d-Matrix's Raptor to ride NVLink Fusion into Nvidia racks by 2027

The arrangement Reuters described on September 10 is unusual because Nvidia is making it easier for a competitor's accelerator to sit inside its own rack design. Under NVLink Fusion, d-Matrix's next-generation Raptor processors connect directly into Nvidia data-center racks, with the systems expected in 2027. That splits the work three ways: d-Matrix supplies the accelerator, Nvidia supplies the rack-scale interconnect technology and architecture, and Astera Labs works on custom connectivity to move data between components.

For Nvidia, the bull case is that NVLink becomes more valuable even when it does not sell every accelerator, shifting the company from merely dominating GPUs toward controlling a system standard — the economics of owning the highway rather than building every vehicle on it. The bear case is that easier deployment of alternatives could cost Nvidia some GPU unit share if inference workloads become more heterogeneous. For Astera, more accelerators, memory pools, CPUs and switches create more connectivity bottlenecks, which is where its retimers and fabric switches come in, though its valuation already assumes substantial AI infrastructure growth.

Hedge-fund positioning leaned bullish into Q2, with tracked holdings in Astera rising to 73 funds from 53 and Nvidia to 285 from 275. Short interest cut the other way for Astera at roughly 6.6% of float on August 14 versus about 1.2% for Nvidia, even as ALAB shorts fell from the prior period. The near-term test is whether these design wins turn into shipped systems by 2027 and steady connectivity demand, or stall on design delays or hyperscaler concentration.