Home / Technology / Nvidia's AI Strategy: Sharing Racks, Owning Standards
Nvidia's AI Strategy: Sharing Racks, Owning Standards
12 Sep
Summary
- Nvidia's NVLink Fusion allows competitors' AI chips in Nvidia racks.
- The new systems integrate d-Matrix's Raptor processors by 2027.
- This move could position Nvidia to control system standards, not just GPUs.
Nvidia Corporation is adopting a strategy that integrates competing AI accelerators into its own data-center systems using NVLink Fusion technology. This move, with systems expected by 2027, allows startups like d-Matrix to connect their next-generation Raptor processors directly into Nvidia's racks. Astera Labs is collaborating on solutions to manage rapid data transfer between components within this novel three-layer architecture.
This approach could enable Nvidia to dominate system standards, akin to owning the infrastructure rather than just the vehicles on it. However, it also presents a risk of Nvidia conceding some GPU market share if AI workloads diversify significantly. For Astera Labs, these increasingly complex heterogeneous systems create opportunities by highlighting the need for advanced connectivity products, such as their retimers and fabric switches, to overcome data-movement bottlenecks.
