Nvidia is not pretending every AI chip in every rack has to say Nvidia on the package. TechCrunch reported on August 31, 2026 that Nvidia is investing $3.5 billion into Taiwanese chipmaker MediaTek. As part of the deal, MediaTek will adopt Nvidia technology that helps it design custom chips for AI companies and hyperscalers that plug straight into Nvidia-based data centers.
Amazon, Google, Microsoft, OpenAI, and Anthropic have been building their own silicon so they lean less on Nvidia GPUs. This MediaTek bet lets Nvidia give ground on the custom chip itself while staying the scaffolding around it. Dion Harris, Nvidia's senior director of HPC and AI hyperscaler infrastructure solutions, told reporters Nvidia expanded beyond pure computing chips years ago and called the company an AI infrastructure company.
The key tech is NVLink Fusion. It lets chips talk to each other quickly even when they are not Nvidia chips. With that stack, MediaTek can keep building customer-designed ASICs while customers still standardize on Nvidia's rack-scale platform. Harris said every cloud and every model builder is already deploying Nvidia's platform in some form, and MediaTek can offer that same rack standard to its own customer base.
MediaTek expects its custom data center ASIC business to generate about $2 billion in revenue in 2026. The companies will also keep collaborating on DGX Spark, Nvidia's small developer desktop AI computer, plus RTX Spark for consumer AI PCs, and platforms for software-defined vehicles that lean on Nvidia graphics and Drive AGX.
Jensen Huang said AI is transforming every computing platform from the biggest AI factories to the PC and the car, and that the partnership gives customers freedom to build differentiated AI systems at huge scale.
For regular folks and for developers, the takeaway is boring and important. The fight is not only "who makes the fastest GPU." It is "who owns the wires, the racks, and the standards the racks speak." If you run local models on a Mac Studio or a small Nvidia box, you are tasting the edge of that same stack. If your company buys cloud GPUs, your bill still rides on whether custom chips can talk to the dominant networking fabric.
I would not cheer for monopoly or for fragmentation. I would watch whether custom chips from big cloud labs still need Nvidia's glue. This deal says Nvidia thinks they will. That is the useful close. Follow the interconnect, not just the GPU SKU.
