The Numbers
Meta’s Q2 2026 results:
- Revenue: $60.8 billion (up ~22% year-over-year)
- Free cash flow: Down 91% year-over-year
- Capital expenditure: Record levels driven by AI infrastructure
- Stock reaction: Fell ~10% after earnings
The revenue growth was solid, but investors focused on the cash burn: Meta spent nearly all its free cash flow on AI infrastructure in a single quarter.
Muse Spark 1.1
Alongside earnings, Meta launched:
- Muse Spark 1.1 API: Proprietary, metered API — up to 83% cheaper than Claude
- Strategic shift: From open-source Llama to proprietary releases
- Zuckerberg’s pivot: The company that championed open models is now selling closed ones
The shift drew criticism: Zuckerberg attacked OpenAI and Anthropic’s closed models even as Meta itself moved to closed releases.
Why the Cash Burn
Meta’s spending drivers:
- AI data centers: Massive compute buildout for training and inference
- GPU purchases: Nvidia and AMD accelerators at scale
- Energy infrastructure: Power for AI workloads
- Muse Spark development: Proprietary models require frontier-scale investment
The company’s $14 billion El Paso data center partnership with BlackRock is one example of the scale of spending.
Industry Context
Meta’s cash burn mirrors the industry-wide pattern:
- Google: $205 billion AI spending projections through 2028
- Microsoft: $80 billion capex in FY2027
- Amazon: $100 billion committed
- Anthropic: $10 billion Volta deal plus $5 billion from AMD
The AI infrastructure race is consuming free cash flow across the industry. The bet: massive compute investment now will pay off in AI-driven revenue later.
Implications
- Investor patience: Wall Street is questioning how long cash burn can continue
- Pricing power: Meta’s cheaper Muse Spark pressures competitor pricing
- Strategic tension: Open vs. closed strategy remains unresolved
- IPO pressure: Public markets are scrutinizing AI spending discipline
For the AI industry, Meta’s 91% free cash flow decline is a warning: the AI buildout is expensive, and the market’s patience is not unlimited.