Pichai reveals Google's computing power allocation priorities: DeepMind first, TPU direct sales for the first time in ten years

CryptoWorld News reports that Pichai revealed during Alphabet’s Q1 earnings call that Google’s compute capacity allocation priorities are as follows: DeepMind comes first, with the rest allocated among search, YouTube, and cloud services according to a return-on-investment framework. He admitted that the company’s compute capacity is constrained in the short term, and that if it could meet demand, cloud revenue could have been higher.

To ease supply pressure, Google will for the first time directly sell TPU hardware to some customers, so they can deploy it in their own data centers. Previously, over the past ten years, TPU was used only for internal purposes and rented out through the cloud, and had never been sold directly.

Target customers include the quantitative trading firm Citadel Securities, users of high-performance computing, and cutting-edge AI labs. CFO Anat Ashkenazi said that only a small portion of revenue from TPU hardware sales will be recognized this year, while the vast majority will be realized in 2027. These hardware agreements have already been included in cloud backlog orders, which have nearly doubled to $462 billion; however, the larger portion still consists of regular Google Cloud Platform contracts, which are expected to be recognized a little more than 50% within 24 months.

Another sign of growing demand is that Google’s self-developed model API token processing volume increased from 10 billion per minute in the previous quarter to 16 billion per minute, a 60% quarter-over-quarter increase.