Nvidia pairs its GPUs with d-Matrix inference chips instead of competing with the startup

Nvidia is combining its GPUs with inference chips from startup d-Matrix in a joint system for running AI models. The first customer for the new product will be AI cloud company Parasail, The Information reports.

Why Nvidia chose partnership over competition

The AI chip market is split into two large segments. The first is model training, where Nvidia is the undisputed leader: its GPUs have become the de facto standard for large AI labs and cloud providers. The second is inference, meaning the direct handling of user requests to already trained models. The requirements are fundamentally different: for inference, low latency, energy efficiency and cost per token matter more than peak compute power.

d-Matrix specializes in inference chips. The startup develops dedicated hardware for fast, cheap production serving of language models. According to The Information, instead of applying competitive pressure, Nvidia chose integration: the companies will build a joint system that combines Nvidia GPUs with d-Matrix inference chips in a single product. This architecture covers both key scenarios: heavy computation on GPUs and economical, high-speed inference on specialized chips.

Who gets the joint system first?

The first customer for the new hardware will be Parasail, an AI cloud company that provides compute infrastructure for running large language models. Parasail focuses on enterprise clients that need scalability, low latency and predictable operating costs when running LLMs. Specific delivery dates and the commercial terms of the deal have not been disclosed.

Key facts:

• Source: The Information
• The system will combine Nvidia GPUs and inference chips from startup d-Matrix
• The first buyer is AI cloud company Parasail
• The product targets AI model inference, not training

Nvidia shifts strategy: integration instead of displacement

The partnership with d-Matrix reflects an important shift in AI infrastructure. Nvidia has traditionally built a closed, vertical ecosystem: GPUs, proprietary drivers, the CUDA platform and developer tools. But the inference market is fragmented, with Groq, Cerebras, Etched, SambaNova, Tenstorrent and other startups competing actively, each with its own incompatible architecture.

Integrating d-Matrix rather than acquiring or ignoring it may be a response to that fragmentation. It benefits Nvidia when niche players become part of its ecosystem without disrupting its core GPU business. For startups, it means a fundamentally new path to scale: reaching the broad market through a partnership with the leader rather than in head-on confrontation.

What this means

The inference market is growing faster than the training segment in 2025–2026 and is becoming one of the key battlegrounds in AI infrastructure. If the joint Nvidia and d-Matrix system succeeds, a strategy of technological integration instead of squeezing out niche competitors could become a template for other partnerships in AI hardware.