Key Takeaways
• Google is considering direct wafer orders from TSMC for its next AI chip
• The move could reduce reliance on MediaTek in the TPU supply chain
• The upcoming TPU codenamed Humufish is targeted for 2027
• Cost optimization is becoming critical as AI infrastructure scales
• Advanced packaging competition between Intel and TSMC is intensifying
Google is reportedly evaluating a significant shift in its artificial intelligence chip strategy as it looks for ways to reduce costs and scale its infrastructure more efficiently.
According to analyst Ming-Chi Kuo, Google has recently approached TSMC to assess potential savings from placing direct wafer orders for its next generation TPU chip. This would bypass MediaTek, which currently plays an intermediary role in the company’s supply chain.
The chip at the center of this discussion is codenamed Humufish, part of Google’s upcoming eighth generation TPU lineup. Scheduled for the second half of 2027, the chip represents a key step in Google’s effort to compete with dominant players like Nvidia in the AI hardware space.
MediaTek has only recently become a major partner in Google’s TPU ecosystem, contributing to inference focused designs alongside Broadcom, which handles training chips. A direct relationship with TSMC would mark a departure from this collaborative model and could reshape the balance of partnerships.
The motivation behind the move is clear. As AI demand continues to surge, the cost of building and deploying chips at scale is becoming a critical factor. Analysts estimate that Google’s current dual sourcing strategy has already reduced costs significantly, but further savings could be achieved by eliminating layers in the supply chain.
At the same time, the Humufish chip is also part of a broader competition in advanced chip packaging technologies. Intel is promoting its EMIB-T packaging approach, which has reached validation yields of around 90 percent. While this represents strong progress, achieving the industry standard of near 98 percent for mass production remains a challenge.
TSMC, on the other hand, continues to develop its own advanced packaging solutions, including CoWoS, which targets similar high yield levels. The competition between these approaches will play a key role in determining the final design and manufacturing strategy for future AI chips.
Google’s evolving chip roadmap reflects a broader push to strengthen its position in AI infrastructure. The company recently introduced its seventh generation TPU and outlined plans for its next generation systems, while also beginning to offer its custom chips directly to enterprise customers.
This strategy puts Google in more direct competition with Nvidia, which has long dominated the AI hardware market. By optimizing both performance and cost, Google aims to expand its reach in cloud computing and AI services.
However, shifting toward direct manufacturing relationships carries risks. It could strain partnerships with companies like MediaTek, which have only recently become integral to Google’s chip ecosystem.
As the race for AI dominance intensifies, decisions around cost, supply chain control, and manufacturing partnerships are becoming just as important as technological innovation itself. Google’s next moves could have a lasting impact on how the AI hardware industry is structured in the years ahead.
Your First 10 AI Skills
10 practical AI skills, copy-paste prompts and a 7-day plan to start using AI with confidence.
Download the guide →