Nvidia is investing three point five billion dollars in Taiwanese chipmaker MediaTek, with the deal granting MediaTek access to Nvidia's NVLink Fusion ecosystem — enabling custom ASIC designs that plug directly into Nvidia-based data centre racks. MediaTek projects two billion dollars in custom data centre ASIC revenue for 2026, designing bespoke silicon for cloud providers and AI laboratories. A similar, non-investment partnership with AWS was announced the prior week, suggesting an emerging pattern rather than a one-off strategic decision.

The move is Nvidia's operational answer to the hyperscaler escape-from-Nvidia thesis that has circulated since major cloud providers announced competing GPU development programmes. Amazon's Trainium and Inferentia, Google's TPUs, Microsoft's Maia, and Meta's MTIA represent serious investments in the premise that proprietary silicon can reduce dependence on Nvidia's GPU supply chain and the pricing power that comes with it. Nvidia's response is not to defend that market share directly — it is to become structurally embedded in the custom silicon supply chain that those hyperscalers depend on.

Every ASIC MediaTek builds for a cloud provider runs on NVLink Fusion-compatible rack infrastructure. The hyperscaler escapes the GPU pricing relationship but not the interconnect relationship. Nvidia owns the fabric through which custom silicon communicates in the data centre — and now owns a significant stake in the company designing much of that custom silicon. The competitive dynamic shifts from GPU versus ASIC to whose interconnect fabric runs the data centre at scale. Nvidia is currently the only player with a deployed, validated answer to that question at the scale hyperscalers require.

At three point five billion dollars, this is among Nvidia's largest single strategic investments outside direct acquisitions. Its logic is consistent with Jensen Huang's stated strategy of owning the full system layer — GPU, CPU, networking, and now interconnect standards and the supply chain building to them — rather than competing on individual component performance as compute becomes increasingly commoditised. The hyperscalers built their escape plan. Nvidia purchased the exits.