The AI investment boom is forcing America's leading technology companies to resemble capital-intensive utilities. Meta Platforms saw its shares decline as much as 10% in after-hours trading on Wednesday after the company reported earnings that fell short of expectations, driven by a 55% surge in costs. The stock later pared losses and was down approximately 7%.

The sell-off underscores a sharp reversal from Meta's 2023 "Year of Efficiency," when Zuckerberg cut roughly 21,000 jobs, flattened management layers, and watched the stock more than triple. Investors who rewarded that austerity are now confronting a company deploying capital at a pace that consumes nearly every dollar of operating cash flow.

Meta's financial results for the second quarter reflected mounting pressure from AI-related spending. Operating income fell 8%, net income declined 14%, and free cash flow dwindled to just $784 million—narrowly avoiding negative territory and far below the approximately $12 billion average the company maintained over the prior eight quarters.

Revenue for Q2 rose 28% year over year, exceeding Wall Street expectations. However, operating income for the Family of Apps segment—which encompasses Facebook, Instagram, WhatsApp, and Messenger—dropped to $23.4 billion from $25.0 billion. In other words, Meta's core business generated more revenue while earning less profit.

The primary driver behind the squeeze was capital expenditure. Meta spent $31.1 billion on capex during the quarter, nearly double the amount from a year earlier. With operating cash flow of $31.9 billion, the company directed almost every dollar generated by its operations into AI infrastructure, including servers, data centers, networking equipment, and semiconductors.

Historically, Meta has invested in data centers to support its global social networking platforms. However, the enormous computing power required to train and deploy AI models has dramatically accelerated spending, disrupting a financial model in which the company's lucrative advertising business enabled substantial cash accumulation. Like its hyperscaler peers, Meta must now construct multibillion-dollar data centers at a continuous pace—acquiring land, securing power sources, purchasing chips, operating cooling systems, and replacing hardware that becomes obsolete within years. The electricity demands have grown so intense that data center power consumption has become a concern for grid operators nationwide, with utilities across the country scrambling to secure enough energy to meet the needs of the AI buildout. Depreciation and amortization in the second quarter rose 46% year over year to $6.4 billion.

Competitors Microsoft, Amazon, and Google have established large-scale cloud businesses that rent infrastructure to external customers, allowing them to generate immediate revenue from their capital investments. Microsoft's shares rose nearly 2% in after-hours trading following its own cloud growth results, even as Meta CEO Mark Zuckerberg faced analyst questions about whether his company would pursue a similar strategy.

Zuckerberg acknowledged the potential for additional revenue by leasing Meta's computing infrastructure to other organizations and confirmed that the company intends to enter the cloud business, saying an update would come soon.

"We're getting a lot of offers for compute at a significant premium for what we paid for it," Zuckerberg said.

However, he characterized the opportunity as secondary rather than central to Meta's strategy, emphasizing that the greater value lies in building AI services on top of its own infrastructure. Unlike competitors such as Google and OpenAI, which keep their most advanced models proprietary, Meta has released its Llama family of large language models as open source, a bet that broad developer adoption will strengthen its ecosystem.

"It would be foolish to basically just sell all of the compute and take a short-term profit," Zuckerberg stated, adding that Meta expects "a significantly higher margin on selling intelligence rather than selling compute directly."

The intelligence Zuckerberg referenced encompasses a full stack of planned businesses: an advertising system that AI has improved by 15.7% in conversion performance; AI agents capable of handling customer messages for up to a million businesses; an API providing access to Meta's AI models; and a personal assistant designed to operate around the clock, building profiles of users' health, finances, and relationships—a product that does not yet exist.

"There's only so much that I can say on an earnings call about this," Zuckerberg remarked.

Meta has also shifted its financing approach. During the quarter, the company issued $24.9 billion in long-term debt and conducted no share repurchases, a sharp contrast from the more than $10 billion in buybacks during the same period a year earlier. CFO Susan Li stated that Meta had been intentionally moving toward "a greater mix of debt" to fund infrastructure projects with long operational lifespans and anticipated passive income.

Meta raised the floor of its full-year capital expenditure guidance to a range of $130 billion to $145 billion. The company spent $50.9 billion in the first half of the year, implying quarterly outlays of $39 billion to $47 billion for the remainder of 2026, against operating cash flow of roughly $32 billion. This trajectory indicates that the second quarter may be Meta's last positive free cash flow quarter of the year.

When asked about projected costs for 2027, Li declined to provide a figure, stating instead that Meta expects to remain demand-constrained—with more profitable applications for its computing power than available capacity.

Zuckerberg struck a confident tone: "My personal bet is that the people who invest in this are going to be rewarded and feel very good over time."

This story was originally featured on Fortune.com.