Meta Raises AI Capex to Up to $145 Billion While Seeing 14% Decline in Quarterly Profit
Revenue grew 28% and surpassed consensus, but the EPS of $6.18 fell short of the projected $7.22. Susan Li increased the 2026 capex range to $130 billion to $145 billion.
Meta reported $60.8 billion in revenue for the second quarter, a 28% increase year-over-year, but delivered earnings per share of $6.18, roughly 14% below the $7.22 consensus from Refinitiv. The stock fell over 7% in after-hours trading on Wednesday (July 29), enough to reverse part of its July rally. The trigger was not revenue, but the new capex range.
Susan Li, Meta’s CFO, raised the guidance for fixed asset investment to $130 to $145 billion in 2026, up from the previous guidance of $125 to $145 billion. In the second quarter alone, capex, including financial leases, totaled $31.08 billion, nearly 50% of revenue. This figure distinguishes Meta from any comparable advertising platform in any previous decade.
Reality Labs No Longer Headlines
The Reality Labs division recorded a loss of $4.62 billion in the quarter, with revenue of $431 million. The loss is nearly identical to a year ago, and Li confirmed that the annual total is expected to be close to the $19 billion lost in 2025. This number is significant in absolute terms but comfortably fits within the company’s new AI capex. Analysts drew a direct comparison during the call: each quarter of Reality Labs equates to less than seven days of current spending on GPUs and data centers.
The Family of Apps continues to fund everything. This segment generated $60.4 billion in revenue and $23.4 billion in operating profit, with advertising revenue of $59.4 billion. Impressions rose 14% year-over-year, and the average price per ad increased by 12%. No other quarter for Meta has combined these two vectors with such a positive spread in double digits.
Where the Money is Going
Mark Zuckerberg reiterated the capex explanation based on the same logic for 2024 and 2025: capacity to internally train frontier models and serve inference at a population scale, both for Meta AI and agents that support feed, Reels, and monetization of conversations on WhatsApp. The new detail is the explicit mention of serving third-party open models within Meta's infrastructure, something the company had previously avoided confirming publicly.
The burden primarily falls on data centers in the United States, where Meta announced expansions at campuses in Richland Parish, Louisiana, and Kuna, Idaho. Outside the country, the focus is Spain, with the new site in Talavera de la Reina under construction, and the United Kingdom, where the company is negotiating contracted energy with Scottish grid operators. Each new gigawatt enters an electrical connection queue that already operates with an average wait time of three years in various North American regions.
How to Read in Light of Microsoft's Day
Microsoft reported on the same day that Azure crossed $100 billion in annual revenue and that FY26 capex reached $115.9 billion. Amy Hood guided the market to approximately $190 billion in the 2026 calendar year. Combining these two announcements, the market saw, on a single Wednesday, two companies confirming a combined total of over $320 billion in fixed investment, close to the annual GDP of Chile.
The structural difference is the revenue model. Each dollar of Microsoft’s capex appears in contracted backlog, with the $678 billion of RPO that the company displays at the end of the report. Each dollar of Meta’s capex still primarily translates into more capacity to serve ads and conversational agents. This is why the market tolerates the expansion of the hyperscaler and penalizes the advertising platform in the same movement, even though both operate identical data centers with the same energy dependencies and the same Nvidia silicon.
The Point Zuckerberg Did Not Answer
The CEO discussed the return on investment in AI using the example of Meta Advantage, the automatic campaign optimization platform that already accounts for a significant share of advertising revenue. What he avoided was presenting a direct metric of return on the incremental capex. Analysts from Bank of America and Morgan Stanley had already asked in recent reports about the incremental LTV per dollar of added GPU. The lack of an objective answer during the call that night is the most likely explanation for the post-market discount.