Nvidia pivots to IP licensing as rivals adopt NVLink Fusion interconnect
The chip giant is monetizing the 'plumbing' of AI infrastructure by licensing its proprietary networking tech to competitors.
Nvidia is expanding its business model by licensing its proprietary NVLink Fusion interconnect technology to rival chip designers and cloud providers. The move allows companies building custom AI accelerators to utilize Nvidia's high-speed networking and rack designs rather than developing their own proprietary standards.
Several industry leaders have already committed to the integration. MediaTek has adopted NVLink Fusion for its datacenter XPU offering, a partnership underscored by a $3.5 billion investment from Nvidia in MediaTek convertible bonds. Amazon is also shifting its strategy, with future Trainium4 accelerators expected to integrate NVLink Fusion to replace or augment its in-house NeuronLink interconnects. Other firms planning to integrate the technology include Marvell, Fujitsu, Qualcomm, and Arm.
The shift to infrastructure plumbing
This strategic pivot comes as major tech firms, including Meta, Microsoft, and OpenAI, accelerate the development of custom AI ASICs to reduce their total reliance on Nvidia GPUs. By licensing NVLink Fusion and providing MGX rack reference designs via the Open Compute Project, Nvidia is effectively monetizing the underlying infrastructure of the AI era.
As part of this ecosystem expansion, Nvidia has introduced NVHBM. This technology integrates the memory controller directly into the HBM base die, which Nvidia claims can free up to 25% of the XPU die area for compute and boost memory bandwidth by up to 30% compared to HBM4e.
Creating a new industry standard
This strategy transforms Nvidia from a traditional chip vendor into an IP powerhouse, mirroring the business model used by Arm. By controlling the interconnect standard, Nvidia creates a systemic lock-in effect. Competitors' custom silicon must now conform to Nvidia's specifications to scale efficiently within production racks, ensuring Nvidia profits even when customers use rival silicon.
"The beauty about this platform is it provides the flexibility and the fungibility in order to meet the customers where they are," said Dion Harris, Nvidia's senior director of HPC and AI Hyperscale Infrastructure.
The road ahead
Nvidia's networking division has already seen massive growth, with quarterly networking revenue reported at nearly $11 billion in Q4 2026. The industry will now be watching to see if NVLink Fusion becomes the universal standard for AI clusters, potentially marginalizing other interconnect efforts. While the adoption among hyperscalers is accelerating, the long-term challenge remains whether these partners will eventually seek to break away from Nvidia's IP to regain full architectural independence.