Meta stock drops 10% as AI capex crushes cash flow

Meta's second-quarter earnings delivered a paradox: revenue jumped 28% year-over-year to beat expectations, yet the company's core advertising business—Family of Apps, which includes Facebook, Instagram, WhatsApp, and Messenger—saw operating income decline from $25.0 billion（約4兆円） to $23.4 billion（約3.7兆円）. The culprit was a tripling of AI infrastructure spending.

Capital expenditure surged to $31.1 billion（約5兆円） in the quarter, nearly double the $15.8 billion（約2.5兆円） spent a year earlier. Operating cash flow came to $31.9 billion（約5.1兆円）, meaning Meta spent almost every dollar its businesses generated on AI infrastructure: servers, data centers, network infrastructure, and chips. Depreciation and amortization rose 46% year-over-year to $6.4 billion（約1兆円）, reflecting the rapid obsolescence of computing hardware. The company has shifted its financing strategy in response: it issued $24.9 billion（約4兆円） in long-term debt during Q2 while buying back no stock, reversing its pattern of $10 billion（約1.6兆円）-plus annual buybacks.

When pressed by analysts on why Meta was not offsetting these costs through a cloud business like its rivals, Zuckerberg acknowledged the opportunity. "We're getting a lot of offers for compute at a significant premium for what we paid for it," he said, and confirmed Meta has plans to enter the cloud business, promising details soon. However, he framed the cloud opportunity as secondary. "It would be foolish to basically just sell all of the compute and take a short-term profit," Zuckerberg said. Instead, he emphasized that Meta expects "a significantly higher margin on selling intelligence rather than selling compute directly."

That "intelligence" refers to a full stack of AI products Meta is developing: an ad system that AI has made 15.7% better at converting; agents capable of answering customer messages for a million businesses; an API selling access to Meta's AI models; and a personal assistant that works 24/7 and builds a profile of a user's health, finances, and relationships—though Zuckerberg noted this assistant does not yet exist and he could not share more detail on an earnings call.

Meta's capital spending will accelerate further. The company now projects full-year 2024 capex of $130 billion（約21兆円） to $145 billion（約23兆円）, raised from prior guidance, after spending $50.9 billion（約8.1兆円） in the first half. That leaves $39–47 billion per quarter for the rest of the year, compared to operating cash flow of roughly $32 billion（約5.1兆円） per quarter. CFO Susan Li confirmed this quarter was the last positive free cash flow quarter in 2024. When asked about 2027 spending, Li declined to provide a number, instead noting that Meta expects to remain "demand constrained"—meaning it has more profitable uses for computing power than capacity available. Zuckerberg echoed the long-term bet: "My personal bet is that the people who invest in this are going to be rewarded and feel very good over time."

Meta's earnings reveal a fundamental shift in how the company finances itself and deploys capital. For years, Meta's advertising business generated enormous cash reserves, allowing it to buy back stock and return cash to investors. The explosion in AI infrastructure spending has inverted that model: the company is now issuing long-term debt (Meta raised $24.9 billion（約4兆円） in Q2 debt versus zero share buybacks, a reversal from the $10 billion（約1.6兆円）-plus annual buybacks of the prior year) to fund data centers with multi-year lifespans and no immediate revenue.

Zuckerberg's framing of the cloud business as secondary is instructive. While Microsoft, Amazon, and Google have built enormous cloud divisions that generate immediate revenue and offset their infrastructure costs, Zuckerberg is betting that Meta's real upside lies in proprietary AI services—better ad targeting, customer service agents, and a personal assistant still in development. His statement that the company expects "a significantly higher margin on selling intelligence rather than selling compute directly" suggests he views renting infrastructure as a low-margin distraction from higher-value AI product development. CFO Susan Li's refusal to project 2027 capex costs and her statement that Meta expects to remain "demand constrained" (meaning more profitable uses for computing power than available capacity) signals the company believes this spending wave will persist and eventually become profitable—but only through proprietary AI services, not cloud rental.

No comments yet. Be the first to share your thoughts!