Big Tech’s AI boom is being built on $3.1 trillion in off‑balance‑sheet commitments — a financial pipeline so large it now exceeds hyperscalers’ annual capex by a factor of five and is already pushing free cash flow negative at Alphabet, Amazon and Meta, reported the Wall Street Journal . As demand moderates and obligations come due, investors may need to reassess how much of the AI build‑out is truly self‑funding.

Since the report came out Aug. 16, shares of four of the leading stocks in this analysis have lost ground. Alphabet stock is down 1% ; Amazon shares lost 2% ; Meta fell 8% — possibly due to concerns about a social media child safety case — and Microsoft stock declined 3% while the Nasdaq lost 2% .

Do Investors Care About These Obligations?

The change in these stock prices suggest the news of the off-balance sheet obligations has not terrified investors.

However, stock prices move based on whether a public company reports and forecasts financial results that exceed investor expectations. If companies beat, the stock generally goes up; otherwise shares drop.

Hence, investors should care about the off-balance-sheet obligations if they contribute to deviations from investor expectations. And when it comes to AI hyperscalers, the latest quarterly reports reveal investors focus on four specific questions:

  1. Demand and Capacity — Is demand already contracted or capacity constrained?
  2. AI Revenue Acceleration — Is AI-related revenue accelerating?
  3. Profitability and Cash Flow — Are operating profit and cash generation holding up?
  4. Capex-to-Returns Bridge — Does management provide a credible bridge from capex to returns?

Indeed, as I wrote earlier this month, CoreWeave stock rose 18% after reporting earnings Aug. 11 because the company was able to answer more of these questions in the affirmative than it had in the previous five quarters.

More specifically, while CoreWeave revenue soared 112% to $3.45 billion, the stock responded to a raised 2026 outlook to $3.52 billion; a $104 billion backlog; more than $25 billion in new customer commitments; and a 5% adjusted operating margin (exceeding the 2.9% expectation, per Bloomberg ).

I think the off-balance-sheet obligations could cause investors to lower their expectations for hyperscalers’ profitability and cash flow.

How so? The cash hit from those obligations has begun to hurt and is forecast to increase.

Alphabet’s free cash flow turned negative in Q2 2026 for the first time since 2004, noted CNBC , adding Amazon’s “flipped into the red in the first quarter, and analysts surveyed by FactSet forecast it will stay there for the full year,” and aggregate hyperscaler capex is forecast to exceed operating cash flow during the current quarter, according to Epoch AI .

If investors have already baked these assumptions into their models, the off-balance-sheet obligations may not be a risk. However, if AI cloud services demand moderates, hyperscalers could have less revenue from which to cover their obligations.

Investors could then conclude the companies fail my four tests.

What’s Inside Hyperscalers’ On‑ And Off‑Balance‑Sheet Debt?

The $3.1 trillion off balance sheet figure includes two components — sums roughly $1.2 trillion in uncommenced leases and $1.9 trillion in purchase commitments, per Stocktwits — disclosed by Alphabet, Amazon, Meta, Microsoft, Oracle, Nvidia, Broadcom, SpaceX and AMD — mostly for for AI infrastructure.

Such special purpose vehicles should be fresh in the memory of anyone who remembers Enron’s 2001 bankruptcy. Enron used special purpose entities to minimize losses, accelerate profits, keep debt off its balance sheets and preserve its credit ratings — which seemed to work as long as Enron’s stock price kept rising, according to the Duke Law Scholarship Repository .

Hyperscalers now use SPVs for many of the same reasons. They keep debt off the balance sheet and preserve credit ratings . Microsoft, for example, is one of two U.S. public companies rated AAA by S&P Global — a notch above the U.S. government itself.

Leasing keeps rapidly depreciating GPUs off hyperscalers’ books, let’s them share risk with private credit, and enables them to build faster than internal capital allows while preserving cash for dividends and buybacks. Moreover, hyperscalers have contracted demand — with Oracle’s remaining performance obligations backlog at $638 billion , Microsoft at $678 billion , and Alphabet at $467.6 billion .

Unlike Enron, companies are not concealing their SPVs. However, until the companies start paying to fulfill those future obligations, they are using different assumptions — notably for calculating depreciation, per Stocktwits — which makes it difficult for analysts to forecast their cash-flow impact.

What Could Push Free Cash Flow Deeper Into The Red?

Negative free cash flow is already here and could get worse. As CNBC noted, Alphabet’s second-quarter 2026 free cash flow was negative — down from $73.3 billion generated in 2025.

Per Amazon’s own filing, trailing-twelve-month free cash flow declined to $1.2 billion from $25.9 billion "primarily reflecting large-scale investments in artificial intelligence infrastructure," and had swung to a $7.6 billion outflow by Q2 2026, noted the Wall Street Journa l. Meta’s FCF dropped from $43.6 billion toward single digits, according to Reuters .

Because many commitments are non-cancelable and take-or-pay, if hyperscaler revenue fell short, cash outflows would remain the same but the companies’ ability to cover those outflows would erode.

For example, if Meta had to recognize its uncommenced leases with no matching revenue, liabilities would rise 148%, Calcbench noted . Oracle is most at risk since S&P estimated half Oracle’s future revenue comes from OpenAI , which loses over $10 billion a year. If Open’s AI’s losses worsen, so might Oracle’s FCF — which was negative $23.7 billion in fiscal year 2026.

Which AI Cloud Providers Face The Highest Risk?

Not all AI cloud services providers face equal risk. The relatively safe companies are Meta and Microsoft, which enjoy advertising and diversified software cash flows. Oracle is the highest-risk hyperscaler due to its negative FCF, single-customer (OpenAI) concentration, and record credit-default-swap spreads .

Neoclouds such as CoreWeave and Nebius and OpenAI itself are the most fragile links.

Not all analysts are gloomy. Hyperscalers now build only if they "can see firm commitments or contracts or demand," noted Moody’s Raj Joshi , who added, “The entire industry remains capacity-constrained because demand for computing capacity to train new AI models and support exploding growth in inferencing and agentic applications exceeds supply.”

What Indicators Should Investors Watch?

Here are three indicators investors should monitor:

  1. Decelerating demand as evidenced by slower growth in hyperscaler RPO or bookings.
  2. Higher credit risk measured by wider credit-default-swap spreads or falling bond deal-coverage ratios — which have dropped from 5 times to under 2 times in 2026.
  3. Rating-agency reclassification of off-balance sheet items from uncommenced leases or SPV debt into adjusted leverage — which Moody’s and S&P have signaled they may do.

Investors do not appear to act as though these risks are a big concern now.

However, these indicators could signal a rapid rise in investor fears that these financial commitments could reduce the cash flows and stock prices of AI cloud services providers such as Microsoft, Amazon, Meta, Google, Oracle, CoreWeave and Nebius.