The bond flood AI built is now the tide rising against it

When more sellers suddenly flood a market with the same product, prices have to fall to move the goods. That is precisely what is happening in the bond market right now.

The US government is pouring out Treasury bonds to plug a ballooning fiscal deficit, while big tech companies are issuing corporate bonds to build AI data centers and power grids. One side is borrowing to keep the government running; the other is borrowing to avoid falling behind in the AI race. The result is a rare spectacle: the US government and cash-rich tech giants competing in the same bond market for the same pool of investor money.

The number of sellers has grown, but buyers have not kept pace. Foreign central banks, once reliable and deep-pocketed buyers of US Treasuries, have faded in significance, and the Federal Reserve is no longer stepping in as a large-scale buyer of long-term government debt. Government bonds and corporate bonds are now effectively fighting over the same limited pool of long-term capital.

To sell, prices must come down. In the bond market, that discount tag is the interest rate — as bond prices fall, yields rise.

The evidence that oversupply is pushing real rates higher becomes clearer when you examine why long-term yields have climbed. Long-term rates rise when inflation expectations or Federal Reserve policy rate forecasts shift, but they also rise when investors demand greater compensation for holding long-dated bonds over time. That extra compensation is the term premium.

Lately, the second force has dominated. An analysis by Mirae Asset Securities of factors driving the US 10-year yield shows that since July, rates have risen 15.9 basis points (1 bp = 0.01 percentage point), with the term premium accounting for 13.0 bp of that move and rising inflation expectations contributing 2.9 bp. The market is effectively asking: "Is it really safe to hold this much newly issued debt for decades?"

The first source of concern is the US government itself. Total US national debt has surpassed $40 trillion for the first time in history. As of Tuesday, the Treasury Department put the figure at $40.05 trillion — of which $32.27 trillion is held by the public and private markets, and $7.78 trillion represents intragovernmental debt. The debt stood at $19.95 trillion in early 2017, meaning it has more than doubled in less than a decade.

What makes the situation more alarming is the self-reinforcing dynamic between debt and interest rates. As debt grows, so does the government's interest bill. When existing bonds mature and are refinanced at higher rates, that cost rises further. Higher interest expenses widen the fiscal deficit, which in turn requires even more bond issuance to cover.

The numbers bear this out. According to the Treasury Department, interest costs for the current fiscal year through the end of July reached $1.17 trillion, up about 15 percent from the same period last year. The vicious cycle — more debt, higher rates, higher interest costs, wider deficits, more issuance — is no longer a distant forecast.

The second source of pressure is the private sector, and specifically AI. For the past several years, big tech companies used their enormous cash flows to fund much of their AI investment from their own balance sheets. But as spending on data centers, semiconductors and power infrastructure has jumped to a new level, debt financing is now being actively deployed. The funding model is shifting rapidly from equity to borrowing, even as AI capital expenditure continues to drive growth.

Data compiled by NH Investment Securities show that corporate bond issuance by hyperscale cloud providers averaged $28 billion a year between 2020 and 2024, then surged to $121 billion in 2025. By August this year, issuance had already reached $165 billion, with the full-year total projected at $400 billion.

A broader tally of AI-related bond issuance by Mirae Asset Securities puts this year's figure at $489 billion, already exceeding last year's full-year total of $322 billion. The scope of each count differs, but the direction is the same: the financing model for AI investment is shifting rapidly from cash-funded data centers to debt-funded ones.

This investment is also difficult to slow with a modest rate increase. The AI industry operates close to a winner-take-all dynamic. If companies calculate that the cost of falling behind rivals outweighs the cost of borrowing, they will keep investing even at higher rates.

That is where the paradox of the AI boom begins. The more companies pour into AI infrastructure, the more corporate bonds flood the market. US Treasuries and investment-grade tech bonds are not identical products, but they ultimately compete for the same long-term investor capital. When corporate bonds offer higher yields, Treasuries must follow to attract buyers.

Those higher rates then turn on AI companies themselves. Equity valuations rest on the present value of future cash flows, and when rates rise, the discount rate applied to those flows rises with them. Companies expected to generate enormous profits years from now are hit harder than those generating cash today. That is why growth stocks — AI, semiconductors, software and platform companies — are particularly sensitive to long-term rates. The capital AI raised to fund its growth is now eroding the valuations of AI companies themselves: a self-cannibalizing loop.

The visible corporate bonds are not the whole picture. According to Kiwoom Securities' monthly asset allocation report, the four major big tech companies — Alphabet, Amazon, Meta and Microsoft — had signed but not yet commenced leases totaling $904 billion, and purchase commitments of $1.52 trillion.

Data center leases signed before the rental period begins, and contracts for chips and power before delivery, may not appear on the balance sheet as liabilities until the obligations are triggered — they sit in the footnotes. This is hidden debt.

It cannot be treated as equivalent to financial liabilities due immediately. But if AI demand and monetization fall short of expectations, the calculus changes. Future revenue gets pushed back while already-committed costs for data centers, chips and power become real. Credit default swap premiums on AI-related companies remain elevated — a signal that bond markets are not entirely dismissing this risk.

With supply growing this fast, who absorbs the bonds matters enormously. That is where Japan enters the picture. Japan's 10-year government bond yield recently climbed to 2.93 percent, its highest level since 1996, as a weaker yen and rising oil prices added to inflation and rate pressures.

As Japanese government bond yields rise, Japanese investors have less incentive to take on currency risk by buying long-dated US Treasuries. Japan, traditionally one of the largest buyers of US government debt, is moving into an environment where it can no longer absorb American bonds as readily as before. If outright sales of US Treasuries to defend the yen become more likely, the supply-demand imbalance at the long end of the US market could worsen considerably.

The recent use by US and Japanese authorities of the Foreign and International Monetary Authorities Repo Facility — which allows foreign central banks to borrow dollars using US Treasuries as collateral rather than selling them outright — reflects an awareness of exactly this pressure.

The core problem is that supply is rising while the demand side weakens. The US government and AI companies are issuing more bonds, but traditional large buyers, including Japan, are less able to absorb them than before. Long-term bond investors have every reason to demand higher yields.

What can the government do? One answer the Treasury has offered is a buyback program — repurchasing bonds already in circulation, effectively stepping in as a buyer in a market short of willing hands.

The Treasury recently doubled the per-operation buyback cap for bonds with 10 to 20 years and 20 to 30 years of remaining maturity, raising it from $2 billion to at least $4 billion per operation. Taking the remaining schedule into account, the additional buyback capacity amounts to at least $14 billion, expanding the long-end buyback ceiling from $16 billion to at least $30 billion.

Revising the buyback plan just two weeks after releasing the quarterly refunding announcement is unusual. Markets have read the move as a signal to investors who had built short positions in 30-year Treasuries — a message that the Treasury can step in if needed, intended to discourage further bets on rising rates.

Long-term Treasury yields did fall immediately after the announcement. The news that the Treasury would at least double its long-end buybacks was read as a strong signal of intent to stabilize the long-dated bond market.

Even so, the firepower is too small to reverse the market's direction. The Federal Reserve's Operation Twist in 2011 and 2012 involved buying long-term Treasuries and selling short-term ones over 15 months at a total of $667 billion — roughly $44 billion a month on average — in a deliberate effort to reduce the duration of bonds held by the private sector.

The current Treasury buyback is far smaller in absolute terms, and the US Treasury market itself has grown more than three times since then. KB Securities estimated in its global insight report that the expanded buyback would reduce the term premium by only 1 to 2 basis points.

There is also one problem today that did not exist then: $40 trillion in debt. Shifting issuance toward short-term bills to ease pressure on the long end would provide near-term relief, but it would shorten the average maturity of government debt, making interest costs more sensitive to the Federal Reserve's benchmark rate. With interest costs already running at $1.17 trillion through July this year, a prolonged period of high policy rates would rapidly compound the fiscal burden.

The Fed cannot set rates based solely on the government's interest bill. If markets begin to suspect that $40 trillion in debt is preventing the central bank from tightening sufficiently even as inflation rises, the effect could backfire. Long-term bond investors would demand a higher premium for future inflation and monetary policy uncertainty, pushing inflation expectations and the term premium higher — and long-term rates with them.

The final fork in the road is whether an inflation problem compounds the supply-demand problem. Much of the rise in long-term yields so far can be explained by the surge in government and corporate bond issuance and the resulting increase in the term premium.

Put another way, inflation itself has not yet risen to a worrying degree. US core consumer prices rose just 0.2 percent month-on-month in July, and producer prices were flat. Long-term inflation expectations have also held steady near 2.2 percent since July.

Yet rates have soared to their highest level in 19 years. If Middle East tensions persist and rising oil prices push inflation expectations higher, the picture changes entirely.

The Federal Reserve faces a dilemma. The government wants lower rates to manage $40 trillion in debt and astronomical interest costs. Companies need cheaper financing to sustain massive AI investment. President Donald Trump has publicly called for rate cuts. But if inflation reaccelerates, the Fed would have no choice but to hold rates high or raise them further.

At that point, supply-management tools like the buyback lose their force — their reach was limited to begin with. Buybacks work by reducing the supply of long-dated bonds in circulation, compressing the term premium. If inflation forces the Fed to raise its benchmark rate, the other half of long-term yields — expectations for future short-term rates — rises entirely beyond the Treasury's reach.

The Treasury also lacks the Fed's ability to create money. To buy back bonds, it must raise cash by selling new ones. The debt the market holds does not shrink; only its maturity shortens. In an inflationary environment, that maturity compression becomes a liability in itself.

The result could be a phase in which both the policy rate and the term premium rise simultaneously — while the supply of government and corporate bonds remains undiminished. The US government cannot stop issuing to cover its deficit, and big tech cannot stop issuing to stay competitive in AI. The forces pushing rates higher are multiplying while the tools to contain them are growing blunter — the combination markets fear most.

"Even if big tech cannot stop investing in AI, capital providers can," said Lee Eun-taek, a director at KB Securities. "If the 10-year US Treasury yield breaks above 5 percent, capital providers may pivot toward locking in safe returns." He added that the common thread running through the Great Depression of the 1930s, the stagflation of the 1970s and the dot-com bubble collapse of the 2000s was a sustained upward trend in interest rates. "When rates rise, massive capital flows will shift," he said.