Technology giants such as Meta and Google face challenges in allocating AI computing power between internal use and external sales.

According to Odaily, Meta CEO Mark Zuckerberg stated during the company's second quarter earnings call that Meta currently does not offer computing power sales to clients, but such services are planned. He indicated that a large portion of Meta's computing power will be used for training AI models, supporting agent products, and developing its core business, and the company also anticipates expanding services for major clients.

A certain exchange and Meta have both slightly increased their capital expenditure forecasts for this year, with the exchange stating that related spending may rise further in 2027, while Microsoft has maintained its capital expenditure outlook. The exchange reported its first negative cash flow in the second quarter, and Meta's cash flow dropped by 91% compared to the same period last year.

Microsoft CFO Amy Hood stated that demand from its cloud business clients still exceeds available capacity. The exchange said last week that, while building more internal computing power, it will also purchase additional third-party computing power to meet client demand.

The exchange CEO Sundar Pichai stated that the primary task for the exchange's self-developed Tensor Processing Units (TPUs) is to ensure the allocation of necessary resources for frontier AGI development. The exchange is working with partners to deploy TPUs to other data centers in order to free up more capacity.