Nvidia invests $3.5 billion in MediaTek for custom AI chips
Nvidia is investing $3.5 billion in MediaTek to integrate its NVLink technology into custom chips, allowing the graphics giant to remain the core infrastructure for rival silicon.
Nvidia has committed $3.5 billion to Taiwanese chipmaker MediaTek in a strategic move to embed its proprietary technology into custom silicon. Under the agreement, MediaTek will integrate Nvidia's NVLink Fusion ecosystem, including its high-speed NVLink interconnect technology, into custom application-specific integrated circuits (ASICs). This integration allows third-party custom chips to communicate seamlessly within Nvidia-based data centers.
The partnership addresses a growing trend where major cloud providers and AI developers, including Amazon, Google, Microsoft, OpenAI, and Anthropic, are designing proprietary chips to reduce their reliance on Nvidia GPUs. By licensing its NVLink technology, Nvidia ensures its architecture remains the essential scaffolding for modern data centers. This follows a similar deal with Amazon Web Services, which will deploy an additional 2 million Nvidia GPUs and integrate NVLink Fusion. MediaTek expects its custom data center ASIC business to generate $2 billion in revenue by 2026.
Beyond data centers, the two companies are expanding their collaboration across multiple computing sectors. MediaTek will continue working with Nvidia on the DGX Spark, a compact desktop AI computer designed for developers, as well as the RTX Spark initiative to bring AI capabilities to consumer PCs. In the automotive sector, MediaTek's vehicle platforms will combine Nvidia RTX graphics with the Nvidia Drive AGX in-car computing platform to power software-defined autonomous vehicles.
For AI practitioners and system architects, this deal simplifies the deployment of heterogeneous hardware. Engineers can now design highly specialized, workload-specific ASICs through MediaTek while retaining the ability to scale them directly alongside standard Nvidia GPU clusters. This standardization reduces integration friction, allowing developers to run custom silicon on the same unified rack-scale architecture without sacrificing the high-speed communication benefits of Nvidia's hardware ecosystem.
This is our own summary of reporting by TechCrunch AI
