Nvidia Corp (NASDAQ:NVDA, XETRA:NVD) is making a US$3.5 billion strategic investment in Taiwan’s MediaTek as the artificial intelligence giant moves to protect its dominance from an emerging threat - its biggest customers increasingly designing AI chips of their own.
The investment, announced on August 31 through the purchase of MediaTek convertible bonds, comes alongside a substantial expansion of the companies’ partnership across data centres, personal computing and automotive technology.
But the most significant part of the deal may be Nvidia’s decision to bring MediaTek into its NVLink Fusion ecosystem, potentially revealing how Nvidia intends to remain at the centre of AI infrastructure even when the processors doing the computing are not Nvidia GPUs.
Big Tech wants its own chips
Nvidia remains overwhelmingly influential in AI computing, but Amazon, Google, Microsoft and some of the biggest AI developers are increasingly pursuing custom silicon designed specifically for their own workloads.
The attraction is straightforward. Custom accelerators can give major cloud operators greater control over performance, energy consumption and costs while reducing their reliance on Nvidia GPUs.
That creates a potentially important long-term challenge for Nvidia.
Instead of trying to prevent that shift, however, Nvidia appears increasingly willing to accommodate it.
Under the MediaTek agreement, customers will be able to develop their own custom AI accelerators, or XPUs, through MediaTek and integrate them into Nvidia’s NVLink-connected rack-scale infrastructure.
In other words, Nvidia does not necessarily need to manufacture every processor inside an AI data centre if its technology remains the infrastructure connecting everything together.
Nvidia changes the battlefield
NVLink Fusion sits at the centre of that strategy.
The platform enables customised processors to connect with Nvidia technologies covering high-speed chip interconnects, memory, networking and rack-scale computing.
MediaTek customers can therefore concentrate on designing the specialised computing element while Nvidia and MediaTek provide much of the supporting architecture required to turn that silicon into a production-ready AI system.
It represents a significant expansion of Nvidia’s competitive moat.
Rather than competing solely on whether a customer chooses an Nvidia GPU over a custom accelerator, Nvidia is attempting to make its architecture useful whichever processor the customer chooses.
Nvidia senior director Dion Harris summed up the shift by telling reporters: “Nvidia is an AI infrastructure company.”
That distinction is becoming increasingly important as the AI hardware market evolves.
From GPU giant to AI infrastructure giant
The MediaTek deal also extends well beyond hyperscale data centres.
The companies are collaborating on future generations of Nvidia RTX Spark and DGX Spark chips for AI PCs, developer systems and workstations, while continuing their partnership on technology for AI-powered and software-defined vehicles.
MediaTek already worked with Nvidia on the GB10 Grace Blackwell Superchip powering DGX Spark, combining Nvidia GPU technology with MediaTek’s system-on-chip expertise.
The US$3.5 billion investment therefore strengthens a relationship spanning everything from enormous AI factories to PCs and cars.
More importantly for investors, it provides another indication that Nvidia increasingly sees its future as much bigger than selling GPUs.
Its competitive advantage is expanding across software, networking, interconnect technology, memory architecture, advanced packaging and complete rack-scale systems.
Custom AI chips may ultimately take some workloads away from Nvidia GPUs.
Nvidia’s emerging strategy appears designed to make sure those chips still have to operate inside an ecosystem that Nvidia helped build.