--- title: "The bond market has started charging more for AI, even to the safest borrowers" description: "US technology companies have issued at least $220 billion of debt this year to fund data centers, against $12.5 billion in the same stretch of 2025. The money is still there, but the price has changed: technology now trades wider than the investment-grade market it used to trade through, and demand for each successive deal has thinned." category: "Companies" category_url: https://boursel.com/category/companies author: "Daniel Okonkwo" published: 2026-08-21T10:57:00.000Z updated: 2026-08-21T10:57:00.000Z canonical: https://boursel.com/article/the-bond-market-has-started-charging-more-for-ai-even-to-the-safest-borrowers tags: ["corporate-bonds", "artificial-intelligence", "data-centers", "credit-spreads", "oracle"] --- # The bond market has started charging more for AI, even to the safest borrowers US technology companies have issued at least $220 billion of debt this year to fund data centers, against $12.5 billion in the same stretch of 2025. The money is still there, but the price has changed: technology now trades wider than the investment-grade market it used to trade through, and demand for each successive deal has thinned. The AI build-out is being paid for with borrowed money on a scale the corporate bond market has not been asked to absorb before, and the market has started to answer. ## The size of it Hyperscaler debt issuance reached [at least $220 billion by early August, against $12.5 billion in the equivalent period of 2025](https://www.investing.com/news/stock-market-news/analysisus-corporate-ai-debt-surge-tests-investor-limits-as-fatigue-emerges-4871227), a roughly seventeenfold increase. Goldman Sachs data cited by [Fortune puts total AI-related issuance this year nearer $500 billion](https://fortune.com/2026/08/21/bond-market-warning-cheap-money-era-over-ai-debt/) on a broader definition that includes the utilities, developers and financing vehicles attached to the same construction. Almost all of it is investment grade. This is not speculative borrowing by weak companies; it is the largest and most profitable companies in the world funding capital expenditure with debt rather than cash flow, because the spending has outrun what the cash flow can cover. ## What a credit spread is, and why it moved A credit spread is the extra yield a company pays over a government bond of the same maturity. If the 10-year Treasury yields 4.69% and a company's 10-year bond yields 5.79%, the spread is 110 basis points, and that gap is what investors charge for the risk that the company does not pay. Two things move a spread: the market's view of the borrower's credit, and the balance of supply and demand for its paper. The AI issuance wave is mostly the second. There is nothing wrong with Amazon's balance sheet, but when every large technology company issues at once, buyers who want to stay diversified run out of room. You can see it in the pricing. Amazon's $25 billion bond sale [came at roughly 120 basis points over Treasuries, about double the level the company paid last year](https://www.investing.com/news/stock-market-news/analysisus-corporate-ai-debt-surge-tests-investor-limits-as-fatigue-emerges-4871227). Alphabet had to concede 10 to 15 basis points against its own existing bonds to place a recent deal. And technology as a sector now trades at [89 basis points, about 9 basis points wider than the overall investment-grade index](https://www.investing.com/news/stock-market-news/analysisus-corporate-ai-debt-surge-tests-investor-limits-as-fatigue-emerges-4871227). That last figure is the one that matters. Technology used to trade through the index, meaning tighter, because the sector was cash-rich and asset-light. It is neither of those things any more. ## Where the limit is Demand is thinning rather than disappearing. Oversubscription on hyperscaler deals, the ratio of orders to bonds on offer, [fell from close to five times in February to below two by July](https://debtexplorer.whitecase.com/leveraged-finance-commentary/high-yield-bond-markets-ride-the-data-center-boom). A deal covered twice still gets done; it just gets done at a price the issuer would rather not pay. The constraint underneath is concentration. Pension funds and insurers typically cap exposure to any single issuer at 2% or 3% of assets. When the same handful of names dominate issuance for a year, those caps bind, and the marginal buyer has to be paid to stretch. ## The credit consequence, so far The rating agencies have started to act on the individual cases rather than the sector. S&P cut Oracle to BBB- from BBB [on 9 July](https://www.spglobal.com/ratings/en/regulatory/article/-/view/sourceId/101695609), one notch above speculative grade, citing a free cash flow deficit that could widen to $42 billion as it builds out, and the concentration of roughly half its $638 billion of remaining performance obligations in a single customer, OpenAI. Moody's has framed the sector-wide version: AI infrastructure spending by six major technology firms reaching [$785 billion in 2026 and $1 trillion in 2027](https://www.cnbc.com/2026/07/24/moodys-ai-spending-credit-quality-amazon-meta-alphabet.html), against $460 billion of direct debt and $1.2 trillion of off-balance-sheet lease and purchase commitments across Microsoft, Alphabet, Amazon, Meta, Oracle and CoreWeave. ## Why a wider spread matters even when the money arrives Every one of these deals has been funded. That is the point most often missed: the market is not refusing, it is repricing. But the cost of capital is the discount rate applied to every future dollar an investment is supposed to earn, and a data center is a long-lived asset whose returns arrive over a decade. Financing it 60 basis points wider does not stop the project; it raises the return the project has to clear to be worth building. Do that across a trillion dollars of planned capital expenditure and the marginal project stops making sense, which is how a bond market slows a boom without ever saying no. The backdrop makes it heavier. Treasury yields set the floor under every corporate coupon, and they rose again on Thursday, to 4.69% on the 10-year and 5.24% on the 30-year. Spread and benchmark are widening together. ## Sources - [Analysis: US corporate AI debt surge tests investor limits as fatigue emerges](https://www.investing.com/news/stock-market-news/analysisus-corporate-ai-debt-surge-tests-investor-limits-as-fatigue-emerges-4871227) - [The bond market is sending CEOs a blunt message: borrowing costs are going to go up](https://fortune.com/2026/08/21/bond-market-warning-cheap-money-era-over-ai-debt/) - [Oracle Corp. downgraded to 'BBB-/A-3' from 'BBB/A-2'](https://www.spglobal.com/ratings/en/regulatory/article/-/view/sourceId/101695609) - [Moody's says 'unprecedented' AI spending threatens credit quality of Amazon, Meta, Alphabet and others](https://www.cnbc.com/2026/07/24/moodys-ai-spending-credit-quality-amazon-meta-alphabet.html)