Intel Gets $5B Investment from Nvidia, Commits to Adopting NVLink (www.hpcwire.com)

Nvidia is investing $5 billion in Intel and the two companies announced a technology partnership in which Intel will adopt Nvidia’s proprietary NVLink interconnect to build next‑generation “superchips” that fuse Intel x86 CPUs with Nvidia GPUs for AI datacenter workloads, plus a consumer SoC that tightly couples an RTX GPU and a CPU for laptops. The deal lets Nvidia integrate Intel CPUs into its NVLink72 rack‑scale architecture (previously tied to Nvidia’s ARM “Vera” CPU), enabling whole racks to behave like single, massively parallel systems. Nvidia will also become a major buyer of Intel x86 CPUs for these systems. Technically, this is a major shift away from the long‑standing PCIe-centric data‑center model toward NVLink’s higher-bandwidth, lower-latency fabric—trading openness for scale. NVLink 5.0 can deliver ~1.8 TB/s bidirectional per Blackwell GPU, and fully connected 72‑GPU systems can reach aggregate bandwidths in the ~100+ TB/s range, outpacing PCIe/RoCE-based solutions like Intel’s Gaudi 3 (PCIe Gen5 host, 128 GB HBM2e, 1.2 TB/s to RAM, ~3.7 TB/s memory BW). For AI/ML this means larger model training and tighter CPU–GPU coupling with less scaling friction, but also greater vendor lock‑in around Nvidia’s interconnect. Intel’s accelerator business (Gaudi 3) remains separate for now, while the collaboration could reshape hardware stacks and supplier dynamics across datacenter and consumer AI markets.