The latest
Meta faces a widening gap between how quickly the bill for artificial intelligence is arriving and how slowly the returns are becoming visible. Free cash flow plunged 91% in the second quarter to $784 million, from $8.5 billion a year earlier. The results showed that a strong advertising business is no longer enough to reassure the market when data centers, chips and expensive research talent consume most of the available cash.
Zuckerberg argues that Meta is not building another digital assistant but a new layer of products and revenue. His vision centers on personal agents that operate around the clock, helping users with health, hobbies, finances and career goals. Meta also plans to offer coding and customer-service agents to businesses. Its broader ambition is a “business-in-a-box” service that helps users create and operate entire companies within Meta’s ecosystem.
The company is entering this phase from an unsettled technological position. Its Llama models fell behind OpenAI, Anthropic and Google, prompting Zuckerberg to restructure Meta’s AI operation and recruit researchers at enormous cost. New Muse Spark models have shown promise, but Meta has yet to demonstrate that those improvements can produce services consumers and companies will buy at a scale capable of justifying the spending.
Details
- Revenue guidance: Meta forecast current-quarter sales of $61 billion to $64 billion, with the midpoint below Wall Street’s expectation of $63.1 billion.
- Rising costs: Quarterly expenses climbed 55% to $42 billion, driven by infrastructure spending alongside legal charges and severance payments.
- Capital expenditure: Meta narrowed its 2026 spending forecast to between $130 billion and $145 billion after investing about $31 billion in the second quarter alone.
- Net income: Profit fell 14% to $15.8 billion, missing analysts’ estimate of $18.5 billion.
- Personal agents: Meta envisions autonomous software that performs continuous tasks and provides customized assistance across users’ daily lives.
- Enterprise products: The company plans to sell coding and customer-service agents while developing an integrated service for launching and managing businesses.
- Compute leasing: Meta is considering selling excess data-center capacity and is in early talks to rent computing resources to Anthropic in a deal potentially worth $10 billion.
- Monetization strategy: The company has begun testing subscriptions that offer advanced features or access to more complex AI tasks across its apps.
- Additional risks: Meta faces trials and regulatory scrutiny over youth safety that could ultimately produce substantial financial losses.
Between the lines
Zuckerberg is trying to transform data centers from an operating expense into a scarce strategic asset that Meta can use internally or rent to rivals. Yet that option also reveals that outside demand for computing capacity is currently clearer than the direct business model for Meta’s own AI agents.
What to watch
Investors will look for two forms of proof: models capable of competing with the market leaders and measurable revenue from agents and subscriptions. Until then, every increase in spending will test whether Meta’s advertising engine can finance the transformation without further weakening cash generation and profitability.
