Overview
- On its earnings call, Nvidia said it will aggressively expand data‑center CPU deployments and signaled confidence that its presence could grow into one of the world’s largest CPU businesses.
- Nvidia’s Grace and Vera chips, introduced in 2023, are framed as high‑throughput CPUs built for data‑intensive operations rather than general‑purpose flexibility.
- Meta agreed to buy Nvidia CPUs on a standalone basis, while AMD separately announced a Meta CPU deal, underscoring a widening supplier contest for AI data‑center build‑outs.
- Nvidia highlighted its NVL72 system configuration of 36 CPUs and 72 GPUs, with analysts expecting some agentic workloads to shift toward a 1:1 ratio or even CPU‑only setups.
- Huang emphasized an architectural path that minimizes chiplet fragmentation to maximize data access and said more CPU roadmap details will be disclosed at next month’s developer conference in Silicon Valley.