Nvidia invests $3.5 billion in MediaTek to maintain grip on AI infrastructure

Nvidia is investing $3.5 billion in Taiwanese chip company MediaTek through the purchase of convertible bonds that can be converted into shares. It is Nvidia's largest direct investment outside the United States to date. The deal forms part of a broader bond offering by MediaTek worth $3.9 billion, in which Alphabet, Google's parent company, is also participating.

The timing is telling. Companies such as Amazon, Google, Microsoft and OpenAI are increasingly developing their own AI chips, with the aim of reducing their dependence on Nvidia's GPUs. By deepening its stake in MediaTek, Nvidia positions itself as an indispensable partner in the broader AI chip supply chain, extending beyond the GPU itself.

What MediaTek does for Nvidia

MediaTek and Nvidia are collaborating on AI platforms for data centres, PCs and automotive applications. The Taiwanese company competes with Broadcom and Marvell Technology in the market for custom chips used by data centre operators. That market is growing rapidly, as large tech companies seek their own silicon for specific AI workloads, separate from standard GPU clusters.

For Nvidia, the partnership with MediaTek provides a route into the custom systems market. Rather than selling only GPUs, Nvidia can co-develop the broader chip architecture that data centres require through MediaTek. Dion Harris, Senior Director at Nvidia, previously confirmed that the collaboration spans multiple product categories.

MediaTek had projected approximately $2 billion in AI chip revenue for this year and intends to accelerate that pace following the deal. Nvidia's investment gives the company both capital and a strategic partner with direct access to the world's largest buyers of AI infrastructure.

Nvidia builds a Taiwanese ecosystem

The investment in MediaTek does not stand alone. Nvidia is working with Foxconn and the Taiwanese government on the construction of Taiwan's first AI supercomputer. Researchers at chip manufacturer TSMC plan to use that system for R&D. In addition, Nvidia has formed partnerships with Acer and Asus for high-performance desktop supercomputers.

Amazon has previously agreed to deploy 2 million additional Nvidia components and to adopt Nvidia's interconnect technology, the connectivity standards that govern how chips communicate with one another inside a data centre. Whoever sets that standard holds a structural advantage in the architecture of future AI infrastructure.

Foxconn Chairman and CEO Young Liu is involved in a number of these Taiwanese initiatives. Jensen Huang, CEO of Nvidia, and Rick Tsai, CEO of MediaTek, confirmed the deal. The expansion of Nvidia's presence in Taiwan signals that the company views the island not merely as a manufacturing location through TSMC, but as a strategic partnership ecosystem.

Response to self-building tech giants

The backdrop to this deal is the shift under way at major tech companies. Google has its own TPU chips, Amazon is developing Trainium and Inferentia, Microsoft is working on Maia, and OpenAI has announced plans for its own silicon. All of these initiatives are partly intended to reduce dependence on Nvidia's H-series and Blackwell GPUs, which have been scarce and expensive in recent years.

Nvidia's response is not only to build better GPUs, but to become deeply embedded in the surrounding chip supply chain. By investing in MediaTek, setting interconnect standards and co-developing national AI infrastructure projects, the company raises the switching costs for anyone considering a move away from Nvidia technology. MediaTek earlier this year also concluded a deal with Google, demonstrating that the same partners are simultaneously committing to multiple parties.

For the European and Dutch AI infrastructure market, alliances of this kind serve as a relevant reference point. The Dutch government and investors thinking about their own computing infrastructure, AI data centres or chip independence can observe here how the global market is consolidating around a handful of platforms and standards. Those who wish to establish a position in time must account for the economies of scale and lock-in effects of multi-year ecosystem deals such as these.