Nvidia puts $3.5B into MediaTek so custom AI chips stay on its rack

On Monday, Nvidia announced a $3.5 billion investment in MediaTek, the Taiwanese chipmaker. Under the deal, MediaTek adopts Nvidia's interconnect technology to design custom chips for AI companies and hyperscalers that plug directly into Nvidia-based data centers.

That last part is the point. A growing list of AI companies and cloud providers, from Amazon, Google and Microsoft to OpenAI and Anthropic, is building its own silicon to become less reliant on Nvidia GPUs. This deal is Nvidia's answer: concede the custom-chip fight, but make sure those chips still run on its hardware.

NVLink Fusion is the piece MediaTek gets

The partnership hands MediaTek access to Nvidia's NVLink Fusion ecosystem, including NVLink, the technology that lets chips, even ones that are not Nvidia's, talk to each other quickly.

Dion Harris, Nvidia's senior director of HPC and AI hyperscaler infrastructure solutions, put it plainly on a Monday call with reporters. "Nvidia is an AI infrastructure company," he said. "We expanded beyond pure computing chips years ago." With the deal, MediaTek can keep building chips that cloud companies and AI labs design for their own workloads, using what Nvidia calls its "proven scale-up and scale-out technology stack and ecosystem" and its "rack-scale architecture."

"Basically, every cloud, every model builder is deploying our platform in some shape, form, or fashion," Harris said. MediaTek's version of that extension, he added, lets customers standardize on rack-scale infrastructure across their AI factories and deploy custom chips right alongside Nvidia ones, on the same standard platform.

It is not Nvidia's only recent move in this direction. Last week it signed a similar partnership with Amazon, without the direct investment: AWS will deploy 2 million more Nvidia GPUs across its infrastructure and integrate NVLink Fusion.

MediaTek's ASIC business is the prize

MediaTek has not publicly shared which customers it works with on custom AI chips. It did say in June that it expects its custom data center ASIC business to generate $2 billion in revenue this year, and it hopes to capture more of the market in coming years. Nvidia's money arrives exactly when that business needs scale.

MediaTek's chip expertise spans smartphones, smart homes, automobiles and wireless communications, and the deal extends the collaborations that come with it. The two companies will keep working on DGX Spark, Nvidia's small developer-focused desktop AI computer, and on RTX Spark, Nvidia's push to put its AI tech into consumer AI PCs. They will also continue "developing platforms for AI-powered, software-defined vehicles in the era of physical AI," per a press release. MediaTek's automotive platforms rely on Nvidia RTX graphics for intelligent vehicle cockpits and work alongside Drive AGX, Nvidia's in-car computing platform for autonomous driving.

"AI is transforming every computing platform," Jensen Huang said in a statement. "Together, we're building platforms that bring NVIDIA accelerated computing to new markets and give customers the freedom to create differentiated AI systems at enormous scale."

For anyone building on custom silicon, the practical line to take away is simple: the chip may stop being Nvidia's, but the rack still will be.