As leading technology firms expand their computing infrastructure to handle rising artificial intelligence needs, they face a complex choice on how much capacity to keep internally versus offer externally.

Meta chief executive Mark Zuckerberg discussed the matter during the firm's second-quarter earnings discussion this week. Although Meta lacks a current operation to sell compute resources, the data center capacity required for AI workloads, to outside clients, Zuckerberg indicated plans are under consideration.

Zuckerberg noted that a substantial share of its resources will support training of Meta's AI systems, operation of agents, and expansion of its primary operations. Yet the company also anticipates developing a sizable offering for major clients, he added.

Technology leaders are moving quickly to increase compute availability amid strong demand for artificial intelligence capabilities. Google and Meta each modestly increased their capital expenditure outlook for the year, while Google indicated spending in 2027 could rise further. Microsoft, in an uncommon step, maintained its prior projections.

The increased outlays are now visible in results: Google's operating cash flow turned negative in the second quarter for the first time, and Meta's declined sharply year over year.

Providing compute to others represents one approach to easing cash pressures, though it carries a cost in forgone internal use. Executives at these firms commonly note that available capacity falls short of demand. The companies are striving to lead in AI development, and staff members at times vie for access to resources.

Microsoft encounters a similar issue. As one observer framed it recently, the question is whether to capitalize on present demand or prioritize longer-term objectives.

During the earnings discussion on Wednesday, Microsoft finance chief Amy Hood observed that customer interest continues to surpass available capacity in its cloud segment.

'Shortsighted' to pursue immediate gains

Zuckerberg revisited the issue later in the call. A frequent decision involves balancing monetization of assets in the present against building capabilities for future periods, he remarked.

It would be shortsighted to sell off all compute capacity for quick returns, he stated, noting that advances in model intelligence would increase the long-term worth of retained resources.

Retaining excessive capacity could also affect the cloud operations of large providers such as Google. Without sufficient resources for business clients, they may turn to competitors like Amazon or Microsoft, according to analyst Mark Shmulik.

Last week Google announced plans to acquire additional external compute to meet client needs while expanding its own internal capacity.

Google is also developing custom chips known as tensor processing units, or TPUs, and placing them in partner data centers to expand overall availability.

Our initial focus remains ensuring we allocate sufficient resources to lead in advanced AI development, chief executive Sundar Pichai stated during the earnings call, describing this as the basis for all other activities.

Although Google relies on its cloud segment for growth, it must also preserve enough capacity to safeguard its position in search, Shmulik noted.

The company further requires resources to support developers and maintain engagement with its platform, which again points to greater allocation of compute.