Nvidia invests $3.5 billion in MediaTek, opens its AI racks to custom chips
The deal extends Nvidia's NVLink Fusion platform so hyperscalers can plug custom-built processors into its rack-scale AI systems instead of relying on off-the-shelf GPUs alone.
Original cover art, generated for this story. THE VISSION does not republish third-party press imagery.
- Nvidia is investing $3.5 billion in MediaTek convertible bonds and extending its NVLink Fusion interconnect platform to let custom chips connect to Nvidia's rack-scale AI systems, the companies announced on August 31.
- MediaTek already supplies custom AI accelerator designs to hyperscale customers; TechCrunch reports the company's custom-silicon business is projected to bring in $2 billion in revenue this year.
- Nvidia's Dion Harris said the company has expanded "beyond pure computing chips" and that "every cloud, every model builder" is deploying its platform in some form.
Nvidia is investing $3.5 billion in MediaTek convertible bonds and extending its NVLink Fusion platform to the Taiwanese chipmaker, the two companies said in a joint announcement on August 31. The deal lets custom processors built by MediaTek's hyperscale customers connect directly into Nvidia's rack-scale AI systems — architecture Nvidia calls MGX — rather than requiring those customers to build entirely around Nvidia's own GPUs.
"AI is transforming every computing platform — from the world's largest AI factories to the PC and the car," Nvidia founder and chief executive Jensen Huang said in the announcement. MediaTek vice chairman and chief executive Rick Tsai said the companies "share a vision for making advanced AI computing pervasive across the technology landscape." MediaTek already designs custom AI accelerator chips, known as ASICs, for hyperscale cloud customers; TechCrunch reports the business is projected to generate $2 billion in revenue this year, citing its own reporting rather than a company-stated figure.
The move addresses a real tension in Nvidia's business: major cloud customers including Google, Amazon and Microsoft have all built their own custom AI chips to reduce reliance on Nvidia's GPUs. Rather than compete purely on chip performance, Nvidia is positioning itself as the connective infrastructure — the racks, networking and software — that custom silicon plugs into. "Nvidia is an AI infrastructure company. We expanded beyond pure computing chips years ago," said Dion Harris, the company's senior director for hyperscaler infrastructure, according to TechCrunch. "Basically, every cloud, every model builder is deploying our platform in some shape, form, or fashion."
For a hyperscaler already committed to its own custom chip, this removes a real switching cost: it no longer has to choose between Nvidia's ecosystem and in-house silicon, which changes the calculus on how much further custom-chip investment actually saves. It also means Nvidia's revenue is becoming less tied to how many of its own GPUs ship, and more to whether its racks and interconnects remain the default AI infrastructure regardless of whose processor sits inside them.