Nvidia Opened Its Server Racks to Rival Chips, and That Is the Real Power Move

By Omar aka Crypt0 · Sep 14, 2026

The most important AI chip story this week is about a company agreeing to stop fighting over chips. Startup d-Matrix said Thursday it will adopt Nvidia's chip linking technology so its own processors can slot directly into Nvidia's data center systems. The company's new Raptor chips will plug into Nvidia server racks using NVLink Fusion, a connector and memory standard that lets custom AI chips live inside Nvidia's larger systems. The racks arrive in 2027, with Raptor's final design stage wrapping by the end of this year. The combined systems target fast, low latency AI services like coding assistants, chatbots, and voice agents, the exact workloads where speed decides winners. Santa Clara based d-Matrix has shipped one AI chip already, in November 2024, and Microsoft has backed the company since its 110 million dollar round in 2023. Last year's 450 million dollar raise valued it at 2 billion dollars.

Here is the part the market should think about. AI workloads are shifting from training models to running them for everyday use, the inference era, and Nvidia's pricey GPUs dominate training while specialists like d-Matrix chase inference. Nvidia could have kept its racks closed and fought every inference startup for the business. Instead it is selling them the socket. Every Raptor that plugs into an NVLink Fusion rack is one more system where Nvidia owns the interconnect, the memory architecture, and the software layer, even when it sells zero GPUs. That is a platform play wearing a partnership costume, and it is a strong one.

This fits a pattern of Nvidia acting less like a chip vendor and more like the central bank of AI compute. The company disclosed 36 billion dollars of commitments under its AI Compute Partnership, a program where Nvidia guarantees to rent GPU capacity itself when cloud providers come up short, taking 50 percent of revenue above a base rate. Anthropic signed a 35 billion dollar cloud deal with Nvidia backed Lambda for a data center that Hut 8 is developing in Texas, with Nvidia holding the lease on the facility. Microsoft and Google both walked away from a similar Nscale deal before Anthropic took it, per Semafor. Nvidia has moved far beyond selling picks and shovels. It is financing the mines, insuring the miners, and now wiring the competitor's tools into its own racks.

For anyone watching the AI hardware race, the lesson is that the interconnect is the moat. Chips get faster every eighteen months. The system that ties them together, bills for them, and guarantees the revenue behind them is where the lasting power sits. Investors, builders, and policymakers should all be studying NVLink Fusion, because that connector quietly decides who wins the inference decade.

Sources

New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.