Nvidia is putting $3.5 billion into MediaTek, a move that says as much about the future of AI infrastructure as it does about one chipmaker investing in another. The core idea is simple: as more cloud companies and AI labs design their own AI chips, Nvidia wants those chips to keep running inside its broader data center ecosystem.
The partnership gives MediaTek access to Nvidia technology for designing custom chips that can be plugged directly into Nvidia-based data centers. It is a strategic response to a market where Nvidia’s GPUs remain central, but major customers are also trying to reduce their dependence on them.
A Deal Built Around Custom AI Chips
MediaTek will adopt Nvidia technology that helps it design custom chips for AI companies and hyperscalers. These chips are meant to connect into Nvidia-based data centers rather than sit outside them as separate, incompatible systems.
That matters because a growing list of companies, including Amazon, Google, Microsoft, OpenAI and Anthropic, are investing in their own chips. Their goal is to be less reliant on Nvidia’s GPUs for AI workloads. Nvidia’s MediaTek deal shows a way to accept that shift while still keeping the surrounding infrastructure anchored to Nvidia.
Instead of treating custom silicon as a direct threat, Nvidia is positioning itself as the platform those custom chips can join. The company can give ground on some chip design while protecting its role in the data center stack that surrounds the chips.
NVLink Fusion Is the Centerpiece
The deal gives MediaTek access to Nvidia’s NVLink Fusion ecosystem. That includes NVLink, the technology that allows chips to communicate quickly with each other, even when those chips are not made by Nvidia.
For AI infrastructure, that communication layer is critical. AI companies and cloud providers do not only need individual processors; they need large systems where many chips can work together across racks and data centers. Nvidia is using NVLink Fusion to make its ecosystem more open to custom chips while keeping Nvidia’s architecture at the center of deployment.
Dion Harris, Nvidia’s senior director of HPC and AI hyperscaler infrastructure solutions, framed the company’s role broadly on Monday during a call with reporters. He said Nvidia expanded beyond pure computing chips years ago and described the company as an AI infrastructure company.
That framing explains the logic of the MediaTek investment. Nvidia is not only selling GPUs. It is selling the connective architecture, software stack and data center blueprint that customers use to build AI factories.
Why MediaTek Fits Nvidia’s Strategy
MediaTek has been building its custom data center ASIC business. In June, the company said it expects that business to generate $2 billion in revenue in 2026 and wants to target more of the market in the coming years.
The Nvidia partnership could help MediaTek expand that work by giving its customers a path into Nvidia’s rack-scale infrastructure. MediaTek can continue building custom chips designed by cloud companies or AI labs for specific workloads, while using Nvidia’s scale-up and scale-out technology stack and ecosystem.
Harris said the arrangement lets MediaTek offer its customers a way to standardize on rack-scale infrastructure across AI factories. It also allows those customers to deploy their custom chips alongside Nvidia systems using the same standard platform.
That is the circular part of Nvidia’s recent financing strategy. Nvidia invests in companies that can send more activity back into its own ecosystem. In this case, MediaTek may gain more room to grow in custom AI chips, while Nvidia keeps those chips tied to its infrastructure.
Part of a Wider Infrastructure Push
The MediaTek deal follows a related move with Amazon Web Services. Last week, Nvidia announced a similar partnership with AWS, though without a direct investment. AWS will deploy an additional 2 million Nvidia GPUs across its infrastructure and will also integrate NVLink Fusion.
Taken together, the AWS and MediaTek partnerships point in the same direction. Nvidia is preparing for a world where hyperscalers and AI model builders use a mix of Nvidia GPUs and their own custom silicon. The company’s answer is to make Nvidia’s platform the common environment where both can operate.
That approach protects Nvidia from a simple either-or market. If customers buy Nvidia GPUs, Nvidia benefits directly. If customers build more custom AI chips, Nvidia still has a chance to benefit if those chips plug into its networking, rack-scale architecture and broader AI infrastructure.
Beyond Data Centers
The companies are also extending collaboration beyond custom data center chips. MediaTek’s experience includes chips for smartphones, smart homes, automobiles and wireless communications. That background is relevant to several Nvidia projects outside hyperscale AI infrastructure.
MediaTek will continue to work with Nvidia on DGX Spark, Nvidia’s small developer-focused desktop AI computer. The companies have also extended their collaboration with RTX Spark, Nvidia’s effort to bring its AI technology into consumer AI PCs.
Automotive work is another part of the relationship. MediaTek and Nvidia will continue developing platforms for AI-powered, software-defined vehicles in what the source describes as the era of physical AI. MediaTek’s auto platforms use Nvidia’s RTX graphics for intelligent vehicle cockpits and work alongside Nvidia Drive AGX, Nvidia’s in-car computing platform for autonomous driving workloads.
Jensen Huang, founder and CEO of Nvidia, said in a statement that AI is transforming every computing platform, from the world’s largest AI factories to the PC and the car. The MediaTek deal reflects that same broad ambition: Nvidia wants its accelerated computing technology present wherever AI systems are being built, even when the chips are customized by someone else.
The result is a more flexible version of Nvidia’s AI chip strategy. The company is not only defending its GPU business. It is building a framework where custom AI chips, AI PCs, developer systems and vehicle platforms can all connect back to Nvidia’s technology stack.