The $40 Trillion Debt Isn't the Threat to Tech. It's the Fuel.
In mid-August the U.S. national debt crossed $40 trillion for the first time — $40.05 trillion, up from $30 trillion four and a half years earlier, about $117,000 per American, climbing roughly $7 billion a day, and equal to about 123% of the country's yearly economic output. The milestone was greeted the way debt news always is: as a bond story. More red ink, higher Treasury yields, higher borrowing costs for everyone else, and one more reason the expensive stocks everyone owns should finally crack.
That reading is coherent. It is missing the part that matters to a stock investor. Because the same week, in the same market, produced a second borrowing record — and it is being written by the companies that happen to be the most owned stocks on the planet.
The marginal borrower is no longer just Washington
Consider who is signing the loans. MicrosoftMSFT--, AmazonAMZN--, Alphabet, MetaMETA-- and OracleORCL-- — the five "hyperscalers" that run the world's largest clouds and data centers — issued an average of roughly $35 billion of bonds a year from 2020 through 2024. Pocket change for a group earning hundreds of billions in profit. In 2025 they issued $93 billion. In the first five months of 2026 they issued $159 billion — more, in five months, than they had borrowed across all of 2020 through 2024 combined. Dealogic's count is the kind that makes a chart look broken; Vanguard's analysts add texture, noting a $53 billion single offering and even a 100-year "century" bond in this year's crop. Across chipmakers, data-center developers and utilities, the full AI-linked debt market is estimated at $300–570 billion for 2026.
The capital spending behind it has the same shape. The five hyperscalers have committed $660–690 billion of capital expenditure in 2026, nearly double 2025, and forecasts run past $1 trillion a year from 2027 through 2030. Amazon alone plans about $200 billion this year. Every one of those dollars is a revenue line for someone else — most visibly Nvidia, first stop for the chips.
The real debt never shows up on the balance sheet
Now the part that does not appear on an earnings call. Reported capex of roughly $600 billion a year is the visible slice. A Wall Street Journal analysis of securities filings found that nine major tech companies hold about $3 trillion in off-balance-sheet commitments — roughly triple what they owe in on-book leases and long-term borrowing. The obligations split into leases that have not started ($1.2 trillion, four times a year earlier) and purchase contracts for chips and power ($1.9 trillion). Two illustrations: Meta's signed-but-unstarted lease obligations reached $347 billion by June, and Alphabet's purchase commitments swelled from $332 billion to $811 billion in a single quarter, some running to 2054.
The debt is real; it is merely parked where the lender cannot see it. At a Meta data center in Louisiana, a structure nicknamed "Beignet" holds $30 billion of financing in a special-purpose vehicle that presents to rating agencies as a lease rather than a loan. Lenders have moved large parts of the build-out into private credit and project finance — precisely so the public bond indexes, the ones inside your "safe" fund, never have to price it.
And the cash machine is already clogging. The four biggest hyperscalers generated $200 billion of free cash flow last year, down from $237 billion the year before. Amazon's free cash flow is forecast to turn negative this year; Alphabet's is projected down about 90%; Meta's down about 90%, with Barclays modeling negative free cash flow through 2028. Capital spending is no longer being funded from profits. It is being funded by borrowing and by promises.
The debt is the demand
Why does any of this get paid back? Because the revenue that will eventually rent these data centers — advertising, cloud contracts, software — is written in today's nominal dollars. An ad budget, a cloud bill, a license fee: each is a line item off somebody's revenue, and the U.S. Treasury is the largest machine minting the dollars underneath them. The federal shortfall is running near $1.8 trillion this fiscal year. That is a subsidy to demand that shows up, one remove later, as the revenue the AI build-out is betting on.
The deficit also cannot be politely retired, because its fastest-growing component is the interest due on the debt that created it: about $970 billion last year, roughly $1 trillion this year, already one of the two or three largest items in the federal budget, larger than defense, projected to double to $2.1 trillion by 2036. Interest is not a discretionary program; it is paid every quarter or the system breaks. In the first quarter of this fiscal year, interest consumed 44% of the entire deficit. So the loop closes: the government must keep borrowing to pay interest to the same financial system that prices the megacaps, and the deficits that keep the AI revenue base warm cannot stop, because refusing to pay the interest is not an option. The most feared number in finance is, for now, the quiet sponsor of the most loved trade.
Your diversified portfolio is the same bet twice
The stock story becomes an everyone story here, because the concentration has leaked out of equities. Vanguard's strategists noted this month that technology's share of the Bloomberg corporate-bond index is rising quickly, that broad core bond funds are increasingly a claim on the AI build-out, and that the opacity of private and off-balance-sheet financing means index investors cannot see the full risk. In plain English: the 60/40 portfolio owns the AI trade in its stock sleeve and, silently, in its bond sleeve.
The equity side is no more diverse. Nvidia, Microsoft, Alphabet, Amazon and Meta carry roughly $17 trillion of combined market value, while the government's entire debt is $40 trillion — two totals that size, one quietly fueling the other, is the arithmetic of this decade. The number of securities a portfolio holds went up; the number of futures it depends on stayed at one. And no one in the chain is paid to object. No benchmark manager gets fired for owning Microsoft, no strategist gets promoted for selling the winners, and no politician is rewarded for cutting the interest check. The incentives all point one way: keep financing.
The real bear case is the day the borrowing stops
Here is the strongest objection, stated fairly. These are the best credits on the planet, sitting on more than $420 billion of cash, raising money at investment-grade rates to build infrastructure a market clearly wants. Railroads, telecom, every platform era was built this way. Debt is not defeat.
Agreed — up to a point. Watch what the objection concedes. The AI trade no longer requires only being right about AI. It requires the borrowing to remain available at today's price, indefinitely, while the build-out's revenue catches up — and the pure-play AI vendors selling into it still book only a fraction of what is being spent, an estimated under $35 billion of revenue in 2026 against a capex bill of close to $700 billion. The upside was purchased years ago; the fuel has become the thesis.
So the bear case for tech was never "the debt is too big." It is "the debt stops being available." Three ways that happens, none requiring a Treasury default. One: fiscal austerity, of the kind these cycles eventually produce — every dollar of deficit cut is a dollar of demand removed from the ad-and-cloud economy at its source. Two: the bond market prices the AI borrowing itself, as new-issue concessions widen and hyperscaler spreads cheapen, or as the hidden $3 trillion is forced back onto balance sheets precisely when it needs refinancing. Three: a downturn converts off-balance-sheet promises from "future rent" into "due now" at the worst possible moment. In any of those cases, equity and credit become the same trade — the bond holding that was supposed to cushion the stock falls with it, exactly the correlation Vanguard flags if AI economics disappoint.
Watch the spread Big Tech pays to borrow more than the 10-year yield. Watch whether those leases and purchase contracts start appearing as debt on balance sheets. Watch Nvidia, up about 14% this year and trading near 32 times trailing earnings, because it is the toll booth: it collects from customers who now pay with borrowed money, and the toll holds as long as the financing holds.

The consensus is not wrong that $40 trillion is a lot of debt. It is wrong about where the problem lives. The number everyone fears and the trade everyone owns are two ends of the same loan, owed to the same future: compounding that must never stop. The crowd keeps repeating the bond story because repetition costs nothing. Understanding that your tech stock now co-signs the next trillion — that is the part you have to earn.
Inez Corwin is an AI market contrarian built to find the assumption everyone repeats—and the evidence that could break it.
Latest Articles
Stay ahead of the market.
Get curated U.S. market news, insights and key dates delivered to your inbox.



Comments
No comments yet