Companies plan $215B of bonds in September. Read who absorbs it.

Wall Street dealers expect about $215 billion of US investment-grade bonds to price in September, against a September record of $207.5 billion set last year, according to a Bloomberg poll reported on 3 September 2026. They expect it while the 30-year Treasury pays 5.25% and has closed above 5% on 55 days this year, the most in any year since 2006.

Those two facts are one fact. Treasurers are pulling 2027 funding into 2026 because they have decided borrowing gets more expensive from here, not less.

The biggest new borrowers in that market build AI data centers. Hyperscalers sold roughly $132 billion of bonds through 31 July 2026, more than the $93 billion they sold in all of 2025 and nearly four times their 2020 to 2024 average, by Vanguard's count. Technology now sits at 10.2% of the investment-grade index, up from 9.5% a year earlier, per Breckinridge.

Someone stands on the other side of each of those deals, and increasingly it is an index fund. If your safe money sits in a total bond market fund or a corporate bond ETF, that someone is you. The compensation for it: under 0.8 percentage points over Treasuries.

The long end stopped cooperating

Start with the price of money at the far end of the curve, because everything else in this story is a response to it.

Treasury's own daily par yield curve put the 30-year at 5.27% on 1 and 2 September and 5.25% on 3 September. The 10-year sat at 4.77% on the third. The 30-year touched 5.34% in mid-August, its highest since 2007 and about 10 basis points from a 22-year high, and it has now closed above 5% on 55 days since January. Bloomberg's framing is the useful one: this is the worst stretch for the long bond since 2006.

Two things did not fix it. The first is buybacks. Treasury doubled the size of its buyback operations this summer, which we worked through here, and John Briggs at Natixis called them "a drop in the bucket" for controlling yields. The second is the global bid. Japan's 10-year touched 3% for the first time in three decades, and Japanese money that used to fund the American long end started staying home.

Then there is the Fed. Traders priced about 17 basis points of tightening into the 15 to 16 September meeting as of 1 September. A cut is barely in the distribution. When the front end might go up and the long end refuses to come down, the treasurer's calculation is simple.

$215 billion, most of it after Labor Day

US Labor Day falls on 7 September this year, so the bulk of that $215 billion has to arrive in the three weeks after it, straddling the Fed meeting.

Issuers did not entirely wait. High-grade sales reached $8.3 billion through Wednesday 2 September, the highest for the pre-Labor Day week since at least 2019. That is a small number in absolute terms and a loud one in context: the week before the US holiday is traditionally dead.

The forecast is not unanimous. Dealers polled by Bloomberg landed near $215 billion, individual estimates ran from $175 billion to $250 billion, and Bank of America put it at roughly $190 billion on the view that some large tech issuers will sit out after recent deals. Take the range seriously. Supply forecasts miss.

What is not in dispute is why anyone would issue into this. Average yields on US high-grade notes are above 5.5%, the highest in more than two years, while spreads have held below 0.8 percentage points. Tom Murphy, who runs investment-grade credit at Columbia Threadneedle, put the trade in one sentence: "Boy, if I was a CFO or treasurer and had something to do in 2027, I'd probably pull it forward into 2026."

Moshe Tomkiewicz at Mizuho Americas described the same logic from the credit side: "the devil you know" beats waiting while spreads are still tight.

Read that as a rate call. A treasurer who believed the long end was heading down would wait six months and save real money. A calendar this size, arriving this fast, says the people who fund these companies have stopped believing that.

The marginal borrower is a data center

Hyperscaler borrowing is the part of this that changed fastest.

The shape matters more than any single bar. These are companies that funded themselves out of operating cash flow for a decade. FactSet's read on the same shift is that incremental debt covered 9% of hyperscaler capex in fiscal 2024 and 32% in the twelve months to mid-2026, against aggregate hyperscaler capex running above $690 billion for 2026 and total debt near $700 billion.

Alphabet is the clean example. In the June quarter it spent $44.9 billion on property and equipment, double the year before, and reported negative free cash flow of $5.9 billion, the first in its history, while raising full-year capex guidance to $195 billion to $205 billion. In August it came to the bond market for $25 billion across ten tranches with maturities out to 40 years and took roughly $115 billion of peak orders.

Forty-year paper against equipment on a much shorter clock is the tension underneath all of this. The chips being financed depreciate on schedules measured in single-digit years, and the vendor commitments behind them are already enormous: Nvidia alone carries $279 billion of supply commitments, a number that only makes sense if the buildout keeps being funded.

Note the accounting caveat, because it cuts against a tidy story. FactSet counts around $820 billion of aggregate lease obligations that never appear as balance-sheet debt. Bond issuance is the visible part of the financing, not all of it.

What the index buyer is actually paid

Here is the arithmetic nobody puts on the fund page.

The average investment-grade note yields above 5.5%. The spread over Treasuries is under 0.8 percentage points. So roughly 85% of that yield is the US government paying you, and roughly 15% is the compensation for taking corporate credit risk, sector concentration, and forty-year duration on assets financing an AI buildout. Breckinridge had the index option-adjusted spread at 74 basis points at the end of the second quarter, after 14 basis points of tightening in the quarter. Both readings agree: credit is not being paid much right now.

Meanwhile the composition of what you hold keeps moving. Technology went from 9.5% to 10.2% of investment-grade market value in a year, which sounds slow until you notice it is climbing on new supply rather than on price, and that the same names dominate the equity index most people hold alongside it. Concentration was an equity-market problem. It is arriving in the sleeve people buy as the diversifier.

None of that makes these bad bonds. It makes them bonds whose risk you should know you own.

What would make this read wrong

Four things, honestly stated.

1. The market keeps absorbing it. Spreads have been tight all year through repeated supply waves. Demand for Alphabet's August deal ran more than four times the size. If that holds through September, the front-running story is just a story about a busy month.
2. BofA might be right. At $190 billion, September is a heavy month and not a record, and the reason given is that big tech sits out. That would weaken the "AI is the marginal borrower" claim for this specific month.
3. The credit is good. Investment-grade tech is mostly high-rated with large cash balances, and 10.2% makes technology the third-largest sub-sector, not the market. A rising weight is not a default.
4. The rate call could be wrong. If the Fed holds and the long end rallies into year-end, everyone who pulled 2027 funding into September simply overpaid for certainty. That costs issuers, not bondholders.

What to watch now

The 15 to 16 September Fed decision, and whether 17 basis points of priced tightening turns into a hike. The final September issuance print against last year's $207.5 billion record. Whether large tech issuers show up or sit out, which decides between the two readings above. And technology's index weight at year-end: 10.2% climbing on supply is the number that tells you whether a bond index fund is still doing the job you bought it for.

Sources

• Rising Yields Seen Pushing Companies to Sell Bonds Sooner, Bloomberg via FA-Mag, 3 September 2026
• US 30-Year Bond Enters September on Its Worst Stretch Since 2006, Bloomberg via Yahoo Finance, 1 September 2026
• Daily Treasury Par Yield Curve Rates, September 2026, US Department of the Treasury
• The AI buildout comes to the bond market, Vanguard, 19 August 2026
• Q3 2026 Corporate Bond Market Outlook, Breckinridge Capital Advisors, 7 July 2026
• Hyperscalers Tap External Financing as AI Capex Outruns Cash Flow, FactSet, 23 July 2026
• Google Raises 2026 CapEx Guidance Again Above US$200B; Free Cash Flow Turns Negative for First Time, TrendForce, 23 July 2026
• Alphabet Raises $25 Billion From Sought-After Bond Sale, Bloomberg Law, August 2026

This is not financial advice.