Taiwanese media outlet Anue Juheng cited a market analysis report from an unknown source on the 1st of this month, stating that Google plans to deploy 12~15 million of its own TPU v9 AI ASIC chips in 2028.
Given analysts' estimate that NVIDIA's AI GPU shipments could reach 12.4 million units in 2028, this means Google's server AI accelerator shipments may catch up with or even surpass NVIDIA in the year after next.
Regarding chip architecture design and process technology, the report suggests that TPU v9 will adopt a 4-compute-die structure, significantly increasing demand for advanced process and packaging capacity. This will force Google to seek capacity outside TSMC, creating new business opportunities for Intel Foundry and Samsung Foundry.