Nvidia plans to take a stake in d-Matrix, a Santa Clara startup that designs chips for AI inference, The Information reported on October 8. The investment, whose size and valuation were not disclosed, is part of a broader Nvidia strategy to make its technology interoperable with rival silicon rather than fighting every inference workload alone.
d-Matrix has been building toward Nvidia's ecosystem for a while. On September 10, the company said it would use Nvidia's NVLink Fusion technology to connect its next-generation Raptor chips directly to Nvidia AI infrastructure. Under the plan, d-Matrix systems would use Nvidia's MGX rack design, Vera processors, and networking hardware, working alongside Nvidia GPU racks including the Vera Rubin NVL72. NVLink Fusion is Nvidia's mechanism for letting other companies' custom chips plug into its scale-up fabric, and MediaTek is building on it too. In August, Nvidia agreed to buy $3.5 billion of MediaTek convertible bonds.
Inference, the business of running a trained AI model to answer questions, is the segment where d-Matrix plays. Demand for it is rising faster than capital, time, and energy budgets can comfortably expand, d-Matrix co-founder and CEO Sid Sheth said in the announcement, which is why efficiency-focused accelerator cards matter to hyperscalers counting tokens per watt. The company's current lineup includes Corsair inference accelerator cards, JetStream network cards, and Aviator software.
Founded in 2019, d-Matrix closed a $275 million Series C round in November 2025 at a $2 billion valuation, according to DCD, and was looking to raise more money this July at a $5 billion valuation. An Nvidia investment would land squarely in that trajectory, following the pattern of strategic checks that buy Nvidia a seat at the table with partners whose chips will live inside its racks.
The broader play is worth watching. Nvidia's dominance in training is settled, but inference is fragmenting across startups, custom silicon, and sovereign-cloud buyers who want vendor choice. By investing in companies that build on NVLink Fusion, Nvidia keeps its networking and rack standards at the center of that fragmentation. For enterprise buyers, the practical takeaway is simpler: the inference market is maturing into an ecosystem where interoperability matters as much as raw throughput, and Nvidia is spending to keep its own fabric as the common language.