Nvidia is investing EUR 3 billion in MediaTek through convertible bonds, the two companies confirmed this week, in a deal that significantly deepens their existing partnership. As part of the agreement, MediaTek will adopt Nvidia’s NVLink Fusion platform across its custom AI accelerators, local AI systems, and automotive products.
The arrangement works in both directions. MediaTek gains the ability to design accelerators built for Nvidia’s fully developed rack-scale platforms, while Nvidia secures a foothold in the fast-growing market for custom AI silicon. For a company that already dominates the merchant AI accelerator business, the move is as much about defending market reach as it is about generating new revenue.
Why Nvidia Is Backing Custom Silicon
Nvidia has become the world’s largest supplier of AI accelerators, with no direct competitor of comparable size. But the landscape is shifting as more organizations pursue bespoke chips. AWS, Google, Meta, Microsoft, and now OpenAI are all developing their own AI accelerators, in part to reduce reliance on costly merchant GPUs.
Rather than resist that trend, Nvidia is extending NVLink Fusion beyond its own hardware to keep custom chips within its ecosystem. If customers swap some Nvidia GPUs for their own XPUs, Nvidia wants those XPUs linked through NVLink, paired with Nvidia CPUs where suitable, and deployed inside Nvidia networking and rack architectures. Under the new structure, a customer can bring an XPU design to MediaTek, and MediaTek and Nvidia will supply much of the surrounding technology around the compute engine itself.
What NVLink Fusion Brings to MediaTek’s Accelerators
MediaTek will use NVLink Fusion as the foundation for custom accelerators that can advance alongside future Nvidia architectures. The platform includes several core components:
The NVLink Fusion chiplet connects custom XPUs to Nvidia’s NVLink scale-up fabric using electrical or photonic interconnects. NVLink-C2C delivers high-bandwidth, energy-efficient links between XPUs, Nvidia Rosa CPUs, and other compatible processors. Nvidia NVHBM supports customized memory configurations while freeing up more silicon area for compute.
By building on this platform, MediaTek’s prospective clients can focus on their differentiated compute architecture while Nvidia and MediaTek handle connectivity, memory design, packaging, manufacturing, and rack-level technologies.
A Pre-Built Rack-Scale Path for Smaller Players
The practical outcome is that MediaTek’s customers receive a pre-developed rack-scale platform for their custom AI accelerators, something not readily available elsewhere. Nvidia typically supplies complete AI infrastructure platforms rather than standalone accelerators, so the MediaTek deal aligns closely with that strategy.
The distinction matters most for smaller companies. Hyperscalers
Source
Image: tomshardware.com
