Nvidia and MediaTek are expanding a partnership that already spans personal computing and automotive systems into a broader edge-to-cloud AI platform strategy.
The companies announced on August 31 that MediaTek will adopt Nvidia NVLink Fusion, giving customers developing custom XPUs a path to connect those processors into Nvidia’s rack-scale AI infrastructure.
Both Nvidia and MediaTek published first-party announcements describing the expanded collaboration.
What is changing
MediaTek is best known by many consumers for smartphone and connected-device processors, but the company also designs custom silicon and data-centre components.
Under the expanded partnership, MediaTek says it will use NVLink Fusion for custom AI silicon intended for hyperscalers, cloud providers and model developers. Nvidia describes the same initiative as a way to bring custom XPUs into NVLink-connected AI factories.
The partnership also continues across local AI computing and automotive platforms.
Why NVLink matters
A large AI system cannot be understood simply by counting accelerators.
When hundreds or thousands of processors work on the same workload, the connections between them become critical. If processors cannot exchange data quickly enough, expensive compute units spend more time waiting and less time calculating.
NVLink is Nvidia’s high-bandwidth interconnect technology for connecting processors inside accelerated-computing systems. NVLink Fusion extends parts of that ecosystem to custom processors designed by other companies.
That is strategically important because the largest cloud operators increasingly want chips optimized for their own workloads rather than relying exclusively on general-purpose accelerators.
Nvidia is opening part of the rack without giving up the rack
Custom AI accelerators could appear to threaten Nvidia: every workload moved to another chip is a workload not running entirely on an Nvidia GPU.
NVLink Fusion offers a different response. Instead of requiring every processor in an AI system to be Nvidia-designed, Nvidia can make its interconnect and rack architecture useful even when customers introduce custom silicon.
That potentially preserves Nvidia’s influence over the wider computing platform even as accelerator designs diversify.
MediaTek benefits from the opposite direction. Its custom-silicon expertise gains a route into large AI systems built around an established high-speed fabric.
The relationship already reaches personal computing
This is not a new alliance.
MediaTek previously worked with Nvidia on the GB10 Grace Blackwell Superchip used in Nvidia’s compact AI systems, and the companies have announced collaboration around personal computing and automotive platforms.
The new announcement broadens that relationship into a more explicit multigenerational infrastructure partnership.
The significance is therefore not one new chip. It is the possibility of MediaTek becoming a recurring silicon partner across multiple layers of Nvidia-centered computing systems.
What to watch
The next evidence will come from actual products and customers.
Important questions include:
- which cloud providers or hyperscalers use MediaTek-designed custom XPUs with NVLink Fusion;
- how much of the resulting rack remains dependent on Nvidia technology;
- whether other custom-silicon vendors join the same ecosystem;
- and whether the partnership produces meaningful alternatives to conventional GPU-only AI clusters.
The larger trend is clear: AI infrastructure is becoming heterogeneous. The competitive question is increasingly not only who makes the accelerator, but who defines the fabric that connects the entire system.