Nvidia to Invest in Chip Rival d-Matrix as Its Challengers Choose Partnership

Nvidia plans to invest in d-Matrix, a seven-year-old developer of server chips for AI that aims to compete with Nvidia, according to three people with knowledge of the deal.

The planned investment is part of Nvidia’s effort to make its technology compatible with rival chip designers and bring them into its dominant hardware ecosystem, even as it has .

It’s also Nvidia’s first deal with a startup designing so-called AI inference chips since its $20 billion licensing agreement with Groq in December, showing that Nvidia isn’t content to bet on just one such firm.

Three months ago, Nvidia said it was working with d-Matrix so the two companies’ chips could work in tandem to handle different parts of the same AI inference workloads, with the goal of increasing the efficiency of such computing.

Rivals have found it difficult to eat into Nvidia’s substantial lead, so they’re increasingly seeking to work in conjunction with Nvidia graphics processing units rather than replace them entirely. Nvidia’s push to bring rivals into its ecosystem could benefit its networking hardware products, which string together different server racks so they can do heavy-duty AI computing tasks as one system, an approach known as parallel processing.

Nvidia in the past 12 months has taken multibillion-dollar stakes in more-established AI chip rivals, Intel and Marvell, in part to support integrations of Nvidia GPUs with those companies’ chips. It’s unclear when these integrations will happen at scale. Nvidia has similarly pitched several other chip startups on making their chips compatible with GPUs powered by Nvidia’s networking technology, according to people at those startups.

Amazon in August also raised the prospect of a similar integration involving Nvidia GPUs and Amazon Trainium AI chips, saying it would be able to use both chips in the same data center with similarly-structured server racks. It also said Trainium chips would tap Nvidia’s networking and memory hardware.

Another chip startup, SambaNova Systems, also demonstrated a way to connect its chips with Nvidia GPUs to jointly run AI models in June. Those integrations are happening as Nvidia separately develops a chip system that combines GPUs with specialized AI inference chips from Groq, whose founders and other employees Nvidia hired in a $20 billion technology licensing deal.

The Department of Justice is investigating whether Nvidia structured the Groq deal that way to avoid a formal review by the agency of the deal’s impact on the chip market, said people familiar with the probe, which was first reported by the New York Times.

Using chips made by different designers to handle the same AI workloads is a fairly new concept, sometimes referred to as disaggregated inference. Even as it pursues disaggregated inference with smaller rivals, Nvidia publicly says its own chips, on their own, are still the most efficient for handling nearly all AI workloads.

Still, cloud providers and major AI developers such as Anthropic and OpenAI don’t want to rely on just one chip vendor. And those two AI firms are separately developing their own chips for AI inference as well as using chips from other firms to lessen their reliance on Nvidia’s.

D-Matrix has focused on powering a process known as speculative decoding that speeds up an AI model’s performance. The forthcoming system with Nvidia will use d-Matrix chips to run a small AI model that guesses what a bigger model’s answer to the customer should be, and the bigger model will run on an Nvidia GPU to verify and accept the guesses.

The Information previously reported that d-Matrix was targeting new funding at a $5 billion valuation, though it’s unclear whether the capital from Nvidia would be part of that round. D-Matrix last raised $275 million at a $2 billion valuation in November 2025. D-Matrix started producing its first chips over the summer and is now testing its second generation chips.