But behind the scenes, a major shift has occurred. The AI boom has evolved from a lean software race into the most capital-intensive, hardware-dependent infrastructure scramble in modern capitalism. Tech hyperscalers are now burning cash at a pace that rivals the budgets of major governments.
Worse, the true financial scale of this arms race is often underappreciated by the average retail investor. A research analysis published by Morgan Stanley in November 2025 identified about $1.8 trillion in off-balance-sheet exposures tied directly to AI infrastructure. These liabilities are legally binding, yet they sit quietly in the footnotes of corporate filings.
As the Bank for International Settlements (BIS) warned in late 2025, the rush to fund AI infrastructure increasingly resembles the great speculative booms of the past, from Railway Mania to the dot-com crash. And in the months since those warnings, the story has accelerated rather than cooled.
To understand what comes next, investors need to look past headline earnings and examine the structural math driving the AI debt loop–without falling for the arithmetic errors that have plagued early bearish models.
The Scale of Spend
The numbers have moved fast, so let’s anchor them to dates.
In 2025, the four largest AI spenders–Alphabet, Microsoft, Meta, and Amazon–allocated roughly 65% of operating cash flow to capital expenditure. In 2026, that figure is tracking toward about 90%, with combined capex projected to exceed $700 billion this year and Wall Street estimates topping $1 trillion for 2027. Epoch AI, after parsing SEC filings from the five largest cloud companies, found capex growing about 70% per year while operating cash flow grew just 23%–a divergence that could push aggregate free cash flow to zero around Q3 2026.
This spending wave is already pressuring the top of the tech hierarchy:
- Alphabet guided 2026 capex to $180–190 billion, nearly double the $91.4 billion it spent in all of 2025. To fund that pace, Google recorded its first zero-buyback quarter in nearly a decade, raised more than $55 billion in fresh debt since November 2025, and in June 2026 priced one of the largest equity offerings in corporate history–an $80 billion raise upsized to $84.75 billion, anchored by a $10 billion Berkshire Hathaway placement.
- Amazon reported a roughly 95% year-over-year contraction in free cash flow through late 2025 as data center spending accelerated.
- Oracle moved into deeply negative free cash flow, raised $43 billion in debt during fiscal 2026, and added a $20 billion at-the-market equity program.
- Meta also came under pressure, with shares falling on reports that it may consider a multi-billion-dollar share sale to fund data center expansion.
The key point is simple: the buyback era that defined mega-cap tech for a decade is starting to reverse into an equity-issuance era. When companies famous for shrinking their share counts begin expanding them to buy chips and concrete, the capital structure of the trade has fundamentally changed.
Inside the $1.8 Trillion Footnote
If public debt and equity issuance are worrying, the undisclosed liability pool is even more important.
Morgan Stanley’s global valuation team found that about $1.8 trillion in future, contractually locked cash outflows was bypassing main corporate balance sheets as of late 2025. This shadow leverage is split into two major buckets.
The Shadow Ledger Breakdown:
- Unexecuted operating leases (~$822 billion): Hyperscalers sign long-term agreements to rent massive data centers built by third-party developers. Under GAAP accounting, those obligations do not fully hit the balance sheet until the physical building is handed over and operational. Until then, the liability sits buried in the fine print.
- Take-or-pay procurement deals (~$982 billion): To secure chip supply, tech giants enter binding purchase obligations with suppliers such as Nvidia. These contracts often require payment whether or not the company ends up deploying every unit, and in many cases, those commitments do not appear as current balance sheet debt.
The result is a sharp increase in apparent leverage once those obligations are included. Morgan Stanley said the aggregate gross leverage ratio for hyperscalers rose from 0.9x to 1.8x in a matter of quarters after these commitments were counted.
Think of it this way: it is like evaluating a family’s finances by looking only at their active bank loans while ignoring the signed contracts committing them to rent three additional homes for the next fifteen years. Nothing illegal is happening–disclosure rules simply weren't built for commitments arriving this fast.
The Revenue Gap: Correcting the Math
For an infrastructure boom of this scale to remain solvent, end-user revenue must grow fast enough to cover the underlying hardware, electricity, and capital costs.
However, early bearish analyses made a critical error: they treated Nvidia’s annual hardware revenue as a one-year expense. In reality, hardware is a capital asset depreciated over 3 to 5 years. Applying the correct accounting changes the math dramatically.
Let’s run a corrected version of the payback framework using mid-2026 inputs. Nvidia’s quarter ended April 2026 delivered a record $75.2 billion in data center revenue, annualizing to roughly $300 billion.
| Metric | Calculation / Basis | Estimated Run Rate |
|---|---|---|
| Nvidia Annualized DC Revenue | Q1 FY27 annualized run rate | ~$300 Billion |
| Effective Annual Hardware Depreciation | Based on standard 4-year useful life | ~$75 Billion |
| Data Center Overhead | Power, cooling, real estate (Matches depreciation) | ~$75 Billion |
| Total Annualized Infrastructure Cost Base | Depreciation + Overhead | ~$150 Billion |
| Required Cloud Service Revenue | Assuming 50% hyperscaler margin | ~$300 Billion |
| Required End-User / Enterprise Revenue | Assuming 50% application layer margin | ~$600 Billion |
For context, the current global cloud infrastructure market (AWS, Azure, GCP) is already running at a ~$350–400 billion run rate and growing at roughly 20%+ annually.
To be fair to the bulls, the demand side is also accelerating faster than skeptics anticipated. Anthropic reportedly crossed a $30 billion annualized run rate in April 2026–up from roughly $9 billion at the end of 2025–and reported a figure near $47 billion by May. OpenAI’s run rate sits near $25 billion.
Additionally, hyperscalers are not relying solely on external AI labs for revenue. They are also generating massive internal cost savings–AI-driven ad targeting at Meta, logistics optimization at Amazon, and coding efficiency at Alphabet–which improves margins without showing up as top-line "AI revenue."
When we combine the existing cloud run rate (~$400B) with the explosive growth of frontier labs and internal efficiency gains, the real gap to the ~$600B required is roughly 4x narrower than the early doomsday models suggested. That does not mean the gap is closed–but it shifts the conversation from an imminent solvency crisis to a race between enterprise adoption and capex growth.
Why This Debt-Funded Boom Is Different
A normal tech bubble usually hurts equity investors first and leaves the broader financial system mostly intact.
Debt changes that equation. As cash reserves tighten, a growing portion of the current data center buildout is being funded by private credit funds and alternative asset managers, which in turn source capital from institutional insurance pools and pension funds. The BIS flagged this transmission path as a key financial stability concern.
If hyperscalers slow capital spending because software revenue lags, the damage does not stop at the top of the market. It can spread to smaller engineering, logistics, construction, and cooling firms that borrowed heavily to support the AI buildout. Because private credit is less transparent than public markets, a localized slowdown could turn into a broader liquidity squeeze faster than many investors expect.
However, it is important to distinguish between a valuation reset and a debt default. The hyperscalers collectively hold pristine investment-grade credit ratings and have just proven they can tap equity markets for $80 billion in days. The most likely outcome of a slowdown is not a wave of bankruptcies, but a sharp deceleration in future orders–which would crater Nvidia’s forward guidance and compress the P/E multiples of the entire Nasdaq.
What Investors Should Watch
The point of this analysis is not to dismiss AI as hype. The technology is real, and the revenue growth at the frontier layer is also real. The issue is whether corporate capital structures can support the pace of investment being demanded by the market.
Going forward, monitor these three indicators rather than adjusted EPS headlines:
- The FCF-to-Capex Trajectory: The big four went from consuming ~65% of operating cash flow in 2025 to ~90% in 2026. Watch whether Q3 2026 marks the crossover into aggregate negative free cash flow, as Epoch AI’s filing analysis projects. A negative print would signal that future spending growth must decelerate sharply.
- Purchase Obligations in the Footnotes: The $1.8 trillion shadow ledger updates every quarter in the 10-Q commitments section. Watch for any deceleration in new commitments–that will be the earliest leading indicator that capital discipline is returning.
- The Financing Mix: Buybacks paused, then record debt, now equity sales. Each step down that ladder signals that cash flow is falling further behind committed spending. Alphabet went first; watch who follows, and at what price.
In every historic infrastructure race, the transition from unchecked optimism to balance-sheet discipline has tended to arrive faster than the consensus expects. The footnotes are usually where it shows up first.
Sources: Morgan Stanley global valuation research (Nov. 2025), Bank for International Settlements bulletins (2025), Nvidia Q4 FY2026 and Q1 FY2027 results, Alphabet Q1 2026 results and June 2026 equity offering filings, Epoch AI hyperscaler cash flow analysis (2026), and company disclosures from Anthropic and OpenAI.
This article is for educational purposes only and is not investment advice.
Check out this breakdown of Alphabet's massive $80B AI capital raise to see exactly how mega-cap tech is pivoting its corporate financing strategy to fund the ongoing infrastructure buildout. This video is highly relevant as it uncovers the direct, real-world mechanics of the public equity shift detailed in the article.