Google Sprints Ahead With TPU Rollout, Doubles Release Pace

Google has significantly accelerated its rollout of Tensor Processing Units (TPUs), increasing the pace from one every two years to two per year. According to Google's chief technologist, Amin Vahdat, this rapid cadence is necessary to meet growing demand for TPUs in AI applications.

Vahdat noted that the company has moved from releasing a TPU every two years to one per year and now two per year. He expressed confidence that the rate of release will continue to increase, with manufacturing lines operating at full speed and testing capabilities fully available.

Google is also undergoing a 'radical change' in how it works with the supply chain, addressing bottlenecks in memory, PCBs, high-voltage power supplies, liquid cooling, AI racks, and other areas. Vahdat warned that managing these constraints requires constant vigilance, as new challenges can emerge at any time.

In addition to its internal use of TPUs, Google is also selling ICs (Integrated Circuits) to external customers, with plans to expand this business further. The company's parent Alphabet has raised its capital spending guidance for 2026 to between $195 billion and $205 billion, partly to build next-generation AI data centers.