NVIDIA Corporation (NASDAQ:NVDA) is doing something that initially sounds counterintuitive: making it easier for competing AI accelerators to live inside Nvidia-designed systems.
Reuters reported September 10 that inference-chip startup d-Matrix will use Nvidia’s NVLink Fusion technology to connect its next-generation Raptor processors directly into Nvidia data-center racks. The systems are expected to become available in 2027. Astera Labs, Inc. (NASDAQ:ALAB) is working with d-Matrix on custom connectivity solutions to move data rapidly across the system.
This creates an unusual three-layer architecture. d-Matrix supplies the accelerator, Nvidia supplies the rack-scale interconnect technology and architecture, and Astera helps solve the increasingly difficult problem of moving data between components.
Nvidia Can Win Without Selling Every Accelerator
For NVIDIA Corporation, the bull case is that NVLink can become more valuable even when Nvidia does not sell every accelerator. If competing XPUs plug into Nvidia’s rack architecture, Nvidia can shift from merely dominating GPUs to controlling an important system standard. That resembles the economics of owning the highway rather than insisting every vehicle on it be manufactured by the same company.
The bear case is obvious: NVLink Fusion deliberately makes alternative accelerators easier to deploy. If inference workloads become more heterogeneous, Nvidia could sacrifice some GPU unit share even while preserving influence over the broader system.
Astera Benefits From a Messier Rack
For Astera Labs, Inc., heterogeneous systems may be an even cleaner opportunity. More accelerators, memory pools, CPUs and switches create more connectivity bottlenecks. Astera sells retimers, fabric switches and other connectivity products designed to address the data-movement challenges created by increasingly complex AI racks. The bearish argument is that ALAB’s valuation already assumes substantial AI infrastructure growth and leaves little room for design delays or hyperscaler concentration.
Hedge-fund positioning became significantly more bullish on Astera Labs during Q2. Insider Monkey tracked 73 hedge funds in the stock, up from just 53 in Q1. Nvidia also increased to 285 hedge funds from 275. Short interest tells a more cautious story for Astera: roughly 6.6% of its float was short on August 14, compared with only about 1.2% for Nvidia. ALAB short interest nevertheless fell from the previous reporting period.
That divergence makes sense. Nvidia is the established platform owner; Astera is the higher-beta bet on connectivity complexity. NVLink Fusion reduces Nvidia’s need to win every accelerator socket, while ALAB benefits from the messier, more heterogeneous AI infrastructure that this strategy encourages. Of the two, Nvidia remains the lower-risk position, but Astera may have more upside if rack-scale connectivity becomes one of the next major AI bottlenecks.
READ NEXT:
The Trillion-Dollar AI Capex Gap: Why Too Much Hardware Could Be Nvidia’s Trap and Microsoft’s Opportunity and Redditors Are Skeptical of SpaceX’s Orbital Data Centers. Are They Right?
Hedge Fund Resource Center