Google aims to produce up to 15 million TPUs by 2028

Google plans to produce between 12 and 15 million ninth generation TPU accelerators by 2028, a scale that would rival Nvidia’s projected data center GPU shipments. According to Fubon’s research note, this would mark a major shift in the AI hardware landscape, with a single cloud provider manufacturing accelerator volumes comparable to the industry’s leading merchant supplier. The TPU v9 uses a multi die chiplet design requiring advanced packaging, and Google may rely on both TSMC and Intel Foundry to meet production demand. Intel has reportedly secured orders after Google tested its EMIB based packaging capabilities. If Google reaches its target, it could become the largest single user of AI accelerators, gaining more control over its compute infrastructure while still purchasing Nvidia hardware. The move signals that competition in AI hardware is increasingly driven by deployment scale rather than raw chip performance.

Read the full story on TechSpot →