Meta Platforms: Free Cash Flow Collapses to $784 Million as Capex Hits $31 Billion, Even as New AI Ad Engine Fuels Fastest-Growing Ad Business in Big Tech
Q2 2026 earnings call, July 29, 2026
Meta Platforms delivered another quarter of accelerating top-line growth alongside a much more strained cash profile, underscoring just how aggressively the company is now spending to defend its position in the AI race. Total revenue reached $60.8 billion, up 28% year-over-year, and advertising revenue of $59.4 billion grew 27%, a pace CEO Mark Zuckerberg said makes Meta's ad business the fastest-growing of any major company on a dollar basis. But free cash flow for the quarter was just $784 million, a figure that stands out sharply against $31.1 billion in quarterly capital expenditures and signals that the infrastructure build-out is now consuming nearly all of the cash the business generates.
Capex Guidance Narrows Higher, Expense Outlook Absorbs Legal Charges
CFO Susan Li raised the low end of Meta's full-year 2026 capital expenditure guidance to a range of $130 billion to $145 billion, up from $125 billion to $145 billion previously, while full-year total expense guidance moved to $165 billion to $169 billion to absorb a $2.4 billion legal charge taken in the quarter. GAAP operating income actually fell 8% year-over-year to $18.8 billion, a 31% margin, though Li noted that excluding the legal charge and $1.2 billion in severance tied to a May 2026 headcount reduction of roughly 8,000 employees, operating income would have risen 9%. The tax rate guidance for the remainder of 2026 was also pushed higher, to 15%-17% from 13%-16%. Management declined to give any 2027 capex figure, with Li stressing that "infrastructure planning remains highly dynamic" and that the company is deliberately preserving optionality rather than locking in long-duration server commitments before demand curves are clearer.
A New Ad Ranking Architecture: Meta Generative Recommender
The most concrete technical disclosure of the call centered on a rearchitected advertising system called Meta Generative Recommender, described by Li as "a paradigm shift in how our ad system works." Rather than scoring each candidate ad individually, the system uses large language models to jointly reason about ad content and user preferences to predict the single best ad for a given person. Meta has already deployed the first generative model into its ads retrieval stack, and early pilots using LLMs to better understand user preferences drove a 1% increase in app event conversions on Instagram. Combined with the company's GEM ranking and sequence-learning model, the changes generated an 8.3% increase in ad clicks and a 15.7% uplift in conversions on Facebook — among the more concrete quantifications of AI-driven monetization gains Meta has disclosed. Advantage+, the company's automated end-to-end ad product, crossed $75 billion in annualized revenue run rate this quarter.
Compute Monetization: Premium Offers on the Table, But Meta Is Holding Back
Zuckerberg confirmed that Meta is fielding a meaningful number of offers to sell its compute capacity directly to third parties "at a significant premium over what we paid for it," but the company is choosing not to pursue this aggressively for now. "We believe that there will continue to be a significantly higher margin on selling intelligence rather than selling compute directly," he said, framing direct compute sales as one component of a broader enterprise strategy that also includes APIs, business agents and productivity tools, rather than the primary lever. This is a notable data point for investors trying to size the option value embedded in Meta's data center buildout beyond its core advertising business.
BlackRock Partnership Signals a Shift in How Meta Funds Its Buildout
Meta announced a new strategic venture with BlackRock to build a 1-gigawatt data center in El Paso, Texas, which Li characterized as a template for how the company will increasingly finance capacity. She noted Meta has been "evolving our capital structure in recent years to include a greater mix of debt" to lower its cost of capital and said the company will keep "broadening our aperture" to include structured partnerships alongside internally generated cash flow — a signal that off-balance-sheet or joint-venture financing structures are likely to become a larger part of Meta's capex funding mix as the scale of investment grows.
Personal Agents Still Not Shipped, But Positioned as the Next Major Bet
Zuckerberg spent considerable time previewing a personal-agent product that has not yet launched, calling consumer agents "almost inevitable" to become a massive market but acknowledging Meta has yet to deliver on it. "I think that it's extremely unlikely if you look out five years from now that you don't have billions of people with a personal agent that understands your goals and is working on your behalf 24/7," he said, while conceding that "there's kind of only so much that I can say on an earnings call about this" since the product has not shipped. He drew a distinction between the current coding-agent boom, driven by technical users willing to tolerate friction, and the harder challenge of building agents that "just work" for billions of mainstream consumers — an area where he argued Meta's distribution and scaling capabilities give it a structural edge over rivals.
Business Agents Show Early Commercial Traction
More tangible progress came from business agents, now live globally on WhatsApp and Messenger and rolling out on Instagram, with over 1 million businesses using them weekly to interact with customers. Management highlighted Movida, a Brazilian car rental company, which saw a 44% increase in daily WhatsApp bookings after deploying a booking agent, with 85% of conversations resolved without human intervention. Zuckerberg described a long-term ambition to build this into a "business-in-a-box" offering monetized similarly to the ad auction system, where businesses pay based on results rather than fixed subscriptions over time.
Model Strategy: Open Source Return Coming, but Frontier Ambitions Remain Non-Negotiable
Asked whether Meta could rely on open-weight models from other labs rather than continuing to fund its own frontier development, Zuckerberg was direct: "Right now, the open source models are not as strong as the frontier models. So no is the basic answer." He confirmed Meta plans to return to releasing open-source models "at some point soon" after a gap, but framed proprietary frontier development as core to Meta's identity as a "full-stack technology company" that builds its own chips, data centers and models, arguing this vertical control is what will differentiate its personal and business agent products from competitors using more generic model stacks.
Legal Overhang Remains a Live Risk
Li flagged that Meta continues to face regulatory scrutiny on youth-safety issues across several markets, with "a number of youth-related trials scheduled for this year in the U.S., which may ultimately result in a material loss" — a reminder that the $2.4 billion legal charge taken this quarter may not be a one-time event, and that litigation risk remains an underappreciated variable in the earnings model going forward.