Industry story

Nvidia Invests $3.5B in MediaTek to Embed Its AI Infrastructure in Custom Chips

by Ken Rona ·September 1, 2026 ·Techcrunch Ai

Nvidia is investing $3.5 billion in MediaTek, and the announcement leads with support for custom AI chips. The mechanics run the other way. The big cloud companies design their own chips to escape Nvidia's prices. This deal pulls those chips into NVLink Fusion, Nvidia's system for wiring chips together inside a data center. Nvidia did the same with Amazon a week earlier. The cloud companies get to tell their boards they're reducing their Nvidia dependence, while their escape hatches are being built on Nvidia's wiring.

Our prediction

By Nvidia's GTC 2027 keynote (March 2027), no MediaTek-designed NVLink Fusion ASIC will be running a frontier training workload at production scale that displaces Nvidia GPUs. The Fusion chips that ship will be inference or offload silicon sitting alongside Nvidia GPUs, not replacing them.

Medium confidence

Why The signal in this deal is that Nvidia set the interconnect spec, the certification, and the upgrade cadence, which means it decides what "interoperable" means and when. Every prior Nvidia platform move has made its own GPUs stickier, not easier to leave, and Fusion's whole design keeps training collectives running best on homogeneous Nvidia parts. The likely 2027 outcome is MediaTek ASICs handling prefill, inference, or offload while Nvidia GPUs still own the training loop, because that's the only configuration where Nvidia gets both the toll and the GPU sale. The opposite outcome, a MediaTek chip actually pulling a training run off Nvidia silicon at scale, would require Nvidia to have engineered away its own moat, and nothing in a $3.5B investment structured to keep customers on its fabric suggests it did.

Revisit by Apr 15, 2027

Right if: We're right if, by then, no hyperscaler or lab has run a frontier-scale training job on MediaTek Fusion silicon displacing Nvidia GPUs, and the shipped Fusion ASICs are inference or offload parts.
Wrong if: any of Google, AWS, OpenAI, Anthropic, or Microsoft publicly runs production frontier training on a MediaTek-designed chip that replaces Nvidia GPUs on the fabric.

Full analysis

Nvidia is putting $3.5 billion into MediaTek and bringing it inside NVLink Fusion, Nvidia's system for wiring AI chips together inside a data center so they can work as one machine. The pitch: cloud companies can design all the custom chips they want, and Nvidia still owns the wiring those chips plug into. The same arrangement landed with Amazon a week earlier.

What's actually being decided for anyone who buys AI computing: whether to let Nvidia own the layer underneath even your non-Nvidia chips. That choice is hard to undo. Once your software and operations assume Nvidia's wiring, replacing it is a multi-year project nobody volunteers for. Standards quietly get set precisely when nothing forces the decision. Now is that moment.

The Skeptic. The $3.5 billion is small change for Nvidia. What it buys is control of the standard: every chip that speaks NVLink Fusion stays in Nvidia's orbit forever. The cloud giants say their homegrown chips reduce their dependence on Nvidia; after this deal, those chips still run inside Nvidia's system, on Nvidia's schedule. Think of it as a franchise: you may open your own restaurant, and the franchisor still collects. In plain terms, Nvidia found a way to get paid even when you buy someone else's chip.

The Compute Pragmatist. Nvidia's wiring was designed for fleets of identical Nvidia chips doing identical work. A MediaTek chip doing a different kind of job has different needs, and nobody yet knows whether it gets the advertised speed in practice or something like half of it. Vendors will quote the number on the slide; measured results will arrive a year later. Anyone budgeting on the press-release figure should build in a large discount.

The Researcher. Underneath the strategy, the engineering credit is real: making chips from different makers work together at speed is genuinely hard, and Nvidia is the company making it work. That is exactly why owning that layer is worth more to them than selling any single chip.

Read the source story →

Also covered this issue

Google and Amazon launch PyTorch forks to challenge Nvidia's AI chip dominance techcrunch-ai
Google and Amazon made PyTorch work on their chips with a config flag instead of code rewrites, letting companies negotiate chip prices instead of staying locked to Nvidia.
Princeton-led paper maps fragile state of AI agent evaluation ai-snake-oil
Published benchmark scores for AI agents don't predict how they'll actually perform in your business workflows, so testing them yourself before deployment is now mandatory.
Hackers breach OpenCode Console, exploit AI API keys at scale techcrunch-ai
Hackers stole API keys from an AI coding tool, and the AI models meant to help defend couldn't analyze the attack because safety filters blocked them.

Nvidia Invests $3.5B in MediaTek to Embed Its AI Infrastructure in Custom Chips

Full analysis

Comments