The numbers are in, and they are staggering. Global debt associated with artificial intelligence has reached approximately $687 billion as of November 3, 2026, according to a Goldman Sachs report . But that headline figure only tells part of the story. When you add the lease obligations and purchase commitments that the world’s biggest technology companies have signed, the true scale of the AI buildout’s financial exposure comes into sharper focus—and so does the question that investors, rating agencies, and the Federal Reserve are all asking: is the AI boom becoming too expensive for its own good?
The $687 Billion Number (and What It Doesn’t Include)
The $687 billion figure represents AI-related debt issuance across global syndicated credit markets year-to-date . What makes it particularly notable is that only 43% of that debt was issued directly by the five major hyperscalers—Alphabet, Amazon, Microsoft, Meta, and Oracle . The rest comes from a sprawling ecosystem of neoclouds, infrastructure providers, and specialized AI companies that are borrowing to build capacity.
But the debt is only the visible layer. Goldman Sachs estimates that the five hyperscalers are facing $1.7 trillion in lease commitments**, roughly **75% of which have not yet commenced** . On top of that, they carry **$1.9 trillion in other unconditional purchase commitments, excluding leases, of which $685 billion is due by the end of 2027 . These are legally binding agreements for cloud capacity, servers, networking equipment, and power supply arrangements—obligations that represent “an additional use of cash” even if some are already folded into capital expenditure estimates .
In other words, the $687 billion debt figure is not the whole bill. It is the tip of a financial iceberg that extends trillions of dollars into the future.
The Shift from Cash to Debt
What makes 2026 a turning point is not the size of the spending. It is how it is being financed.
For the first years of the AI boom, the hyperscalers funded their data center buildout largely from the cash generated by their core businesses—search advertising, cloud computing, e-commerce, and software. That era is over. In their most recent quarterly results, the five hyperscalers committed $33 billion more to capital spending than they generated from operations, marking the first time their free cash flow turned negative .
The response has been a wave of borrowing. The five hyperscalers issued **$296 billion in debt year-to-date**, exceeding Goldman Sachs’ full-year forecast of $250 billion . Meta raised $95 billion in stock in a single offering, underlining the new dependence on outside financing . And the neoclouds—the smaller, debt-funded operators that provide GPU capacity—are under even more pressure. CoreWeave’s bonds have sold off sharply, with yields approaching 15%, a signal that lenders are growing wary of the riskiest borrowers in the AI ecosystem .
The Warning Signs: Market Saturation and Rating Agency Skepticism
Goldman Sachs’ report is not alarmist, but it is cautious. The bank notes that “market saturation remains the most binding constraint to the AI-related debt issuance wave,” with companies themselves beginning to acknowledge the issue . Investors are expected to become “more selective about their exposure to AI-related companies and the prices they are willing to pay” as financing requirements continue to grow .
The rating agencies are watching closely. S&P has warned that greater adoption and revenue generation from AI will be necessary to justify the rising investment commitments . Goldman Sachs suggests that “evidence of AI monetisation by 2028 would be important in easing pressure on credit ratings” . That is a polite way of saying that the debt is sustainable only if the revenue arrives—and it is not arriving fast enough to satisfy the skeptics.
The Federal Reserve has taken notice as well. A Fed survey has placed AI at the top of financial stability risks, with advanced-computing debt reaching $680 billion by October 2026—a fivefold increase on all of 2023 . The concern is not that AI is a bubble in the traditional sense, but that the financing structure of the boom has become increasingly dependent on debt markets that may not remain as accommodating if sentiment shifts.
Is the Boom Too Expensive?
The honest answer is: it depends on whether the returns materialize—and on a timeline that is measured in years, not quarters.
The optimists point to the earnings. AI is generating real revenue for the hyperscalers: Google Cloud revenue has surged, Microsoft’s AI services are growing, and Nvidia’s data center business has become one of the most profitable enterprises in corporate history. The spending, in this view, is an investment in a platform shift as significant as the transition to cloud computing or mobile—and the companies that build the infrastructure will capture the value.
But the mechanism that makes the earnings look strong also makes the cash flow look weak. When a hyperscaler buys chips from Nvidia, it capitalizes the cost and depreciates it over 5 to 7 years . Nvidia recognizes the revenue immediately. This creates an “earnings powerhouse” for the hyperscalers, but the true cost shows up in free cash flow—and free cash flow is now negative . If capex stalls, the hyperscalers will still be carrying years of depreciation on equipment they have already bought .
The pessimists focus on the rising indebtedness and increasingly precarious financial structures that the boom is built on . One forecast puts the cost of the AI buildout at some $5 trillion by the end of 2030, $4 trillion of which could be in the form of debt—equivalent to the entire US municipal bond market . Pimco’s CEO has warned that AI companies risk borrowing “too much,” projecting that they could borrow up to $10 trillion in the next five years, which would mean rising interest rates and falling stock prices .
A Bet on the Future, Financed by the Present
The AI boom is not a bubble in the sense that the technology is fake or the revenue is imaginary. The technology is real, and the revenue is growing. But the financing structure of the boom has become dependent on debt markets that are beginning to show strain. The $687 billion in AI-related debt is not a crisis in itself. It is a signal that the era of cash-funded expansion is over, and that the next phase of the AI buildout will be financed by borrowing—borrowing that must be serviced, rated, and ultimately repaid.
Goldman Sachs’ conclusion is measured but pointed: “The cost is only one part of the equation—market saturation remains the most binding constraint” . In other words, the problem is not that debt is expensive. It is that there may not be enough lenders willing to provide it at the scale the industry requires.
The AI boom is becoming more expensive. Whether it is too expensive depends on whether the monetization arrives before the patience of the credit markets runs out. The evidence of AI monetization by 2028 will be the deciding factor . Until then, the industry is making a multitrillion-dollar bet that the future will pay for the present. The debt is the receipt.