MediaTek, the Taiwanese chip designer, has hired Douglas Yu, a former senior executive at TSMC who helped develop the advanced packaging technology at the heart of every major AI accelerator, as a part-time adviser to guide its push into the artificial intelligence chip market.
Yu joined TSMC in 1994 and retired last year after more than three decades spanning backend research and development, during which he played a central role in creating CoWoS (Chip on Wafer on Substrate), the advanced packaging process that binds multiple chiplets and high-bandwidth memory into a single module.
CoWoS is the packaging technology used in Nvidia's data centre GPUs, Google's TPUs, AMD's Instinct accelerators and a growing number of custom AI chips designed by hyperscalers, and TSMC's capacity for the process has been one of the tightest bottlenecks in the global AI supply chain for more than two years.
"We look forward to leveraging his extensive industry experience and technical expertise to support the company's exploration and roadmap planning for future advanced packaging technologies," MediaTek said.
The appointment signals how seriously MediaTek is pursuing the AI accelerator market, where it has historically had no presence.
Last week, the company said it expects to generate multiple billions of dollars in annual revenue from custom AI accelerator ASICs by 2027, a business line that did not exist 18 months ago.
The timing aligns with MediaTek's reported partnership with OpenAI and Qualcomm to develop custom smartphone processors for a new AI-first device, as well as Qualcomm's separate confirmation that it is entering the custom hyperscale silicon market with bespoke data centre chips.
Advanced packaging has become a strategic chokepoint because modern AI chips are too large and complex to manufacture as a single piece of silicon, requiring multiple smaller chiplets to be stitched together with high-bandwidth interconnects on a single substrate.
Related reading
- OpenAI deploys GPT-5.5 across Nvidia on Blackwell infrastructure in enterprise-scale AI agent rollout
- Google in talks with Marvell to develop two new AI chips as it builds multi-supplier TPU strategy
- Nvidia launches open-source quantum AI models as it deepens grip on chip design workflows
The process determines how much memory can sit alongside the processor, how quickly data moves between components, and ultimately how efficiently the finished chip performs inference and training workloads.
Yu's knowledge of TSMC's packaging roadmap, capacity planning and process capabilities gives MediaTek an unusual degree of insight into the manufacturing partner it will depend on to produce its AI accelerators.
The recap
- MediaTek appoints Douglas Yu as a part-time adviser.
- Yu worked at TSMC from 1994 until 2025.
- MediaTek expects multiple billions of dollars revenue from AI ASICs by 2027.
