NextFin News - The US Treasury is no longer the only borrower moving the bond market. Nearly $500 billion of artificial-intelligence-related debt has been issued this year, Goldman Sachs Research estimates, and the flood is large enough that it is now competing with record government borrowing for the same pool of savings — a "reverse crowding out" that is helping push the 10-year Treasury yield to its highest level since 2007, touching 5% in mid-September. The question for investors is no longer whether tech giants can afford to borrow. It is whether the bond market can absorb what they still need to raise without repricing risk across every other borrower in the economy.
The Scale of the Deluge
The numbers mark a sharp break from the past. Goldman Sachs Research puts AI-related debt issuance at $489 billion so far in 2026, already above its $322 billion estimate for all of 2025. The five largest AI hyperscalers — Amazon, Alphabet, Meta, Microsoft and Oracle — issued $194 billion of debt in the first seven months of 2026 alone, compared with $108 billion globally across the full year of 2025, according to Amanda Lynam, head of credit strategy research at Goldman Sachs. Hyperscalers account for only about 40% of the total; the rest comes from data-center financing, chipmakers and the wider AI ecosystem.
The shift started in 2025, when those five companies sold $121 billion of US corporate bonds, versus an average of $28 billion a year between 2020 and 2024. That is more than four times the pre-AI pace. By October 2025, debt tied to AI had reached $1.2 trillion, or 14% of the investment-grade market — making it the largest sector in the JPMorgan US Liquid index, ahead of US banks. A single technology theme now accounts for roughly one in every seven dollars of high-grade credit. Worldwide AI spending is forecast at $2.52 trillion in 2026, a 44% increase year over year, according to Gartner; the industry spent $1.6 trillion in total between 2013 and 2024. More AI money is being spent this year alone than in the entire prior decade combined.
The borrowing is arriving at the same moment the US government is tapping the market at a record pace. The federal deficit is on track for roughly $2 trillion in fiscal 2026, the national debt has crossed $40 trillion, and the 10-year Treasury yield touched 5% in mid-September, its highest level since 2007. The two demands are colliding inside the same fixed-income portfolios, and the collision has a price.
Why Cash-Rich Tech Is Borrowing Anyway
The first puzzle is why companies sitting on large cash piles are issuing bonds at all. The answer is that AI capital spending is catching up with, and in some cases overtaking, internal cash generation.
"We're already at a point where CapEx is quickly approaching cash flow from operations, and so that paves the way for the debt markets to play a role,"Lynam said. The hyperscalers are not funding a single project; they are pre-funding a multi-year buildout. Lynam describes it as a "waterfall of capital" — exhausting debt, equity and internal cash in sequence to position themselves for several years of investment.
This is a structural change in what these companies are, not just a cyclical spending wave. For two decades, the largest technology firms were asset-light: their value sat in software, intellectual property and scalable cloud services that required modest capital investment. The AI buildout has inverted that model. Data centers, power infrastructure and GPU clusters are heavy, long-lived industrial assets with financing profiles closer to utilities than to software companies.
"The transition from asset-light to asset-heavy models requires unprecedented levels of investment and capital raising,"Moody's said in a recent report.
The borrowing that appears on balance sheets understates the total. A study found that so-called hidden debt at the five US tech giants has risen eightfold in four years to $1.65 trillion — more than the $1.35 trillion of debt that actually appears on their books. These are long-term purchase commitments for chips and servers, and lease agreements for data centers that do not show up as loans. Moody's counts roughly $1.2 trillion of such debt-equivalent liabilities, with more than $820 billion tied to data centers still under construction. The bond market is seeing only part of the tab, which means the true claim on future tech cash flow is larger than headline leverage ratios suggest.
The Market's Absorption Ceiling
Scale alone does not create a problem if investors are eager to buy. The concern is concentration: a single theme is claiming a growing share of a finite risk budget. Lynam and Goldman's credit desk estimate that the hyperscalers could add roughly $2 trillion of debt and still remain investment grade. But the US bond market, by its own historical rules of thumb for issuer concentration and market saturation, can comfortably absorb only about $510 billion before investors push back. That gap between what borrowers can issue and what buyers will tolerate is where the risk lives.
The pushback has begun. Insurance-company participation in 30-year AI bond tranches has already halved between the first and second quarters of 2026. New-issue concessions — the extra yield underwriters must offer to clear a deal — have widened to about 20 basis points from a historical norm of 2 to 3 basis points. Spreads on AI credit have nearly doubled from their tights. S&P Global Ratings counts $225 billion of bonds issued by hyperscalers and related entities such as Nvidia so far in 2026, a 973.7% jump through midyear, putting them on pace for a record $400 billion for the full year.
"Market participants are growing leery of quickly rising leverage from issuers previously characterized by strong and reliable cash flow,"the ratings firm said.
The pressure is not confined to tech. Barclays expects total US investment-grade corporate issuance to reach about $2.46 trillion in 2026, up 11.8% from 2025. When one theme dominates that flow, every other borrower competes for what is left.
"The significant increase in hyperscaler issuance raises questions about who will be the marginal buyer of IG paper,"said Torsten Slok, chief economist at Apollo Global Management.
"Will it come from Treasury purchases and hence put upward pressure on the level of rates? Or might it come from mortgage purchases, putting upward pressure on mortgage spreads?"
Reverse Crowding Out
Economists have long warned about crowding out: when government borrowing soaks up savings and pushes up the cost of capital for everyone else. The AI buildout has produced the mirror image. Corporate demand is now large and inelastic enough to move the price of government debt. In March, the busiest single day on record for US corporate bond sales saw investment-grade issuance top $65 billion, beating the previous one-day record of $52 billion set in 2013. Amazon led with a $37 billion offering that drew about $123 billion of orders. Analysts at Deutsche Bank said the wave of sales added upward pressure on the 10-year yield, which climbed 6 basis points to 4.16% at session highs that day.
The arithmetic is unforgiving. If the first-half trend in combined corporate and government issuance continues into the second half, the total as a share of GDP would be the highest in any year outside the pandemic, according to Capital Economics.
"The reverse crowding out in the corporate bond market has even gotten the Treasury Secretary's attention,"Jurrien Timmer, director of global macro at Fidelity Investments, wrote on X.
Yet this is a repricing story, not a failed-auction story. Demand for both types of debt remains strong: a 30-year Treasury auction earlier this year drew the highest demand in the history of the series, led by overseas buyers, and the investment-grade credit spread compressed to 71 basis points in late January, its tightest level since 1998. The problem is not that capital is refusing to show up. It is that capital is showing up only at a higher price, and the margin is being set by a theme whose cash returns have yet to be proven. That is the transmission mechanism in one line: AI capex does not need to default to raise the cost of capital for everyone — it only needs to be large enough to bid away scarce fixed-income risk capacity.
The Counter-Thesis: Fortresses, Not Fragility
The strongest argument against alarm is balance-sheet strength. Even after billions in issuance, the hyperscalers will carry leverage of only about 0.4 to 0.7 times, compared with an average of just under three times for the US investment-grade market. Tapping debt markets reflects capital-structure optimization and commitments to return cash to shareholders, not a shortfall in internal resources. Oracle scaled back a debt requirement by using equity when balance-sheet concerns surfaced — evidence that issuers themselves are managing the pace.
But credit markets do not price leverage in isolation; they price certainty about the end of a borrowing cycle.
"The credit market works best in funding periods of active re-leveraging when there's an end in sight, it's quantifiable, and ideally there's a period of time where companies can grow into their capital structures,"Lynam said.
"Right now, none of those conditions is in place."The asymmetry is fundamental: equity investors buy the upside of AI, while bondholders take a multi-year commitment with no participation in the gains. For a bondholder, concentration and market saturation matter more than they do for a shareholder, because a bondholder's best case is a fixed coupon while the worst case is a write-down.
Oracle is the canary. The company ended its recent fiscal year with $149 billion of long-term debt, up sharply from $96 billion a year earlier, while holding $31.3 billion in cash. S&P Global Ratings cut Oracle to BBB-minus, one notch above junk. The stock has fallen more than 50% from its June 2 high.
"There's big execution risk,"said Barbara Doran, chief executive and chief investment officer of BD8 Capital Partners.
"There's no question they have a huge backlog, although there is big customer concentration risk there. But it's really the debt issue, as we know. And S&P just downgraded to triple B minus. That is one notch above junk status. And so they're really betting the house on this continued demand and the fact that there is plenty of room for more capacity to be added, but it's going to take time."
That is the distinction the market is drawing: the hyperscalers as a group are not Oracle, but the market is beginning to price the theme as if the marginal borrower matters more than the average one.
What Comes Next
The outlook splits cleanly by time horizon. Over the next three to six months, flows and positioning will dominate. A large supply wall is landing as the Federal Reserve weighs its next move, and the long end of the curve is likely to stay volatile. Over the next 12 to 24 months, the test is monetization: whether AI revenue grows fast enough to cover debt service on data-center leases and project-finance structures that will need refinancing in 2027. Over the long term, the structural shift is durable — technology has become an infrastructure-heavy borrower, and AI debt will remain a permanent, large claimant on investment-grade capital. The cyclical leg is the issuance wall; the structural leg is the asset-heavy balance sheet. They point in the same direction for the cost of capital, but over different horizons.
Three scenarios frame the path. In the base case, issuance moderates toward the market's roughly $510 billion absorption ceiling, with overflow migrating to private credit, infrastructure funds, project-finance joint ventures and non-US bond markets; the 10-year yield trades in a wide range rather than trending. In the upside case for bonds, hyperscaler capital-expenditure guidance decelerates for two consecutive quarters and new-issue concessions revert toward the 2-to-3 basis-point norm, allowing yields to drift lower. In the downside case, AI-related issuance exceeds $600 billion in 2026 while the 10-year yield holds above 5%, forcing wider spreads across the investment-grade complex and accelerating the shift to off-balance-sheet funding.
The signals to watch are concrete. First, the rate of change in hyperscaler capital-expenditure guidance, which has ratcheted up every year. Second, new-issue concessions: whether large deals keep clearing at 20 basis points or revert toward the historical norm. Third, insurance participation at the long end: whether the halving in 30-year AI orders stabilizes or continues. Fourth, explicit issuance roadmaps in which a company commits to sitting out after a quarter's sales. And fifth, whether the spread of indigestion from investment-grade deals into data-center joint ventures and the BB-rated curve broadens.
The falsifying signal is quantifiable: if AI debt issuance exceeds $600 billion in 2026 while the 10-year Treasury yield holds above 5% for a sustained period, the pressure thesis is confirmed. Conversely, if capital-expenditure guidance decelerates and concessions compress back toward 3 basis points, the market is telling investors that the absorption ceiling was never binding.
The AI buildout was priced as an equity story. It is becoming a bond-market story — and bond markets care less about how much you might earn than about when, and from whom, you will be paid.
Explore more exclusive insights at nextfin.ai.
