NVIDIA launches NVLink Fusion for custom AI factories

NVIDIA has announced NVLink Fusion, a program to connect custom XPUs to NVIDIA’s AI infrastructure for semi-custom AI factories.

• The program is positioned as a way to improve performance, time to market, and risk mitigation for hyperscalers and AI-native companies building custom AI platforms.
• NVIDIA says NVLink Fusion supports NVLink scale-up networking, NVLink-C2C, the MGX rack-scale architecture, and the DSX reference architecture for AI factories.
• The article includes partner quotes from Intel, MediaTek, GUC, QCT/Quanta Computer, and Annapurna Labs (Amazon).
• It provides technical claims such as 3x lower latency versus off-the-shelf Ethernet, 10x higher packet rate, up to 1,152 accelerators in future NVLink roadmap configurations, and up to 6x energy efficiency for NVLink-C2C versus PCIe.
• This is a product/technology announcement and explanatory article, not a report of a completed deal or funding round.