Google says AI servers pay back in under two years

Google Cloud CEO Thomas Kurian said the company’s AI servers have an aggregate payback period of less than two years, while Google’s own silicon pays back in roughly one year.

Kurian made the comments at the Goldman Sachs Communacopia + Technology Conference, according to reporting published by Yahoo Finance. He said Google’s accelerator business, including its Tensor Processing Units, is more than twice the size of the next-largest hyperscaler’s accelerator business.

Google also claims to offer 2.7 times better price performance for AI training and 80% better price performance for inference compared with relevant alternatives. The company’s use of proprietary chips allows it to design accelerators around its specific workloads.

Alphabet expects 2026 capital expenditure of US$195 billion to US$205 billion, with much of the spending directed towards AI infrastructure. The company reported second-quarter Google Cloud revenue of US$24.77 billion, up 82% year on year, while cloud backlog reached US$514 billion.

Kurian said most Google Cloud infrastructure contracts run for about five years and that customers committing to US$100 typically spend more than US$150 over time. The payback figures are company estimates; the supplied material does not provide calculation methodology or independent corroboration.