• Nvidia invested $3.5bn in convertible bonds issued by MediaTek as the companies expanded their collaboration across AI infrastructure, local computing and automotive
  • MediaTek will use NVLink Fusion as a design foundation for custom XPUs that can connect to Nvidia's rack-scale AI systems

The fact

Nvidia and MediaTek said on 31 August that they had expanded their collaboration across cloud AI infrastructure, local computing and automotive. Nvidia also invested $3.5bn in convertible bonds issued by MediaTek.

For data-centre infrastructure, MediaTek will use Nvidia's NVLink Fusion platform as a design foundation for customers developing custom XPUs. Nvidia had already named MediaTek among the platform's early adopters when NVLink Fusion was introduced earlier in 2026.

The latest announcement provides more detail on MediaTek's role, including work around connectivity, memory, packaging, manufacturing and integration with Nvidia rack-scale systems. Nvidia describes the platform as a prebuilt and prequalified foundation intended to reduce the engineering required around a custom accelerator.

The assessment

MediaTek's role gives customers a defined route from a custom XPU design into an Nvidia-compatible rack-scale system. The compute element can remain customer-specific, while MediaTek and Nvidia provide more of the surrounding work needed to connect the chip to memory, high-speed interconnects, packaging and the scale-up fabric.

Nvidia says this approach can reduce development complexity and shorten the path to production. That has not yet been demonstrated through a named customer deployment. The companies have not disclosed a customer, XPU tape-out, qualification result or timetable showing how quickly a MediaTek-designed accelerator can move from silicon design into an operating rack.

For BTW readers, custom silicon in this model does not mean designing every part of the AI system independently. Customers can differentiate the accelerator while using Nvidia-compatible infrastructure around it. The first production XPU and rack deployment will show whether that division of work can deliver the integration benefits Nvidia is promising.

What to watch

Watch for the first named customer using MediaTek and NVLink Fusion for a custom XPU, followed by tape-out, packaging, system qualification and rack deployment. A disclosed development timetable or production deployment would provide the first measurable evidence that the design framework reduces the work needed to move a custom accelerator into an operating AI system.