July ended on Friday with a picture that contains no contradiction at first glance and almost nothing but contradiction at second. The S&P 500 closed at 7,489.72, up 0.70 percent. The Dow Jones Industrial Average added 0.53 percent to 52,485.03, and the Nasdaq Composite rose 1.00 percent to 25,373.85. Amazon jumped roughly 15 percent after earnings; Apple fell about 7 percent. That is the headline: a friendly end to the month, carried by the largest technology companies in the world.
Two numbers from the same session do not fit that story. The yield on the ten-year Treasury note climbed to 4.73 percent, its highest level since January 15, 2025. The thirty-year bond finished at 5.28 percent, a level the American bond market last saw before the financial crisis. And the Russell 2000, the index of 2,000 smaller American companies, fell 0.50 percent that same day to 2,931.34. While the top of the index celebrated, the belly of it gave way.
The obvious explanation is that rising rates hurt small companies more than large ones. That is true, and it is not enough. Because in the same month that the price of money reached its highest level in a year and a half, something quietly changed that rearranges the relationship between the equity market and the bond market. The companies building artificial intelligence have become the largest borrowers in the American capital market. For two years the AI buildout was an equity story. This year it became a credit story. And credit has a price that no board of directors sets.
Three numbers tell the whole story: 28, 121, 159
Between 2020 and 2024, the five large American cloud and platform companies — Amazon, Alphabet, Meta, Microsoft and Oracle — issued an average of roughly $28 billion in corporate bonds per year, combined. For companies of that size, that is close to nothing. These were businesses with net cash that tapped the capital markets opportunistically rather than out of need. Anyone analyzing Microsoft or Alphabet never had to think about their refinancing.
In 2025 that figure rose to $121 billion, more than four times the prior average, in a single year. And by the middle of 2026 those same five companies had already raised roughly $159 billion in the bond market, an increase of about 47 percent over the full prior year — meaning they borrowed more in six months than in the five years before 2025 combined. Add Nvidia, and the six largest technology companies had issued roughly $244 billion globally by mid-July.
That sequence — 28, 121, 159 — is the real story of the year, and it has stayed off the front pages remarkably well, because bond markets look boring. It describes a change in financing, not a swing in the business cycle. Through 2024, the data center buildout was paid for out of operating cash flow. That was the strongest argument the optimists had, and it was a good one: a company investing out of its own earnings can stop at any time, and no creditor gets a say. That argument is now disappearing.
The deals that rebuilt the market
The individual transactions are more striking than the total. In September 2025, Oracle issued $18 billion in bonds. Meta followed in October with $30 billion — the largest single investment-grade bond sale on record that was not tied to a merger. In November, Alphabet raised $17.5 billion and Amazon $15 billion.
In 2026 the scale moved again. In March, Amazon brought a transaction totaling roughly $54 billion, split between dollar and euro tranches. The $37 billion dollar portion alone was the fourth-largest corporate bond sale in American market history, joined by €14.5 billion. Demand was overwhelming: the dollar tranche drew orders of about $126 billion, and the euro tranche saw peak demand above €35 billion. Meta returned in late April with $25 billion. Alphabet raised roughly $32 billion across the year, including the first hundred-year sterling bond ever issued by a technology company. Nvidia placed $25 billion in June. And on July 7, Amazon announced a further offering of at least $25 billion.
It is worth pausing on what a hundred-year bond means in this context. A company whose principal product did not exist three years ago is borrowing money for a term that exceeds the life expectancy of nearly everyone alive today — in order to buy graphics processors whose useful life is measured in years. That need not be a mistake. But it is a wager on the durability of earnings that barely exist yet.
The moment technology passed the banks
By October 2025, the volume of debt connected to the AI buildout had grown to roughly $1.2 trillion. That made technology the single largest segment of the American investment-grade market, displacing US banks from the top spot; in the JPMorgan US Liquid Index the sector accounts for about 14 percent. According to Dealogic data, technology accounts for roughly 20 percent of all dollar-denominated investment-grade issuance in 2026, a record share.
This is more than a statistic for bond fund managers. For decades the large banks were the dominant issuers in the American bond market. Their issuance is a byproduct of a regulated business model, it is reasonably predictable, and it is backed by a business that has existed for generations. The six largest American banks together average roughly $157 billion of issuance per year. For the hyperscalers, Bank of America analysts project $140 billion to $175 billion annually over the next three years — with the explicit possibility that it exceeds $300 billion.
One scenario illustrates the force of this shift better than any forecast: if these companies financed even 20 percent of their AI capital budgets with debt, Alphabet would rise from 67th to 8th place by weight in the investment-grade index. A company that the bond market simply did not notice until recently would become one of its heavyweights.
Why this pushes the interest rate itself higher
Here the circle closes back to Friday’s 4.73 percent, and here lies the most interesting part of the story analytically. Bond markets work like any other market: price is set by supply and demand. When several hundred billion dollars of new corporate paper arrives within a few quarters on top of the already enormous financing needs of the US Treasury, and it all competes for the same pool of capital, the consequence is not neutral. The buyers of these securities — insurers, pension funds, bond funds — have a finite budget for long-duration assets. Anyone who wants more of it has to bid more.
Confirmation came in the form of a forecast revision from UBS. The bank raised its expectation for total US investment-grade issuance in 2026 to $1.8 trillion, up from $1.725 trillion — roughly 22 percent above the prior year. Its forecast for the technology sector alone rose from $300 billion to $360 billion, with hyperscaler supply expected at $230 billion to $240 billion. The stated trigger for the revision was a $145 billion increase in the capital spending guidance of those same companies, whose total capex UBS now puts at roughly $770 billion for this year. Morgan Stanley sits well above that, estimating $400 billion of hyperscaler issuance in 2026 against $165 billion last year.
What has emerged is a mechanism that did not exist in this form before: the AI buildout has grown large enough to move the interest rate at which the AI buildout is valued. Every increase in the capex guidance raises the supply of bonds; every additional bond weighs on prices and lifts yields; and every higher yield lowers the present value of precisely those future earnings used to justify the investment. This is not a conspiracy and not proof of a bubble. It is simply the scale an industry has reached when its capital needs have become macroeconomically relevant.
The central bank supplies the other half of the bill
The other half comes from monetary policy, and it turned sharply this week. The Federal Reserve left its policy rate at 3.50 to 3.75 percent on July 29, but the vote was the real message: 9 to 3. Beth Hammack of Cleveland, Neel Kashkari of Minneapolis and Lorie Logan of Dallas all voted to raise the target range by a quarter point. It was the most divided FOMC vote since 2016, and three dissents pointing the same direction are rare in the history of the institution.
The market took the point. The implied probability of a September hike rose, depending on the data source and the moment, to somewhere between roughly 57 percent and above 80 percent. The width of that range is itself information — it shows how unreliable expectations have become. The causes are known: the core rate of the Fed’s preferred price index stands at 3.3 percent, and oil rose more than 20 percent in July, with Brent recently above $88. For an investor holding corporate bonds, both forces push in the same direction: a higher risk-free base and a growing supply of paper.
Who is already paying the higher rate
The gap between the top of the index and the rest of the economy can be quantified unusually precisely right now. The average American thirty-year fixed mortgage rate stood at 6.76 percent at month-end, the highest in 51 weeks; Freddie Mac had reported 6.58 percent for the week ending July 23. The large homebuilders — Lennar, D.R. Horton, PulteGroup and NVR — all declined over the course of the week. That is the channel through which 4.73 percent on the ten-year note reaches household budgets, at a spread of roughly 1.8 percentage points.
The same logic runs through the regional banks, whose loan books are full of commercial real estate priced off the long end, and through the smaller companies in the Russell 2000, a large share of which borrow at floating rates rather than the fixed thirty-year terms available to Amazon. This is the practical case for looking at what a broad index fund actually owns: an S&P 500 position today is, to an unusual degree, a position in the handful of balance sheets that can still afford these rates. Meanwhile a long-end yield above 5 percent offers something that has not existed for most of the past fifteen years — a serious alternative to equities inside the same portfolio. For US investors the tax treatment matters here in a way it rarely does elsewhere: bond coupons are taxed as ordinary income at rates up to 37 percent, while qualified dividends and long-term capital gains top out at 20 percent, which argues for holding taxable bond exposure inside a tax-deferred account rather than a brokerage account. Treasury interest, unlike corporate coupons, is exempt from state and local income tax.
The case against this reading
The counterarguments are strong enough that they should not be treated as a formality. First, these are excellent credits. Microsoft, Alphabet and Amazon carry interest coverage ratios an average industrial company could only dream of, and the sums raised remain modest relative to their operating earnings. A bond market that suddenly receives more paper from borrowers of this quality is not in trouble — it is getting better collateral than it is used to.
Second, and this is the strongest objection: Amazon explicitly committed, alongside its July transaction, to issuing no further debt during the remainder of 2026. If the sector’s largest issuer declares its needs met, extrapolating the recent trend forward may simply be wrong. Third, demand has been enormous — an order book of $126 billion for a $37 billion deal is not a sign of market stress but the opposite. That said, demand at Amazon’s most recent offering was reported to have cooled relative to the March sale, and details of exactly that kind are where a turn shows up first.
Fourth, a substantial part of the yield increase comes from oil and monetary policy rather than bond supply. The issuance wave is an amplifier, not a sole cause. And fifth, the spread of the forecasts is itself a warning against confidence in either direction: when UBS expects $230 billion to $240 billion and Morgan Stanley expects $400 billion, nobody knows.
What matters starting Monday
The coming week brings the next data point that will help decide the September meeting: the July employment report is due Friday. Consensus expects payroll growth of roughly 88,000 and an unchanged unemployment rate of 4.2 percent. A materially stronger print would support the three dissenters and push yields further; a weak one would be the first real relief the bond market has had in weeks.
More important than any single forecast is the metric that should survive this month. For two years the standard question put to every technology company was: how large is the capital budget? Then, certainly since this summer, came the questions about free cash flow and assumed useful life. A third now joins them, and it is the most uncomfortable: what share of this investment is paid for out of earnings, and what share is borrowed? Because the self-funded portion can be halted at any time. The borrowed portion has to be serviced, in good quarters and bad, at a price set by the bond market rather than by management.
So anyone wanting to know when the index will start to feel the interest rate no longer needs to watch only the earnings calendar. They need to watch how often the names at the top of the index appear on the bond market’s issuance calendar. This year there have been 159 billion reasons to look closely. On Friday, long-term money cost as much as it did before the financial crisis. Both things at once are new, and both things at once are the real reason the smaller stocks fell on the day the index set its high.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
