NextFin News - The 10-year US Treasury yield is trading above 5%, and a growing number of bond-market strategists are pointing to an unlikely culprit: the artificial-intelligence buildout. For decades, "crowding out" meant government borrowing squeezed the private sector. Now the trade has flipped — six AI-heavy technology companies are issuing so much long-duration debt that they are competing with the US Treasury for the same pool of capital, and the price of that competition is showing up in government borrowing costs.
The reversal is stark. Michael Cembalest, chairman of market and investment strategy at JPMorgan, estimates that the five major hyperscalers plus Nvidia have issued roughly $320 billion of debt so far in 2026, including special-purpose vehicles where the companies ultimately stand behind data-center lease obligations. In 10-year-equivalent terms, the long-duration component comes to about $303 billion — equivalent to roughly 68% of new long-duration Treasury borrowing this year. Six private issuers now sit across the table from the US government, bidding for the same long money. Data as of late September 2026.
The backdrop is a Treasury market already stretched by fiscal math. US debt has surpassed $40 trillion, of which $32.27 trillion is held by the public, and the Treasury raised a net $2.4 trillion last year to finance the deficit. The share of marketable debt has risen to 22.4% of debt outstanding, above the 20% limit the Treasury itself has recommended. Into that fragile absorption environment, the AI borrowing wave is not a marginal disturbance — it is a second large issuer competing for the same long-duration buyers.
The Mechanism: How AI Borrowing Moves Treasury Yields
The channel is not mystical; it runs through the marginal buyer. A 10-year corporate bond and a 10-year Treasury note are not perfect substitutes, but their long-duration buyers overlap heavily — pension funds, insurers, and asset managers with long-dated liabilities who need duration, not credit risk per se. When AI issuers flood the market with 10- and 30-year-equivalent paper, they do not just widen their own spreads. They pull duration demand away from the Treasury curve, forcing the government to pay more to clear the same auction.
This is the inversion of the textbook crowding-out story. The classic version holds that excessive government borrowing leaves little room for companies and drives their borrowing costs to punishing levels. What is happening now is the opposite: private borrowers, racing to fund data centers and chips, are absorbing a share of long-duration capacity large enough that the marginal buyer of a 30-year corporate bond and the marginal buyer of a long Treasury are increasingly the same investor. That investor sets the clearing price for mortgages and long yields alongside AI capital expenditure.
The scale makes the mechanism plausible rather than theoretical. Goldman Sachs estimates roughly $300 billion of AI-related issuance has already come to market this year, a figure that independently converges with JPMorgan's $320 billion read. Goldman strategist Amanda Lynam puts total AI-related debt issued this year at $489 billion, already above the $322 billion that came to market in all of last year. Two banks, two methods, one conclusion: the supply wave is real and it is large relative to the Treasury market's incremental long-duration absorption.
The transmission shows up in the term premium — the extra yield investors demand for holding long-duration risk instead of rolling short-term bills. When supply of long paper rises faster than the pool of patient capital, that premium rises. The 30-year Treasury yield illustrates the pressure: it climbed to 5.31% in August, its highest level since June 2007, before the Treasury moved to intervene. The 10-year yield, which anchors mortgages and corporate borrowing across the economy, has been trading above 5% in late September, with major banks lifting year-end targets — JPMorgan raised its 10-year forecast by 20 basis points to 5.05%, and Citi now sees the 10-year at 5.0% by year-end.
The Scale: A Capital-Intensive Boom Funded by Debt, Not Cash
The debt surge is the financing side of a larger capex story. The five largest hyperscalers — Amazon, Microsoft, Alphabet, Meta, and Oracle — are on track to spend between $700 billion and $900 billion on capital expenditure in 2026, a 36% increase over 2025, according to CreditSights estimates. Amazon alone has guided to $200 billion this year, more than doubling its 2025 outlay, while Alphabet has roughly doubled its guidance to $175 billion to $185 billion.
Crucially, this spending is not being funded from cash alone. Goldman Sachs projects that investment-grade global bond issuance from the hyperscalers will reach roughly $250 billion in 2026, about 33% of capex, up from $108 billion — roughly 26% of capex — in 2025. In 2027, the debt-funded share is expected to rise again, to about $400 billion against an estimated $1.14 trillion in capex, or roughly 35%. Barclays forecasts overall US corporate bond issuance will reach $2.46 trillion in 2026, up 11.8% from $2.2 trillion in 2025, with net issuance of $945 billion, up 30.2% from $726 billion. Its analysts identified AI hyperscaler capex as the biggest upside risk to that net-supply number.
The acceleration is visible in the deal tape. The five hyperscalers issued $121 billion in US corporate bonds last year, versus an average of $28 billion per year between 2020 and 2024, according to BofA Securities. Globally, Alphabet, Amazon, Meta, Microsoft, and Oracle have issued $159 billion of bonds this year, up from $108 billion in all of 2025 and just $17 billion in 2024, according to Dealogic data. BofA expects the Big Five to borrow roughly $140 billion annually over the next three years, a pace that could exceed $300 billion a year — putting them on par with the Big Six banks' expected average issuance of $157 billion annually.
More supply to fund AI could make the five hyperscalers some of the largest issuers in the IG index, BofA analysts wrote in a January research note.
The widening extends beyond the hyperscalers themselves. AI-linked debt issuance by companies outside the group — spanning software, semiconductors, and data-center financing — has already totaled roughly $412 billion in 2026, according to Goldman Sachs. Add that to the hyperscaler wave and the total capital absorbed by the AI buildout approaches $900 billion for the year. That is capital that is not available to absorb Treasury supply at prior pricing.
The Market's Message: Absorption Is Getting Harder
The bond market has begun to push back. Over a recent stretch of weeks, the investment-grade market struggled to absorb a combined $75 billion of issuance from Nvidia, SpaceX, and Amazon. While Nvidia and SpaceX borrowed at reasonably low rates, their newly issued bonds quickly slumped in the secondary market, disappointing investors who often flip such deals. Amazon, meanwhile, had to pay unusually steep rates by its standards to complete its debt sale — a sign that investors were no longer happy to hand money to AI issuers at any price.
Credit protection is pricing the same caution. The cost to insure hyperscaler debt through credit-default swaps has risen since June 2025, most notably for Oracle, whose five-year CDS more than tripled after its September deal, according to MUFG. Oracle is where the risk concentrates: its fiscal 2026 capital expenditure reached $55.7 billion against operating cash flow of roughly $32 billion, and the company plans to raise around $40 billion in debt and equity during fiscal 2027. Bank of America projects Oracle running negative free cash flow until 2029, leaving little capacity for more debt. Oracle shares are down 17.91% year to date as of Wednesday's close.
Goldman Sachs credit trader Jeffrey Papai calls 2027 the real stress test, with a scenario pointing toward roughly $340 billion of senior hyperscaler and chip issuance in the coming year, before another large layer of data-center and structured chip financing is added on top. Limited hyperscaler maturities mean there is far less natural recycling of investor capital than in sectors such as banks — the market is dependent on fresh demand for fresh supply.
The Treasury has not been idle. In an effort to curb the sharp increase in long-end borrowing costs, the Treasury Department said it would roughly double its bond buyback program to at least $4 billion per operation starting in September, targeting longer-dated maturities to smooth term-premium pressure. The intervention briefly tamed yields, but the relief proved short-lived — a reminder that buybacks can manage the shape of issuance, not eliminate the underlying competition for long money.
Cyclical or Structural: This Is a Regime Shift, Not a Wave That Reverts
The central judgment: the AI-driven demand for long-duration capital is structural, not cyclical. A cyclical call would require a mean-reverting driver — an inventory cycle, a liquidity impulse, a temporary supply glut — and evidence that the pressure dissipates on its own. None of those apply here. The driver is a technology platform shift that changes the capital intensity of the entire economy: compute demand, power infrastructure, and data-center construction are multi-year commitments, not a quarter's worth of restocking.
Three pieces of evidence support the structural read. First, the spending is guided, not guessed — Amazon's $200 billion and Alphabet's $175 billion to $185 billion are management commitments, not cyclical forecasts. Second, the financing is shifting structurally toward debt: the hyperscalers' debt-funded share of capex is rising from 26% to 33% to a projected 35%, a deliberate balance-sheet choice, not a one-off funding gap. Third, the demand base is widening beyond the hyperscalers: the roughly $412 billion of AI-linked debt from non-hyperscaler issuers shows the capital intensity is spreading through the supply chain.
That said, there is a cyclical leg riding on top of the structural shift: the pacing of issuance. Deal flow clusters in windows of market openness, and the recent secondary-market slumps suggest the near-term pipeline could slow into year-end, as Goldman noted. If issuance pauses, term premia could compress and the 10-year yield could drift lower on technicals. But a pause is not a reversal. The structural pressure returns as soon as the next funding window opens, because the capex commitments do not disappear.
The Adversarial Case: Cash Flow Is Enormous, and the Bulls Have a Point
The strongest counter-thesis rests on the cash flow behind these balance sheets. Bank of America argues that the big five hyperscalers' debt-to-cash ratio actually improved, dipping from 0.94 to 0.75, and projects operating cash flow jumping to $1.1 trillion by 2029, a 95% increase. On that read, this is balance-sheet optimization by cash-rich businesses, not a solvency scare. The hyperscalers are among the most profitable companies in history, and their bonds carry investment-grade ratings with spreads that, while wider, remain far from distressed.
The bull case is credible at the aggregate level, and it is the reason this is a pressure on yields rather than a crisis. But it papers over two problems. First, the aggregate masks wide dispersion: Oracle's negative free cash flow and tripled CDS show that the marginal issuer is not Microsoft. Bond markets price the marginal deal, not the median balance sheet. Second, the cash-flow argument assumes the AI revenue arrives on schedule. Sequoia Capital's David Cahn calculated an annual revenue gap of approximately $600 billion between what hyperscalers are spending on AI infrastructure and what the AI ecosystem is actually generating in sales — a gap that is widening in 2026 as capex accelerates faster than revenue projections. If that gap persists, the debt must be refinanced, not retired, and the duration demand becomes permanent rather than transitional.
There is also a foreign-demand wrinkle the bull case tends to underweight. US Treasuries have long relied on official-sector buyers abroad — central banks and sovereign funds — to absorb a large share of issuance. Those flows have plateaued in recent years as reserve managers diversify and geopolitical considerations reshape portfolios. If foreign official demand for 30-year bonds has peaked, as several fixed-income strategists have argued, then the marginal long-duration buyer is increasingly a domestic institution choosing between a Treasury and a hyperscaler bond. That is precisely the choice that lifts the term premium.
Who Wins, Who Loses, and What to Watch
The implications split cleanly across asset classes. Long-duration Treasury holders are the most directly exposed: if AI issuance keeps absorbing a share of long money near two-thirds of new long-duration government supply, the term premium embedded in the 10-year and 30-year yields has a structural floor even when growth data softens. Investment-grade credit investors face a double-edged market: more high-quality paper to buy, but wider spreads and a secondary market that has shown it will punish deals priced for perfection. Banks and mortgage markets are the indirect channel — as long-duration corporate paper and Treasuries compete for the same investors, the clearing rate for fixed-rate mortgages and long-term corporate loans rises with them.
The short-term view is technical and can cut either way: issuance windows close, secondary spreads widen, and the 10-year yield can ease on a quiet deal calendar. The medium-term view is fundamental: the debt-funded share of AI capex keeps rising, and the supply pipeline into 2027 is already visible at roughly $340 billion for senior hyperscaler and chip paper alone. The long-term view is structural: if AI proves to be a general-purpose technology on the scale of electrification, the economy's demand for long-duration capital has shifted to a higher regime, and the era of structurally suppressed term premia is over.
Three signals would falsify the structural call. First, if hyperscaler debt issuance falls back toward the 2020-2024 average of $28 billion a year while capex guidance is cut, the pressure was cyclical after all. Second, if the 10-year Treasury yield trades sustainably below 4% while issuance remains above $300 billion annually, the transmission mechanism is broken. Third, if AI revenue closes the $600 billion gap Cahn identified and hyperscaler free cash flow turns decisively positive across the group, the debt wave becomes self-financing and the crowding pressure lifts. Watch the bid-to-cover ratios on long-end Treasury auctions and hyperscaler spreads versus Treasuries as 2027 supply comes into view — Goldman Sachs identifies the latter as the metric to watch.
The bond market spent years worrying that Washington would crowd out the private sector. The irony of 2026 is that the private sector is now crowding out Washington — and the price is being paid by anyone holding a 30-year mortgage or a long Treasury.
Explore more exclusive insights at nextfin.ai.
