AI costs billions. When do shareholders get paid?

The boom comes with a bill

Artificial intelligence can transform businesses. That does not automatically make the investment attractive for shareholders. Between an impressive model and an attractive stock sit electricity bills, depreciation and a great deal of committed capital.

October 2026 earnings reports bring this distinction into focus. Meta offers a particularly clear example: strong business growth and negative free cash flow can coexist. The question is no longer just who has the best AI. It is who earns an attractive return after paying for it.

Meta: a strong business with enormous capital needs

Meta reported third-quarter 2026 revenue of $64.60 billion, up 26% year over year. Operating cash flow reached $32.79 billion. Capital expenditure, including principal payments on finance leases, was $34.91 billion. Under Meta’s definition, that left negative free cash flow of $2.12 billion.

The company expects full-year capital expenditure of $135 billion to $145 billion. This is not a measure of AI-only spending: the infrastructure also supports its existing business. It nevertheless illustrates the scale of capital being committed.

Operating profit needs context as well. It was $12.99 billion, including a $9.89 billion legal settlement charge. Attributing the entire decline in operating margin to AI would mix separate issues.

Three roles, three different bets

The suppliers: Chip, networking and infrastructure companies sell the tools. They can generate revenue while customers are still developing their business models. That does not eliminate order cycles, customer concentration or the risk of a future investment slowdown.

The operators: Cloud providers and platforms fund data centers and must put capacity to profitable use. Microsoft, Alphabet and Amazon are important companies to examine here. Their total cloud revenue should not simply be relabeled AI revenue.

The adopters: Businesses can monetize AI through better advertising, software subscriptions or more efficient processes. At Meta, benefits can appear within the core business without customers purchasing a separate AI product. Management says stronger recommendations and monetization are already producing incremental advertising revenue. Consolidated accounts cannot cleanly isolate how much sustainable return comes from each investment.

Earnings are not the same as available cash

A data center is paid for today but depreciated over several years. Heavy investment can therefore hit free cash flow immediately while the income statement reflects the cost over time. Energy, staffing and model operations add ongoing expenses.

A weak quarter of cash flow is not proof of wasted investment. Equally, rising profit is not proof of attractive returns on capital. What matters is whether today’s spending eventually generates additional, durable cash flows.

A checklist for the next few quarters

• Monetization: Are paid usage, advertising effectiveness or software revenue growing, rather than just free requests?
• Cash flow: What remains of operating cash flow after investment? Leases and stock-based compensation belong in the broader assessment.
• Utilization and pricing: Is new capacity actually used? Does competition reduce prices faster than costs fall?
• Depreciation: Are assumed useful lives consistent with technological change?
• Valuation: How much future success does the share price already assume?

The opportunity — and the counterargument

The optimistic case is that AI improves existing products, creates revenue streams and allows usage to grow faster than costs. Companies with customers, distribution and resilient cash generation have useful starting advantages.

The counterargument is that simultaneous capacity expansion can create oversupply and price pressure. A technology can deliver enormous economic benefits while more of the value accrues to customers than to shareholders.

The SmartDeer view

The AI boom does not have to fail for individual AI stocks to disappoint. Expectations merely need to outpace the cash flows eventually delivered. Business quality, investment requirements and valuation deserve separate analysis — not an automatic winner’s badge for every company building a data center.

The most useful question is not who talks loudest about the future. It is who can fund it and still have money left over.

As of October 22, 2026. For information and analysis only; not investment advice. Hero image: conceptual illustration.