If you are a cloud provider trying to design your own AI chip, you will still need Nvidia's plumbing. That is the practical implication of a $3.5 billion investment Nvidia announced on 31 August 2026 in MediaTek, a Taiwanese chipmaker, structured through convertible bonds.
The headline figure matters less than what comes with it. Under the deal, MediaTek gains access to Nvidia's NVLink Fusion platform, a technology that gives chipmakers ready-made connectors and specialised memory so that custom AI chips can plug directly into Nvidia-based data-centre systems. For any company building its own accelerator silicon, that compatibility is hard to walk away from.
Amazon, Google, Microsoft, OpenAI and Anthropic are all investing in their own AI chips, in part to reduce reliance on Nvidia's processors. The NVLink Fusion arrangement does not block that ambition, but it does mean those chips are more likely to work alongside Nvidia infrastructure rather than instead of it. Nvidia's founder and chief executive Jensen Huang is positioning the company as the connective tissue of AI computing, a role that persists even as customers build around it.
For MediaTek, the benefits are concrete. NVLink Fusion saves engineering time and cost by providing a tested, standardised interface. The company, which is competing with Broadcom and Marvell Technology for contracts to help data-centre operators design custom silicon, gains a significant technical advantage by having Nvidia's interconnect built in from the start.
MediaTek vice chairman and chief executive Rick Tsai leads the company into a partnership that extends well beyond data centres. The two companies have been working together for some time. MediaTek previously collaborated with Nvidia on the GB10 Grace Blackwell Superchip, which powers the DGX Spark developer computer. The new deal deepens that relationship across three areas: AI data-centre infrastructure, consumer computing, and automotive systems.
On the consumer side, the partnership extends collaboration on Nvidia RTX Spark and DGX Spark chips for personal and enterprise computers. In automotive, it builds on AI-powered platforms using MediaTek's Dimensity Auto line. MediaTek had also previously partnered with Alphabet's Google, giving it a track record of working with the large technology companies that are now the most active buyers of custom AI silicon.
Several things remain unclear. The terms of the convertible bonds, including conversion price, maturity date and interest rate, have not been disclosed. There is no public timeline for when products built on NVLink Fusion through this partnership will reach customers, and no specific market-share or revenue targets have been announced. Whether regulatory approvals are required, given MediaTek's Taiwanese status, is also an open question.
What the deal does confirm is a pattern. Rather than competing directly with the companies designing custom chips, Nvidia is making itself useful to the designers. The major cloud providers may reduce their purchases of Nvidia graphics processing units over time, but the infrastructure logic of NVLink Fusion keeps Nvidia embedded in whatever replaces those GPUs. The gap between the stated goal of independence from Nvidia and the practical reality of building within its ecosystem remains wide.
Comments (0)