Analysis
Nvidia plans to invest in d-Matrix, an AI inference-chip startup it competes with, as part of a deal that will also connect d-Matrix's next-generation Raptor chips to Nvidia's NVLink Fusion interconnect, The Information reported. Neither company has confirmed the investment's size or timing.
A Rival Nvidia Is Also Backing
d-Matrix, founded in 2019, raised a $275 million Series C in November 2025 at a $2 billion valuation. The startup builds inference-specific chips, a category distinct from the training-focused GPUs Nvidia dominates.
“## A Rival Nvidia Is Also Backing d-Matrix, founded in 2019, raised a $275 million Series C in November 2025 at a $2 billion valuation.”
d-Matrix has more recently been reported to be targeting a fresh round near a $5 billion valuation -- a potential markup of roughly 2.5x in under a year if it lands. It already counts Microsoft as a production backer, per CNBC's June reporting on d-Matrix starting chip production. Competitors in inference chips include Groq, Cerebras and SambaNova, all of which have raised at multibillion-dollar valuations on the same thesis: that running trained models cheaply at scale is a different, and increasingly separable, business from training them.
NVLink Fusion is the mechanism that makes this deal make sense for Nvidia despite the competitive overlap: it's Nvidia's interconnect standard for linking non-Nvidia chips into Nvidia-based AI infrastructure, and getting d-Matrix's Raptor XPUs compatible with it means Nvidia earns a toll on inference workloads even when customers pick a rival chip. It's the same playbook Nvidia has run with other AI bets this year, treating competitors as distribution rather than purely as threats.
What It Means
For chip buyers, Nvidia hedging against its own dominance by backing d-Matrix signals that inference, not training, is where the next capacity crunch is expected. For founders raising chip-adjacent rounds, a potential jump from $2 billion to $5 billion in under a year is a reminder of how compressed AI infrastructure financing timelines have become; d-Matrix's prior round closed barely eleven months ago. Pulse has tracked d-Matrix's acquisition activity as the company has built out its software stack alongside the hardware.
The risk nobody's pricing yet: an investment from Nvidia doesn't guarantee NVLink Fusion compatibility translates into design wins, and d-Matrix still has to prove its chips beat Nvidia's own Blackwell-generation parts in head-to-head inference benchmarks, not just interconnect compatibility.
Nvidia has not disclosed the investment size, and whether d-Matrix's next round actually prices at the reported $5 billion target -- rather than below it -- is the number that will tell the real story.
