Google's TPU production is expected to surpass NVIDIA by 2028, with Intel likely to be key in the expansion

According to the latest estimates from Fubon Research analysts, Google is significantly ramping up production for the upcoming ninth-generation TPU (TPU v9) set to launch in 2028. The target deployment volume is expected to be between 12 million and 15 million units, a scale that not only aims to match but may even surpass Nvidia's total shipment volume of AI GPUs for data centers in previous years. This forecast indicates that Google's self-developed AI chip strategy has transitioned from early testing to large-scale commercial deployment, and it is highly likely to reshape the global AI hardware supply chain landscape.

Analysts point out that Google's TPU v9 will adopt a design with four compute dies, meaning that the production capacity consumption in 2028 will show exponential growth compared to 2027. In contrast, analysts estimate that Nvidia's data center AI GPU shipments will be around 8.2 million units in 2026, increasing to 12.4 million units in 2028. If Google's deployment plan comes to fruition, this would represent a demand for AI accelerators from a single company that approaches or even exceeds the overall supply scale of the market leader.

Fubon- “the production push out in Rubin will bring some negative impact to the supply chain as Rubin has higher dollar content compared to Blackwell"$MU: “recent checks suggest that $NVDA HBM4 cost will increase to US$31-32 per GB from US$17-18 per GB (HBM 3e) in 2026."

“We… pic.twitter.com/wLcWwcqcsH

— Sean (@sean_________) July 28, 2026

However, achieving such a massive output may be difficult solely relying on TSMC's production capacity. Analysts believe that for Google to successfully reach the target of 15 million TPUs in 2028, support from Intel's foundry services will be crucial. There have also been rumors in the market that Intel has successfully secured over 3 million TPU foundry orders after Google tested its advanced packaging technology. Considering that Intel's EMIB/EMIB-T technology is incompatible with TSMC's CoWoS-L packaging architecture, if Google decides to shift to Intel for foundry services, the related chip design and packaging plans will inevitably need to be adjusted simultaneously Analysts believe that if the above forecast data comes true, Google will become one of the largest consumers of AI accelerators in the world and will possess the largest proprietary AI hardware fleet. In the context of rapidly expanding AI demand, this development may not weaken Nvidia's market dominance, but it clearly highlights another industry trend: large cloud providers are accelerating their move towards "vertical integration"—from designing chips in-house to optimizing for their own software stack and data center needs, thereby constructing a more complete AI infrastructure.

• Google could build more AI accelerators than Nvidia sells in 2028, analyst claims — could push the company to use Intel Foundry to meet its goals

(Featured image is Google TPU, source: Google)

Further Reading:

• Google rumored to be developing a new AI chip to improve Gemini's computing efficiency
• Is it wafer foundry or advanced packaging? Controversy arises over Google's order of 3 million chips from Intel
• With self-developed TPU and powerful models, Google controls about a quarter of global AI computing power
• Tech giants are increasingly developing their own chips! Microsoft, Google, and Meta all join the fray to challenge Nvidia's AI dominance