AI Earnings Test Shifts From Demand to Financing

The earnings test facing the AI trade is no longer a simple referendum on demand. Semiconductor orders, equipment sales and platform revenue indicate that demand remains substantial. The emerging question is whether the companies building the infrastructure can translate that demand into returns quickly enough to cover capital costs without weakening cash generation or continually expanding their financing commitments.

Meta illustrates the distinction. Its third-quarter revenue rose 26% year over year, and the company said AI was improving engagement, recommendations and advertising monetization. Those are signs of benefits inside an established, profitable business. Yet the same quarter produced $32.79 billion of operating cash flow against $34.91 billion of capital expenditure, including finance-lease principal payments, leaving the company’s free-cash-flow measure negative by $2.12 billion. Meta also reported $24.91 billion in net proceeds from long-term debt issuance during the first nine months of 2026. (investor.atmeta.com)

These figures do not establish financial distress. Meta held $82.45 billion in cash, cash equivalents and marketable securities at quarter-end, and its operating business continued to generate substantial cash. Nor can its lower reported operating income be attributed solely to AI spending, because the quarter included a $9.89 billion litigation charge. The evidence instead shows that infrastructure spending has become large enough to change cash-flow timing and the company’s mix of financing. (investor.atmeta.com)

That timing creates the central market risk. Data centers, processors and power capacity require expenditure before the resulting services generate revenue. During the interval, companies absorb financing costs, lease obligations and eventually depreciation. If utilization or monetization develops more slowly than capacity, returns can fall even while end demand continues growing. Higher interest rates strengthen this mechanism by raising the cost of external capital and reducing the present value investors place on distant earnings.

The supplier results explain why weak stock reactions should not be confused with evidence that AI activity has stopped. Lam Research reported record revenue and stronger margins, while TSMC maintained an unusually large capital budget and described AI-related demand as robust. Suppliers can therefore experience strong current sales at the same time investors become less willing to give infrastructure buyers—or their vendors—credit for returns that remain years away. (investor.lamresearch.com)

There is, however, a credible alternative explanation for part of the market weakness. AP reported that rising oil prices pressured the wider market on October 23, while Reuters identified elevated Treasury yields and energy costs as broader headwinds. The reaction cannot therefore be assigned entirely to deteriorating confidence in AI. One week of trading and a limited set of earnings also cannot establish the top of a multiyear investment cycle. (investing.com)

The more defensible conclusion is narrower: strong demand is no longer sufficient evidence for the investment case. Companies must increasingly demonstrate who funds the build-out, how quickly capacity becomes productive and whether incremental AI revenue exceeds the combined cost of equipment, power, financing and depreciation.