NVLink Fusion
Nvidia's platform for letting other companies' custom AI chips plug into Nvidia's high-speed interconnect and rack-scale systems.
NVLink is Nvidia’s high-speed connection between chips, much faster than the standard PCIe slot a normal graphics card sits in. In a modern AI rack, dozens of chips have to behave like one big machine, and the interconnect is what makes that possible. NVLink Fusion opens that fabric to hardware Nvidia did not build.
The point is strategic. Amazon, Google, Microsoft, OpenAI and others are designing their own AI accelerators, often called XPUs, to depend less on buying Nvidia GPUs. NVLink Fusion supplies everything around such a chip, the interconnect, the memory architecture, the packaging and the rack qualification, so the custom accelerator still lives inside an Nvidia system and runs on Nvidia’s software. Losing the chip while keeping the rack is a far better outcome for Nvidia than losing the customer.