S&P Global’s AI Capex Warning: $5.3 Trillion Bet Tests Hyperscaler Credit

The defining financial risk in the AI infrastructure boom is not that hyperscalers lack profitable businesses. It is that their spending commitments arrive years before the cash returns required to support them.

S&P Global’s September 2026 analysis frames the AI build-out as a credit-quality stress test. Amazon, Alphabet, Microsoft, Meta, and Oracle—dubbed the “Hyper 5”—have collectively invested about $1.1 trillion in capital expenditures over the past five years. Visible Alpha consensus forecasts another $5.3 trillion of spending from 2026 through 2030. 116

That scale may ultimately be justified by AI-driven revenue and productivity gains. But it also means the investment cycle is testing free cash flow, liquidity buffers, leverage, and the resilience of corporate credit markets before those returns are fully proven.

The central risk: spending now, monetizing later

S&P’s concern is fundamentally about timing. Building AI capacity requires enormous upfront outlays for data centers, chips, networking, power, and related infrastructure. The expected payoff depends on sustained demand, high utilization, and the conversion of AI services into durable, high-margin revenue.

The Hyper 5’s capex intensity, measured relative to depreciation, has surpassed peaks seen during the dot-com era and the Great Recession, according to S&P Global. The firms are expected to account for more than half of S&P 500 capital expenditure by 2028, compared with 13% in 2020 and 30% in 2025. 1

Those comparisons are cautionary, not a declaration that the AI boom is the same as either earlier episode. The point is that a concentrated group of companies is making historically large, synchronized bets on capacity while the pace and economics of AI adoption remain uncertain.

Free cash flow is becoming the pressure point

The companies can still generate substantial operating earnings and access capital markets. The issue is that infrastructure spending can consume cash faster than operations replenish it.

S&P Global Ratings projects that combined capital expenditure by six major U.S. hyperscalers—including the Hyper 5 and SpaceX—will exceed $1.3 trillion by 2027. It expects all six to post negative free operating cash flow in 2026 and 2027, with recovery not projected until 2029. 3

This distinction matters:

• Profitability reflects whether the underlying business earns more than it spends in operations.
• Free cash flow shows how much cash remains after investment spending.
• A company can remain profitable while free cash flow falls sharply or turns negative if it is building infrastructure at an exceptional rate.

Alphabet offers one illustration of the pattern. Its capital expenditure reached $44.9 billion in the period covered by S&P Global’s July analysis, more than double the level a year earlier, while free cash flow fell 74.1% year over year. Alphabet subsequently raised its 2026 capex guidance to $195 billion to $205 billion. 4

Why financing structures matter as much as headline debt

As internally generated cash is absorbed by capex, hyperscalers are increasingly using debt, equity issuance, lease commitments, and other financing arrangements to fund AI infrastructure. 3

This does not automatically signal distress. Large companies routinely use multiple funding sources, and leases or partnership structures can help match financing to long-lived assets. The credit concern is that these arrangements can create fixed or debt-like obligations that are not captured by looking only at conventional borrowings.

For investors and lenders, the practical questions are therefore broader than reported debt:

• How much AI capacity is supported by long-term lease or financing commitments?
• How quickly must new capacity be filled to meet return expectations?
• How much cash remains available if AI revenue scales more slowly than planned?
• Are revenue commitments sufficiently durable to support the associated infrastructure obligations?

The more that spending is financed with obligations that persist regardless of short-term demand, the greater the downside if utilization or monetization disappoints.

Oracle is the clearest early warning case

Oracle shows how quickly an AI infrastructure push can reshape a company’s cash-flow and credit profile.

The company reported negative free cash flow of $23.7 billion in fiscal 2026 as capital investment in data centers outpaced operating cash generation. S&P Global Ratings lowered Oracle’s long-term issuer rating from BBB to BBB-, leaving it one notch above speculative grade. 17

Oracle’s case does not prove that its AI investment will fail. It shows the mechanism S&P is highlighting: even a company with growing operations can face ratings pressure when capex, financing needs, and long-dated obligations rise faster than cash generation.

What could turn the build-out into a broader credit problem?

S&P’s framework suggests that the risk becomes more serious if several conditions occur together:

1. AI revenues arrive later than expected. Capacity may be built before customers are ready to pay enough for it.
2. Utilization disappoints. Underused data centers and computing equipment would weaken returns on the investment.
3. Capex stays elevated for longer. A delayed spending slowdown would extend the period of weak or negative free cash flow.
4. Financing obligations accumulate. More debt, leases, and similar commitments can reduce flexibility if conditions deteriorate.
5. Investment cuts occur simultaneously. Because spending is concentrated among a few companies, a broad pullback could have wider implications for suppliers, lenders, and credit markets.

That is why S&P’s late-cycle concern is not simply about whether AI is valuable. It is about whether the credit system is being asked to fund a build-out whose returns have not yet been demonstrated at the same scale as the spending.

The upside case remains real—but it must be earned

The constructive case is straightforward: AI could create new software, cloud, and enterprise-service revenue while lifting productivity enough to support today’s infrastructure costs. S&P Global’s AI Monitor projected total AI-exposed revenue to rise from $468 billion at the end of 2023 to $1.354 trillion at the end of 2026. 2

If demand proves durable and high-margin AI revenue scales quickly, the current capex cycle may look like a necessary investment phase rather than financial excess. S&P Global Ratings’ expectation of free-operating-cash-flow recovery by 2029 reflects that possibility. 3

But the forecast is conditional. The essential watchpoints are AI revenue realization, utilization of new capacity, the persistence of capex, and the full scale of debt-like commitments. Until those improve, the AI race remains a high-stakes test of whether extraordinary infrastructure spending can produce returns before weaker cash flow and rising financial obligations constrain the companies funding it.