Nvidia pours $3.5B into MediaTek to widen its AI chip moat

Nvidia is putting $3.5 billion into MediaTek, the Taiwanese chipmaker best known for smartphone processors, in a deal announced August 31, 2026 that pulls another major silicon player deeper into Nvidia's orbit. The investment comes as Amazon, Google, Microsoft, OpenAI, and Anthropic all push to build their own custom AI chips, a trend that threatens to chip away at Nvidia's grip on data-center hardware.

Rather than compete head-on with those custom-silicon efforts, Nvidia is folding them into its own ecosystem. Under the agreement, MediaTek will use Nvidia's technology to help design custom chips for AI companies and hyperscalers, chips built specifically to plug into Nvidia-based data centers rather than stand apart from them.

Central to that integration is NVLink Fusion, Nvidia's interconnect standard that lets chips from different vendors talk to each other at high speed. MediaTek now gets access to it, meaning any custom silicon it helps design can slot into Nvidia infrastructure without the usual performance penalty of mixing hardware from multiple sources.

The MediaTek deal mirrors a partnership Nvidia struck with Amazon Web Services just a week earlier, in which AWS agreed to deploy two million additional Nvidia GPUs alongside NVLink Fusion integration. Together, the two deals suggest Nvidia's strategy for the custom-chip wave is less about blocking rivals and more about making sure every alternative chip still depends on Nvidia's plumbing.

Beyond data centers

The collaboration extends past hyperscale infrastructure. MediaTek and Nvidia are also working together on DGX Spark, a compact desktop AI computer aimed at developers, and expanding their partnership around RTX Spark chips for AI features in consumer PCs.

MediaTek expects its custom data-center chip business alone to bring in $2 billion in revenue during 2026, a figure that underscores how quickly demand for bespoke AI silicon has grown even for a company historically focused on mobile chips.

Nvidia executives used the announcement to tout performance gains from its upcoming Vera Rubin architecture, which pairs the Rubin GPU with a new Vera CPU built to handle data movement more efficiently. Jason Hardy of Nvidia said the Vera CPU delivers roughly threefold improvements in memory operations compared to prior designs, a detail the company is clearly using to argue that even as rivals build their own chips, none can yet match the throughput of staying inside Nvidia's stack.

Dion Harris, Nvidia's senior director of HPC and AI hyperscaler infrastructure solutions, was named among the executives discussing the deal, framing it as part of a broader push to keep Nvidia's architecture at the center of AI infrastructure decisions regardless of who manufactures the underlying chip.