Google Targets 15 Million TPUs by 2028

Google, which began developing custom AI processors a decade ago, aims to deploy between 12 and 15 million Tensor Processing Units (TPUs) by 2028. According to a Fubon Research note, this expansion centers on the upcoming TPU v9 generation, which will feature four compute dies. To meet these targets, Google's capacity consumption is expected to more than double between 2027 and 2028. While Nvidia is projected to supply 12.4 million data center GPUs in 2028, its current stock remains sold out. Fubon researchers indicate that Intel Foundry's supply is essential for Google's goals, with reports claiming Intel received orders for three million TPUs. Google has reportedly tested Intel's advanced packaging, though Intel's EMIB/EMIB-T technologies remain incompatible with TSMC's CoWoS-L.