Nvidia gains a route to preserve its dominance in data centers even as cloud companies and AI labs develop their own chips.
The company announced a $3.5 billion investment in Taiwan’s MediaTek that will see MediaTek adopt Nvidia technology to design custom chips tailored for AI customers and hyperscalers, chips that can be deployed alongside Nvidia hardware in existing data centers. As part of the arrangement MediaTek will gain access to Nvidia’s NVLink Fusion ecosystem, including NVLink, the interconnect technology that lets different chips communicate at high speed.
The deal lands amid a broader industry push by Amazon, Google, Microsoft, OpenAI and Anthropic to build proprietary silicon to lower reliance on third-party GPUs. Nvidia frames the partnership as a strategic way to concede some custom silicon work while keeping control of the rack-scale infrastructure that binds AI deployments together. “Nvidia is an AI infrastructure company,” Dion Harris, Nvidia’s senior director of HPC and AI hyperscaler infrastructure solutions, said on a call with reporters.
Harris argued the arrangement lets MediaTek offer customers chips that standardize on Nvidia’s rack-scale architecture, and that customers can deploy custom silicon “right alongside those using the same standard platform.” He added the move opens Nvidia’s ecosystem to MediaTek’s base, a point that underlines how Nvidia’s investments often circle back to benefit its own platform.
MediaTek has been expanding its custom data center ASIC efforts and told markets it expects that business to generate $2 billion in revenue in 2026. The company will continue collaborating with Nvidia on developer hardware like DGX Spark and on consumer-focused RTX Spark, and both firms said they will keep developing platforms for AI-powered, software-defined vehicles. MediaTek’s automotive work already relies on Nvidia RTX graphics for intelligent cockpits and pairs with Nvidia Drive AGX for autonomous workloads.
The announcement follows a recent Nvidia tie-up with Amazon Web Services, where AWS said it will deploy an additional 2 million Nvidia GPUs and integrate NVLink Fusion. Together these moves extend Nvidia’s influence by making custom chips interoperable with Nvidia’s data-center scaffolding, even when those chips are not made by Nvidia.
The immediate result is a clearer commercial path for companies that want bespoke silicon without rebuilding their infrastructure. For Nvidia the wager buys continued centrality in AI data centers; for MediaTek it accelerates a push into larger, cloud-scale customers. Expect more partnerships that stitch non-Nvidia ASICs into Nvidia-centric server racks as the industry balances bespoke designs with standardized infrastructure.
