Big Tech's Hidden Debt: What the $1.65 Trillion Really Means

The $1.65 Trillion Nobody Talked About

The five largest US tech companies are carrying $1.65 trillion in financial obligations that don't appear on their balance sheets. Not the debt you can already see. A second, larger pile sitting quietly behind it.

That number came from reporting by Nick Aasia, and it was already out of date by the time most people read it. Days after publication, three of the biggest hyperscalers — Microsoft, Alphabet, and Meta — disclosed nearly $900 billion in new AI commitments in a single quarter. Then the Financial Times uncovered another $50 billion in leases Nvidia had quietly signed for a single data center in Texas, stuffed with its own chips. A commitment that had appeared nowhere in prior disclosures.

So $1.65 trillion is a floor, not a ceiling. The real number is almost certainly higher, and it's still moving.

When figures like these surface, the internet reaches for its most dramatic comparison. And right now, that comparison is Enron. Financial commentators with large followings and limited accounting backgrounds have looked at these disclosures and concluded that big tech is running the same playbook as the most infamous corporate fraud in American history.

They're wrong. But not for the reasons you might expect. And the truth is both more reassuring and more alarming than the Enron framing suggests.

Why the Enron Comparison Doesn't Hold Up

Enron wasn't just aggressive accounting. It was criminal fraud. The company hid enormous debts and losses inside a web of secret off-books entities specifically designed to deceive investors. The financial statements that shareholders could see were, in effect, fiction. When it collapsed in 2001, it wiped out the retirement savings of thousands of employees and took down Arthur Andersen — one of the five largest accounting firms in the world — with it.

That's the accusation the Enron comparison carries. Not "confusing footnotes." Deliberate, criminal deception at scale.

Big tech's off-balance-sheet obligations are a different beast entirely. The bulk of what's missing from the main balance sheet consists of two things:

• Purchase commitments: long-term agreements to buy chips and hardware that haven't been delivered yet
• Operating leases: contracts on data centers that haven't started operating

Under standard accounting rules — specifically US GAAP and IFRS — you don't record a liability for goods not yet delivered or a building not yet in use. You disclose it in the footnotes to the financial statements. The information is there. It's just not in the headline number.

The phone contract analogy is instructive here. When you sign a two-year mobile contract at $50 a month, you've committed to paying roughly $1,200. That's a real obligation. But you don't book a $1,200 liability on day one. You record the cost as you use the service. Tech companies are doing the same thing — just with data centers in Ohio instead of handsets in your pocket.

The details are in the accounts. You just have to go looking for them. That's camouflage, not fraud. And it only works on people who don't read past the first page.

What the Debt Signal Actually Tells Investors

When a company chooses debt over equity to fund growth, it sends a specific signal to markets. The logic runs like this: if you're convinced you've built a machine that turns $1 into $5, you don't sell half the machine to raise the money to build it. You borrow, build, keep all the upside, and pay back the loan. You only dilute ownership when you're less certain the bet will pay off.

By that logic, the wave of debt issuance across big tech should read as a confidence signal. Management believes the return on AI infrastructure investment justifies keeping ownership concentrated rather than selling shares.

Except these companies are also raising equity — at record scale. In June, Alphabet completed what was reported as the largest equity raise in corporate history, approximately $85 billion, anchored by a $10 billion investment from Berkshire Hathaway. Berkshire, notably, bought in at roughly a 6% discount to market price. This is not a firm that overpays for a story or funds moonshots out of sentiment.

When a company is raising capital by every available route simultaneously — debt, equity, convertibles, lease financing — the signal isn't in which instrument they chose. It's in the sheer scale of the raise. You don't mobilise capital like a wartime economy unless you believe there's something on the other side worth the fight. Whether the AI buildout justifies that belief is a separate question. But it's difficult to argue the spending is being hidden when part of it was announced in the largest stock offering ever filed.

The Accounting Tricks That Are Hidden in Plain Sight

If big tech isn't committing Enron-style fraud, it isn't playing it straight either. The more legitimate concern isn't what's buried in the footnotes. It's what's sitting on the front page of the earnings report under a heading that says "adjusted earnings."

Two practices deserve particular scrutiny.

EBITDA and depreciation: Companies spending tens of billions on data centers and chips prefer to report EBITDA — earnings before interest, tax, depreciation, and amortisation. The late Charlie Munger suggested that whenever you see the word EBITDA, you should mentally replace it with "BS earnings." His point was precise: depreciation isn't an accounting abstraction. Physical assets wear out and need replacing. Stripping it out assumes that servers and buildings last forever, which they don't. The strain shows up in the cash numbers. The four biggest hyperscalers recently posted their lowest combined free cash flow in a decade — roughly $7 billion between them — and Alphabet went cash-flow negative for the first time since going public.

Stock-based compensation: Tech companies routinely pay staff in equity and then add that cost back out of adjusted earnings on the grounds that it's "non-cash." NYU finance professor Aswath Damodaran has called this one of the worst abuses in modern financial reporting. The cost isn't non-cash in the way depreciation is. It's a barter transaction. If the company sold shares on the open market and used the proceeds to pay employees, everyone would call it a cash expense. Handing shares directly instead of selling them doesn't make the economic cost disappear. Warren Buffett put the question simply: if options aren't compensation, what are they? If compensation isn't an expense, where should it go?

To prevent stock-based compensation from inflating the share count, companies buy back their own shares using real cash — presented to investors as "returning capital." In reality, they're running on a treadmill. A buyback only rewards remaining shareholders if shares are bought cheaply. A company mopping up its own stock-based compensation doesn't get to wait for a good price. It has to buy on schedule, at whatever the market charges. Lately, that hasn't been cheap.

The Circular Financing Problem at the Heart of AI

The more structurally interesting concern isn't accounting presentation. It's the web of financial relationships between the companies at the centre of the AI boom.

Nvidia is currently working on AI financing deals reportedly worth more than $750 billion. It is in talks to backstop $250 billion to help OpenAI lease computing power and to help finance another $350 billion in chip purchases. Google has agreed to cover lease payments for Anthropic, effectively extending it a $35 billion credit line. SoftBank committed $65 billion to OpenAI and took out a $40 billion bridge loan just to fund that stake.

Draw the diagram of who owns what and who's financing whom, and the companies at the centre of the AI boom turn out to be mostly investing in and lending to each other.

Nvidia's CEO Jensen Huang has called the suggestion that any of this is circular "ridiculous" — which is a strong word to reach for while backstopping a quarter of a trillion dollars in purchases of your own product. To be fair, there's an established precedent. Telecom equipment makers and aircraft manufacturers have been providing vendor financing — lending customers the money to buy their products — for decades. The logic for Nvidia doing it is coherent: the AI buildout is moving fast enough that companies like OpenAI can't raise sufficient conventional capital to buy the computing power they believe they need. By stepping in, Nvidia locks in a customer, ensures its chips get deployed, and acquires equity stakes in companies that could be worth multiples of the investment if the bet pays off.

The trouble with vendor financing is the downside scenario. If a customer fails, the loss isn't just the equity stake. If Nvidia has guaranteed the customer's debts, a valuation problem can become a solvency problem. Right now, Nvidia generates approximately $200 billion a year in cash. It can absorb a startup failure or two. The question is what happens as the guarantees climb into the hundreds of billions and a company that historically carried minimal debt finds itself standing behind everyone else's obligations.

The credit market noticed. The cost of insuring Nvidia's debt against default jumped by the most on record in a single day as these deals were announced. The people whose job is to price the risk of Nvidia not paying its bills had a look at the structure and got noticeably less relaxed.

The Big Market Delusion: When the Story Prices Out the Numbers

All of this circular financing rests on a shared assumption: that the market for AI will be so large that whatever is spent today will look like a rounding error by the time it matures. Finance professor Aswath Damodaran and his co-author Bradford Cornell have a name for what happens when an entire industry organises itself around this kind of thinking. They call it the big market delusion.

The mechanism works like this. A transformative technology arrives attached to a genuinely enormous potential market. Multiple companies emerge to compete for it. Investors price each company as if it's going to be the dominant winner. The problem is that they can't all be right. Add up what the market expects each AI company to earn, and you get a number that exceeds what the market itself will generate. Everyone has been priced to come in first in a race with one winner.

The Economist's analysis puts some shape around this. The AI buildout is on track to represent the largest investment surge in recorded history — approximately $900 billion in a single year on chips, data centres, and power infrastructure, with more than $400 billion of it borrowed. To service that capital at a reasonable return, the industry would need to be generating roughly $2.5 trillion a year in AI revenue. That figure is larger than the entire global technology sector earns from everything it does today — software, hardware, cloud, devices, advertising, everything combined.

The actual figure being generated by AI right now is not close to that.

Adoption is real: around a fifth of American firms report using AI in some form. But a Bank of England study found that the average American executive spends approximately 100 minutes a week using AI tools. And nine out of ten executives surveyed said AI had made no measurable difference to their company's productivity over the prior three years. When a technology is genuinely transforming an economy, people tend to notice.

The more instructive data point comes from the bottom of the market. Fintech firm Ramp analysed actual company spending and found the median firm was spending $10.66 per employee per month on AI tools. The largest capital investment surge in history is being financed to capture spending that, at the median, amounts to roughly two cups of coffee per employee per month.

The clearest near-term beneficiaries of AI, based on available data, aren't the giants spending hundreds of billions. They're small businesses and solo founders using tools like ChatGPT to handle tasks — building a website, processing paperwork, drafting contracts — that previously required hiring someone. According to payroll firm Gusto, the share of new business founders who used AI during startup doubled to 60% in two years. They're paying around $20 a month. The unit economics look very different from the hyperscaler model.

The Practical Takeaway for Investors

None of this means big tech is going to collapse next quarter. These are companies with genuine competitive moats, exceptional cash generation histories, and management teams that have compounded shareholder value through multiple cycles. The comparison to Enron is not just wrong — it actively misleads people about the real risks.

But the real risks are worth taking seriously:

• Off-balance-sheet obligations are growing faster than disclosed revenue from AI. The timing gap between commitment and cash flow is widening, not narrowing.
• Adjusted earnings metrics systematically exclude real costs. EBITDA and non-GAAP figures strip out depreciation and stock-based compensation that represent genuine economic drains on the business.
• Circular vendor financing creates correlated risk. If the AI revenue build is slower than expected, the companies most exposed aren't just the startups — they're the chip and infrastructure providers who guaranteed the debt that funded the revenue.
• Market pricing reflects winner-take-all outcomes that can't all occur simultaneously. The math of what's been priced in versus what the market can realistically generate doesn't close at current valuations.

The story is doing a lot of work. Investors who want to participate in the AI infrastructure build without overpaying for the narrative should focus on free cash flow — not adjusted earnings — and read the footnotes, not just the headlines.

Because the money is real. The commitments are real. The question is whether the revenue on the other side will be real enough, fast enough, to justify all of it.

---

This article is for informational purposes only and does not constitute financial advice. Always consult a qualified financial professional before making investment decisions.

Frequently Asked Questions

Why don't off-balance-sheet obligations appear on tech companies' balance sheets? Under standard accounting rules (US GAAP and IFRS), a liability is only recorded on the balance sheet once the economic event that triggers it has occurred — goods delivered, a building in use, a service rendered. Long-term purchase commitments for chips not yet shipped or leases on data centres not yet operational don't meet that threshold. They must be disclosed in the footnotes to financial statements, which they are. The information isn't missing; it's just not in the headline number.

Is what big tech is doing comparable to the Enron scandal? No. Enron created secret off-books entities specifically to hide debts and losses from investors. The financial statements shareholders saw were deliberately falsified. Big tech's off-balance-sheet items are disclosed in company filings — they're in the footnotes. That's a fundamental difference. The Enron comparison makes for dramatic content; it doesn't accurately describe the accounting practice in question.

What is vendor financing, and why is Nvidia using it? Vendor financing is when a supplier lends money to a customer to buy its own products. It's been used for decades in industries like aerospace and telecoms, where customers need large capital outlays that conventional debt markets can't fully support. Nvidia is using it because AI companies like OpenAI need computing infrastructure faster than they can raise capital through conventional channels. By backstopping debt or leases, Nvidia secures long-term customers and equity stakes. The risk is that if those customers fail, Nvidia loses both the customer and the money it lent them.

What is the 'big market delusion' and how does it apply to AI? The big market delusion, identified by finance professors Aswath Damodaran and Bradford Cornell, occurs when multiple companies in a new industry are each priced by investors as if they will be the dominant winner of a genuinely large market. Aggregating those individual valuations produces an implied total market size that exceeds what the market can realistically generate. In AI, The Economist estimates the industry would need to generate roughly $2.5 trillion annually in AI revenue to justify current capital spending — a figure larger than the entire global technology sector earns today from all sources combined.

Why do analysts focus on free cash flow rather than adjusted earnings for big tech? Adjusted earnings figures — particularly EBITDA and non-GAAP earnings — strip out costs that represent real economic drains on the business: depreciation on assets that genuinely wear out, and stock-based compensation that dilutes shareholders if not offset by buybacks funded with real cash. Free cash flow, by contrast, measures actual cash generated after capital expenditure. It's harder to flatter. The four largest hyperscalers recently reported their lowest combined free cash flow in a decade — approximately $7 billion between them — which tells a materially different story than their adjusted earnings figures suggest.