Zuckerberg Unveils Meta's First Paid AI as Free Cash Flow Falls 91%

Meta Platforms posted Q2 2026 revenue of $60.8 billion, a 28% year-over-year increase, but free cash flow collapsed 91% to $784 million as quarterly capital expenditures hit $31 billion in AI infrastructure, according to the company's July 29 earnings release. On the same earnings call, Mark Zuckerberg defended the spend as the price of AI infrastructure leadership, even as total expenses climbed 55% to $42 billion and the company booked a $2.4 billion legal charge.

A $31 Billion Quarter for AI Compute

Meta's capital expenditures, including finance lease payments, reached $31.08 billion in Q2 2026, 83% above the $17.01 billion recorded in Q2 2025, according to the earnings press release. That single quarter consumed nearly all of Meta's operating cash flow, leaving $784 million in free cash flow compared with $8.55 billion a year earlier, CNBC reported.

By June 30, Meta's net property and equipment balance reached $225.72 billion, a $49.32 billion increase from the end of 2025. For the full year, the company narrowed its capex guidance to $130 billion to $145 billion, tightening the lower bound by $5 billion from its prior forecast of $125 billion to $145 billion. Zuckerberg told investors on the call: "AI is accelerating our core business today, powering our next generation of products, and opening the door to entirely new enterprise opportunities."

CNBC also reported that Zuckerberg addressed a tension specific to a company at this scale of AI capacity: deciding how much compute to sell to external developers versus reserving for Meta's own products. That question does not have a settled answer. The company is still making the calculation in real time.

The First API Business Meta Has Ever Run

Three weeks before the earnings call, Zuckerberg told Bloomberg in an exclusive interview: "Since this isn't an open-source model, this is really the first time we're seriously launching an API business." The model is Muse Spark 1.1, a closed proprietary frontier model built for agentic reasoning and tool use, Bloomberg reported on July 9.

The commercial strategy is built on aggressive pricing. Where rivals charge $5 to $10 per million input tokens and $30 to $50 per million output tokens, Meta priced Muse Spark 1.1 at roughly one-quarter of those rates, making it among the cheapest frontier APIs available to developers, according to 247 Wall St. on July 13. Developers get a free tier to a token threshold, then pay per usage.

Muse Spark 1.1 is separate from Meta's open-source Llama family, which remains freely available. Zuckerberg's pitch is that a proprietary, commercially priced model running at 75% below OpenAI and Anthropic can pull developers onto Meta's stack before they build a dependency elsewhere. The agentic shopping tools he previewed in January 2026, which would connect Meta's advertising catalog to AI-driven purchase agents, represent the most direct potential revenue payoff from that developer relationship, TechCrunch reported.

One Billion Users, One Narrow Revenue Line

Zuckerberg announced at Meta's May 2025 annual shareholder meeting that Meta AI had reached 1 billion monthly active users across its apps, TechCrunch reported. That milestone made Meta AI one of the fastest consumer AI products to reach that scale. Yet Meta's Q2 2026 revenue breakdown tells a sharply different story about monetization: advertising accounted for $59.36 billion of the $60.8 billion total, approximately 97.6% of all revenue. Non-advertising revenue, including the nascent Muse Spark API, was $1.44 billion.

The gap between user scale and commercial AI revenue is the defining tension in Zuckerberg's current position. StartupHub.ai data shows more than 150 companies currently building in the foundation model and AI API category, the large majority still pre-Series B, which reflects how early commercial AI API monetization remains across the industry. Meta's move to price Muse Spark 1.1 aggressively is partly a response to this window: the market for AI API customers is still forming, and the company with the lowest price and the largest developer surface area has an early structural advantage.

The Q2 total expenses figure of $42 billion, up 55% year-on-year, is the number that absorbed investor attention on July 29. Net income came in at $15.8 billion, with diluted earnings per share of $6.18, both below analyst consensus. Meta also booked a $2.4 billion charge related to legal proceedings during the quarter, per the earnings release.

What It Means

Zuckerberg's Q2 2026 framing is a calculated bet on infrastructure timing: spend at scale now, while the AI buildout is still early enough that acquiring compute capacity is competitively differentiating. The 91% free cash flow decline is the visible cost of that bet. The Muse Spark 1.1 pricing, positioned at 75% below rivals, is the matching commercial logic: establish a developer base and an agentic commerce pipeline while competitors are still charging premium rates. The risk embedded in both moves is the same: the cycle takes longer than projected, and the balance sheet absorbs the wait. Meta's advertising business, generating $59 billion per quarter, gives Zuckerberg considerably more room than most to absorb that risk.

Sources

• Meta Investor Relations: Meta Reports Second Quarter 2026 Results
• Bloomberg: Meta Starts Charging for AI With Muse Spark 1.1 Agentic Model
• CNBC: Meta's Stock Falls on Light Revenue Guidance, Dwindling Free Cash Flow
• 247 Wall St.: Zuckerberg's New AI Model Costs 75% Less Than Rivals
• TechCrunch: Meta AI Now Has 1 Billion Monthly Active Users
• TechCrunch: Zuckerberg Teases Agentic Commerce Tools and Major AI Rollout in 2026

Editorial standards: every claim is sourced. Tips: [email protected]