Nvidia Opens Racks to Rival Chipmaker, Hyperscalers Eye Shift in AI Compute

Nvidia has opened its NVLink Fusion infrastructure to d-Matrix, a Microsoft-backed inference chip startup valued at approximately $2 billion. This move allows d-Matrix to wire its Raptor processors into Nvidia-powered systems using the high-speed interconnect.

This development is significant because it gives hyperscalers a credible on-ramp to workloads where Nvidia's GPU margins are fattest, specifically in inference. If Raptor and similar parts prove more efficient per token, hyperscalers could shift workloads off Nvidia GPUs while keeping the fabric, resulting in Nvidia collecting a smaller toll.

Nvidia's Chief Executive Jensen Huang framed this logic as 'AI has reached its inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue.' The company now pegs data-center revenue per gigawatt at roughly $18 billion for Hopper, $25 billion for Blackwell, and $40 billion for Vera Rubin.