Nvidia Builds IP Licensing Empire with NVLink
In the AI race, Nvidia dominates competitors through technology licensing model.
1Nvidia's NVLink Strategy
Nvidia has started licensing its NVLink Fusion technology to maintain dominance in the AI chip market. MediaTek agreed to use this technology in exchange for Nvidia's $3.5 billion investment in convertible bonds. NVLink was initially developed for high-speed GPU connections but has now become a broad semiconductor IP licensing portfolio. This move encourages competitors to use Nvidia's solutions rather than developing their own.
2Strategic Partnership with MediaTek
MediaTek became the latest company to license NVLink Fusion technology for its data center chips. Under the deal, Nvidia invested $3.5 billion in MediaTek's convertible bonds. MediaTek's designs will also be used in Nvidia's future DGX and RTX Spark systems. Major chipmakers like Amazon, Fujitsu, and Qualcomm also plan to integrate NVLink technology.
3Market Dynamics
As companies developing custom AI chips (XPUs) increased, Nvidia's market share was questioned. However, the company strengthened its position by licensing NVLink instead of letting competitors develop their own solutions. It took years for rivals like Broadcom to develop NVLink alternatives. Nvidia's licensing model enables customers to access scalable solutions faster.
4Technological Advantages
NVLink Fusion offers a chip-to-chip (C2C) variant for connecting CPUs to GPUs (or XPUs) in a memory-coherent fabric. It also provides a switched fabric option for connecting multiple accelerators. This technology is compatible with Nvidia's MGX rack designs, part of the Open Compute Project. Companies like MediaTek can equip existing NVL72 racks with their compute blades.
5Amazon's Example
Amazon uses NVLink Fusion technology in its Trainium series of AI accelerators. The company employed Nvidia's MGX NVL72 reference design for its Trainium3-based rack systems launched last year. The upcoming Trainium4, expected later this year, will switch from Amazon's in-house NeuronLink to NVLink Fusion. This demonstrates the rapid adoption of Nvidia's technology.
6Nvidia's Gains
By licensing NVLink to competitors, Nvidia doesn't just earn licensing fees. It also ensures companies become dependent on its other products. There's potential to sell two GPUs per partner CPU, one Grace or Vera CPU per two XPUs, and up to nine NVSwitches per 72 accelerators. Customers using NVLink may also opt for InfiniBand or Spectrum-X Ethernet kits.
7Network Expansion Strategy
Nvidia offers five different networking technologies through NVLink Fusion: from CPU to GPU, scalable fabric to Spectrum-X and InfiniBand. Dion Harris, Nvidia's Senior Director of HPC and AI Hyperscale Infrastructure, states this platform provides flexibility to customers. The NVLink Fusion licensing structure hasn't been disclosed yet, but Nvidia won't give it away for free.
8Dominance in AI Market
Thanks to the AI boom, Nvidia is now the world's largest data center Ethernet networking provider. The company's quarterly networking revenue exceeds $15 billion. NVLink Fusion licenses and NVSwitch sales will further boost this revenue. Nvidia's licensing ambitions continue with new technologies like NVHBM. This strategy positions Nvidia at the center of the AI ecosystem.
Original article: www.theregister.com - Articles — Nvidia'nın NVLink ile IP Lisans İmparatorluğu