Nvidia's $3.5B MediaTek investment targets custom AI chips for its data center racks

Nvidia puts $3.5 billion into MediaTek

Nvidia is investing $3.5 billion in Taiwanese chipmaker MediaTek, according to TechCrunch. Under the deal, MediaTek gains access to Nvidia technology that will let it design custom processors for AI companies and hyperscalers — silicon built to slot directly into data centers constructed around Nvidia hardware.

Conceding custom silicon to keep the scaffolding

The backdrop is a land grab. Major cloud providers and AI labs, including Amazon, Google, Microsoft, OpenAI and Anthropic, have been spending heavily on their own chips to reduce their reliance on Nvidia GPUs, as TechCrunch reports. Rather than resist that shift, Nvidia appears to be accommodating it: if customers insist on custom accelerators, Nvidia wants those accelerators to live in Nvidia racks, linked by Nvidia interconnects, inside what the company calls AI factories.

Dion Harris, Nvidia's senior director of HPC and AI hyperscaler infrastructure solutions, told reporters on a Monday call that Nvidia is "an AI infrastructure company" and expanded beyond pure computing chips years ago.

NVLink Fusion is the hinge

The practical core of the partnership is access to Nvidia's NVLink Fusion ecosystem. NVLink is the interconnect that lets chips — including ones Nvidia did not manufacture — exchange data at high speed. With that access, MediaTek can keep producing custom ASICs designed by cloud companies or AI labs for their specific workloads, while those chips take advantage of Nvidia's scaling technologies and rack-level architecture.

Harris argued that essentially every cloud and model builder already runs Nvidia's platform in some form, so MediaTek offering this extension to its customers lets them standardize on a single rack-scale platform across their AI factories and deploy their custom silicon alongside it.

TechCrunch also notes the circular character of Nvidia's financing in recent years: the company has repeatedly invested in businesses that then feed revenue and design wins back into its own ecosystem.

A pattern, not a one-off

The MediaTek deal follows a similar arrangement announced the week before with Amazon Web Services, TechCrunch reports — though that one came without a direct investment. AWS committed to deploying an additional 2 million Nvidia GPUs across its infrastructure and to integrating NVLink Fusion.

What's in it for MediaTek

MediaTek has been quietly assembling a custom data center ASIC operation. In June, the company said it expects that business to generate $2 billion in revenue in 2026 and aims to capture more of the market in the years ahead, according to TechCrunch. It has not publicly named its custom AI chip customers. The company's design pedigree spans smartphones, smart home devices, automobiles and wireless communications — a capability Nvidia now wants pointed at hyperscaler workloads, with Nvidia riding along.

Beyond the data center

The partnership also extends work the two companies already do together. MediaTek will continue collaborating on DGX Spark, Nvidia's small desktop AI computer aimed at developers, and on RTX Spark, Nvidia's push to bring its AI technology to consumer AI PCs. On the automotive side, MediaTek's cockpit platforms already rely on Nvidia RTX graphics and sit alongside Nvidia Drive AGX, the in-car computing platform for autonomous driving workloads. The companies say they will keep developing platforms for AI-powered, software-defined vehicles.

Nvidia CEO Jensen Huang said in a statement that AI is transforming every computing platform, "from the world's largest AI factories to the PC and the car," and that the companies are building platforms that let customers create differentiated AI systems at scale.

Why it matters

The investment reveals Nvidia's answer to the custom-silicon wave sweeping Big Tech. Instead of fighting hyperscalers' in-house chip programs, Nvidia is trying to make them irrelevant at the infrastructure layer: whatever accelerator a cloud provider designs, the surrounding racks, interconnects and networking remain Nvidia's. Custom chips become tenants in a building Nvidia owns.

For MediaTek, Nvidia's capital and ecosystem access could accelerate its ASIC ambitions and turn it into a credible factory for hyperscaler silicon. For Nvidia's customers, the arrangement lowers the friction of mixing homegrown accelerators with commercial GPUs — but it may deepen their dependence on Nvidia at a different layer of the stack. Whether opening the interconnect ecosystem entrenches Nvidia's platform or eventually commoditizes it is one of the defining questions of the next phase of the AI buildout.