Nvidia is plowing $3.5 billion into MediaTek, giving the Taiwanese chipmaker access to the NVLink Fusion ecosystem so its custom silicon can plug straight into Nvidia racks. The move lets Nvidia collect rent on the data-center scaffolding even as Amazon, Google, Microsoft, OpenAI and Anthropic build their own accelerators to wean off Nvidia GPUs.
The circular financing loop
Nvidia has a habit of funding companies that turn around and buy its gear. MediaTek gets NVLink, the interconnect that lets any chip talk to any other chip at speed, and Nvidia gets a deeper moat around the rack-scale architecture that every hyperscaler is standardizing on. Dion Harris, who runs HPC and AI hyperscaler infrastructure solutions at Nvidia, put it plainly: the company stopped being a pure compute-chip vendor years ago and now sells the whole infrastructure stack.
MediaTek's ASIC ambition
MediaTek told investors in June that its custom data-center ASIC business should hit $2 billion in revenue this year. The company has deep experience in phone, auto and wireless silicon but a thin track record in AI accelerators. Nvidia's capital and interconnect tech are meant to change that, while also feeding Nvidia's own DGX Spark desktop box and the RTX Spark push into consumer AI PCs.
The auto tether
The partnership also extends to software-defined vehicles. MediaTek's cockpit platforms already lean on Nvidia RTX graphics and sit beside Drive AGX for autonomous workloads. Jensen Huang framed the whole thing as bringing accelerated computing to every platform from the largest AI factories down to the car.
What to watch
AWS just signed a similar NVLink Fusion deal without taking Nvidia money, adding 2 million GPUs to its fleet. The test is whether MediaTek can actually win custom-ASIC sockets at scale or whether the $3.5 billion simply buys Nvidia a seat at the table while the hyperscalers keep designing in-house.