AI borrowers now pay about 115 basis points over Treasuries, versus 78 for the rest of high-grade debt

Investment-grade bond buyers are demanding extra pay to lend to the companies building out artificial intelligence, even as they compete hard for bonds from everyone else, Reuters reported on Tuesday. Spreads on AI-related issuers are running around 115 basis points over Treasuries, according to Goldman Sachs data cited by Reuters, against 78 basis points for the broad investment-grade market on ICE BofA figures.

The issue is supply, not credit risk

The fund managers Reuters spoke to were clear that they do not expect hyperscalers to default. The problem is volume. Goldman expects gross debt issuance from hyperscalers to hit a record $420 billion next year, up 60% from its 2026 estimate. Working that backwards puts this year's figure at roughly $260 billion. For scale, total U.S. corporate issuance through August was $1.9 trillion, up 30% from a year earlier, according to the Securities Industry and Financial Markets Association.

The split shows up deal by deal. Alphabet had to offer a large concession to get its August bond sale done, according to a BNY research note cited by Reuters. Insurance broker Aon's $13.5 billion acquisition financing this month drew $65 billion of orders, close to five times the amount on offer, and the pricing on its 30-year piece tightened by 35 basis points during the sale. BlackRock's Russell Brownback described some AI deals as double-A credits pricing closer to triple-B levels.

What 37 basis points is worth

The gap between 115 and 78 basis points is 37 basis points, or 0.37 percentage points a year. On $10 billion of bonds, that is $37 million of extra interest every year, or $370 million over a 10-year bond's life, for borrowers whose balance sheets are among the strongest in the market.

In yield terms, using the 10-year Treasury as a rough benchmark (it closed Monday at 4.96% on Treasury's daily curve, and was near 4.97% on Tuesday per CNBC), a 115-point spread puts a 10-year AI bond near 6.1%. At the market-wide 78 points it would be near 5.7%. Actual spreads are measured against Treasuries of matching maturity, so treat those as ballpark figures. The direction is what matters: the companies with the most cash are paying up because there is so much of their paper coming.

Who this actually touches

• Holders of high-grade bond index funds. Most of these funds weight holdings by the amount of debt outstanding, so every big AI bond sale automatically raises their share of AI exposure. That is the same concentration issue active managers told Reuters they are trying to manage. Wellington's Loren Moran said some institutions are nearing single-name limits once debt from data-center financing vehicles is added back to the parent company.
• Everyone else raising debt. The other side of the trade is good news for non-AI borrowers. With investors "starved for anything ex-hyperscaler," in Moran's words, industrial, financial and pharmaceutical issuers are getting oversubscribed deals at tight spreads.
• Equity traders in the AI names. A wider spread is a real cost of capital. Companies keep borrowing at these levels because they expect AI projects to earn well above that cost, BlackRock's Brownback said. If returns on those projects disappoint, the bond market is where the pressure would show first. Breckinridge's Nick Elfner told Reuters some hyperscaler deals have already traded poorly after pricing.

What traders are watching

The next big test is the next surprise deal. Elfner said issuance arriving months after a previous sale, often at wider spreads, undermines confidence and raises what investors want next time. Oracle's bonds have at times shown that pattern, according to Marty Fridson of Income Securities Investor. For rate context, see our 10-year Treasury chart and our story on Goldman's talks to buy Palmer Square, another sign of how much capital is chasing credit.

Sources: Reuters via Yahoo Finance, citing Goldman Sachs, SIFMA, ICE BofA and BNY; U.S. Treasury; CNBC. Dollar and yield figures in the "37 basis points" section are our own arithmetic from those numbers. This is market information, not investment advice.