S&P Warns Hyperscalers Will Burn Cash Until 2029 as AI Capex Hits $1.3 Trillion

The AI infrastructure boom has produced plenty of eye-watering numbers, but the one S&P Global Ratings dropped on August 27, 2026 may be the most consequential yet: combined capital expenditure at the world’s six largest hyperscalers is projected to exceed $1.3 trillion by 2027, and every one of them is expected to generate negative free operating cash flow in both 2026 and 2027 — with recovery not projected until 2029.

The report, “S&P Global Ratings’ View On Artificial Intelligence And Hyperscalers,” examines how Alphabet, Amazon, Microsoft, Meta, Oracle, and SpaceX are funding the largest infrastructure build-out in corporate history, and what that means for the credit quality of the companies underneath it.

The headline numbers

The scale of the commitment is without precedent. Six companies — four of which were, until recently, among the most prodigious cash generators in the history of capitalism — are now collectively spending more each year on data centers, GPUs, networking, and power than most countries’ GDP.

The key findings from the report:

• Combined hyperscaler capex is projected to exceed $1.3 trillion by 2027. For context, that single-year figure approaches the annual economic output of Spain.
• All six hyperscalers are expected to generate negative free operating cash flow in 2026 and 2027, with recovery not projected until 2029.
• Debt, leases, guarantees, and other financing structures are playing an increasingly important role in funding AI infrastructure growth.
• S&P’s models assume a 2028 inflection point, with revenues accelerating and capex growth moderating as monetization improves.

Why negative free cash flow — at this scale — matters

Free cash flow is the money a company has left after paying for its operations and capital investments. Tech giants have historically used it to fund buybacks, dividends, and acquisitions. Now that cash is being poured — along with enormous volumes of borrowed money — into GPU clusters and gigawatt-scale campuses.

The squeeze is already visible. Alphabet recently posted its first negative free-cash-flow quarter, driven by AI infrastructure spending. Oracle’s free cash flow turned negative as its cloud backlog swells with AI demand it must build out capacity to serve. Raymond James noted in mid-August that by the second quarter of 2026, aggregate hyperscaler capex had begun to exceed operating cash flow, pushing free cash flow into negative territory across the board.

An individual year of negative FCF is not itself alarming for companies with massive balance sheets. What concerns credit analysts is the duration and the debt-like commitments stacking up around it.

The hidden liabilities: JVs, SPVs, and residual value guarantees

The most technically interesting part of the S&P analysis concerns how the build-out is being financed — not just how much.

Hyperscalers are increasingly turning to:

• Joint ventures that pool capital across multiple balance sheets
• Special purpose vehicles (SPVs) that isolate data center assets off the parent’s books
• Residual value guarantees (RVGs) — promises to make asset owners whole if equipment loses value faster than expected
• Debt, equity issuance, and lease commitments layered on top

Each structure moves risk somewhere less visible. SPVs keep construction debt off the headline balance sheet. RVGs convert depreciation risk into contingent liabilities that appear only in footnotes. JVs split obligations across partners whose own credit positions may differ wildly.

S&P explicitly warns that this growing complexity is increasing the difficulty of credit analysis. The rating agency’s key monitoring areas now include: monetization of AI investments, demand durability, overcapacity risk, and the treatment of contractual commitments and other debt-like obligations.

A bet on 2028

Strip away the financing mechanics and the entire edifice rests on one assumption: that AI revenue inflects upward around 2028, before the financing structures begin to bite.

That is why S&P’s models assume a 2028 inflection point — revenues accelerating as enterprise AI adoption matures, while capex growth moderates because the first wave of builds is complete. If that timing holds, today’s negative cash flow reads in hindsight as disciplined aggression. If it slips — if monetization lags by even a year or two — the sector enters 2029 with trillions in debt-like obligations and a revenue base still catching up.

The overcapacity question is particularly pointed. These companies are building ahead of demand they can see (cloud backlog is real and measurable) but also ahead of demand they can only hope for (end-user AI monetization remains uneven). History is not encouraging on this point: telecom carriers in the late 1990s also had real demand, real traffic growth — and still managed to bury themselves under fiber they couldn’t pay for.

What to watch

For anyone tracking whether the AI capex cycle is a bubble or a build-out, S&P has essentially published the checklist:

1. AI revenue growth at each hyperscaler — does it accelerate toward the 2028 inflection?
2. The pace of new financing structures — are SPVs and RVGs proliferating faster than disclosure can keep up?
3. Free cash flow trajectories in 2027 — is the trough shallower or deeper than modeled?
4. Credit spreads on hyperscaler debt — the market’s real-time verdict on the bet.

One more number worth remembering: Nvidia, whose chips sit inside much of this build-out, now expects hyperscaler capex to top $800 billion this year alone. The people selling shovels are getting paid upfront. The people buying them are financing it — and the bill comes due in 2029.

---

Sources are listed in the frontmatter of this post.