The promise of hardware-agnostic AI isn't just about faster models; it's about democratising access to high-performance computing for artificial intelligence.

It suggests a future where the best AI models can flourish regardless of the chip they call home, allowing businesses to choose infrastructure based on need, cost, and availability, rather than being dictated by a single vendor.

This shift could unlock new levels of innovation and efficiency across every sector, from retail analytics to smart city solutions. The race to free AI from its hardware shackles has just gotten more interesting.

Quick Answers

What problem does Daisytuner solve in AI?

Daisytuner addresses the "hardware lock-in" problem in AI, where models are often optimised for specific chips, leading to dependency on a single vendor's software stack and limiting hardware choices for businesses.

How does Daisytuner's technology work?

Daisytuner uses its Daisyflow inference engine and an open-source compiler called DOCC. DOCC translates AI model code into native code for various chips (NVIDIA, AMD, etc.), while AI agents continuously optimise and store performance results.

What are the benefits of hardware-agnostic AI for businesses?

Hardware-agnostic AI offers businesses greater flexibility in choosing hardware, potentially reducing costs and accelerating deployment by eliminating the need for extensive code rewrites when moving between different chip architectures.

Sources

Facts are credited to the outlets above. The analysis, calculations and Signal Check are by the Finnect AI Insights desk.