All news

Meta's spending now outruns its cash, and the AI build-out changes how it is paid for

A model data centre on a graphite plinth beside a brass scale tipped by a stack of paper

Meta's first negative free-cash-flow quarter in more than a decade marks the point where AI spending starts to lean on balance sheets, not earnings.

Meta reported its third quarter after the close on 21 October, and by the usual measures it was a good one. Revenue rose 26% to $64.60bn, and adjusted earnings of $7.40 a share beat a consensus near $6.52. The figure that mattered sat lower in the cash-flow statement. Capital expenditure, including finance-lease principal, reached $34.91bn against operating cash flow of $32.79bn. Free cash flow was minus $2.12bn, compared with plus $10.63bn a year earlier, the company's first quarterly outflow in more than a decade. Meta narrowed its 2026 capex range to $135bn to $145bn, said it would keep investing aggressively into 2027, and announced a non-voting Class C share to be paid as a one-for-one stock dividend. The shares fell 7.0% on 22 October. The Philadelphia Semiconductor Index lost 6.7%, its worst day since 23 June, with AMD down 8%, Micron 6% and Nvidia 4%. The Nasdaq Composite fell 2.6% to 26,008.86 and the S&P 500 fell 1.21% to 7,663.43.

The significance lies less in the price action than in the change in funding. Until recently the largest buyers of AI infrastructure paid for it mostly out of operating cash, which let the rest of the supply chain treat their orders as close to riskless, even as bond sales and off-balance-sheet vehicles such as Meta's Hyperion data-centre joint venture were already growing. When a hyperscaler spends more than its business generates, the build-out leans harder on balance sheets and outside capital. Operating margin fell to 20% from 40%, or to roughly 35% excluding the legal charge. The tremor reached the bond market. The 10-year Treasury closed at 4.96%, down from 5.13% and below 5% for the first time since 22 September, as money sought safety. Semiconductors are more than 26% below their June record yet still up 51.5% for the year, and those valuations assume capex keeps rising.

One quarter is not a trend, and a business growing revenue at 26% is not in distress. GAAP earnings of $3.83 a share also absorbed a $9.89bn legal charge, which clouds the margin picture. The selloff had other causes too. On the same day Brent rose 4% to $113.17 after Houthi strikes on Saudi oil infrastructure, and markets are pricing a Fed hike before year-end, so some of the de-rating reflects oil and rates rather than doubts about AI. The haven bid also proved fragile: the 10-year moved back up to 4.999% on 23 October. This reading would be falsified if the other hyperscalers report cash flow that comfortably covers their spending, or if Meta's 2027 plans show capex easing while revenue holds.

The cycle has not ended, but the way it is financed has changed, and that is what patient capital such as Fanaura has to price. When a build-out moves from internal to external funding, return and risk move with it. They shift toward lenders, lessors and the owners of power and land, and away from the assumption that the end buyer's balance sheet has no limit. For portfolios with heavy AI supply-chain exposure, whether through listed equities or private vehicles that finance data centres, the question is whether those positions were underwritten on the old funding model. The next round of hyperscaler reports and their 2027 guidance should show whether Meta was the first to cross this line or the only one.

For founders

Building an AI-native company? Send us the project.

Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.

Apply in five minutes
Subscribe

Get the news by email

One short note each weekday. No marketing, and you can unsubscribe at any time.