📰 Key Summary

Nvidia announced a $3.5 billion investment in Taiwanese chipmaker MediaTek. As part of the deal, MediaTek will use Nvidia’s technology to help design custom chips for AI companies and hyperscalers—chips that plug directly into Nvidia-centric data center architecture. This comes against a backdrop where more AI companies and cloud providers, including Amazon, Google, Microsoft, OpenAI, and Anthropic, are pouring resources into building their own chips to cut their dependence on Nvidia GPUs. Nvidia executive Dion Harris told reporters on a Monday call that the company shifted from being a pure compute-chip maker to an “AI infrastructure company” years ago. The deal gives MediaTek access to Nvidia’s NVLink Fusion ecosystem, where NVLink tech lets chips from different vendors—even non-Nvidia chips—connect to each other at high speed. Just last week, Nvidia announced a similar partnership with AWS (without a direct investment), where AWS will deploy an additional 2 million Nvidia GPUs across its infrastructure and integrate NVLink Fusion. Through this partnership, MediaTek can keep building custom chips for cloud providers and AI labs while leveraging Nvidia’s “rack-scale architecture” tech stack, letting customers’ custom chips sit side-by-side with other systems built on the same standard platform. MediaTek said back in June that its custom data center ASIC business is expected to generate $2 billion in revenue in 2026, with plans to grow that share further. MediaTek will also keep collaborating with Nvidia on DGX Spark, the compact desktop AI computer built for developers.

💬 JudyAI Lab Take

This isn’t Nvidia just selling GPUs—it’s putting $3.5 billion directly into MediaTek in exchange for MediaTek using Nvidia’s tech to design custom chips for AI companies and cloud providers, chips that plug straight into Nvidia-centric data center architecture.

Given that Amazon, Google, Microsoft, OpenAI, and Anthropic are all building their own chips, cutting reliance on Nvidia GPUs is clearly an industry-wide trend. But Nvidia isn’t playing defense here—it’s positioning itself as an “AI infrastructure company.” By using NVLink Fusion to pull non-Nvidia chips into its own rack-scale architecture, Nvidia makes sure that even customers trying to route around its GPUs stay inside its ecosystem. Add in last week’s deal with AWS for 2 million GPUs, and the pattern is worth watching: competition at the infrastructure layer stopped being just about selling hardware a while ago.

For AI builders, it’s worth keeping an eye on cross-vendor interconnect standards like NVLink Fusion—they could end up shaping architecture decisions the next time you’re picking a cloud or chip vendor.

📅 Original Source Info

🔗 Further Reading