U.S. AI Debt Boom Reaches $220 Billion as Bond Investors Start Demanding More Yield

Debt issuance by U.S. technology companies racing to finance artificial intelligence infrastructure has climbed to $220 billion in 2026, according to BNP Paribas data as of August 10 cited by Reuters. That is up from $12.5 billion in the comparable period last year, a jump large enough to change the way bond investors are pricing even some of the strongest corporate borrowers.

The immediate concern is not that companies such as Amazon and Alphabet suddenly look unable to repay their debts. The bigger issue is supply. A small group of hyperscalers is returning to the bond market with unusually large offerings and more long-dated securities, forcing investors to decide how much exposure they are willing to hold and what extra yield they need for taking it.

Reuters reported that technology-sector corporate bond spreads have widened to about 89 basis points over U.S. Treasuries, roughly 9 basis points wider than the overall investment-grade market, based on figures cited by Capital Group portfolio manager Karen Choi. DWS and Schroders portfolio managers also told Reuters that recent offerings have required larger pricing concessions than earlier in the year, a sign that buyers are becoming more selective even though the underlying issuers remain highly rated.

Bond buyers are charging more as AI supply builds

The shift is important because hyperscalers entered the AI buildout with unusually strong balance sheets and historically low funding costs. As capital spending moved sharply higher, however, the amount of debt available from the same group of issuers began to matter as much as their credit quality. Investors can like Amazon or Alphabet as borrowers and still demand more spread if their portfolios are already carrying large positions in technology debt.

BNP Paribas forecast in May that investment-grade AI hyperscaler bond supply would reach $250 billion in 2026 and estimated that their combined capital spending could reach $725 billion. The bank said issuance had already exceeded the 2025 record at that point. With the latest figure at $220 billion through August 10, borrowing has reached about 88% of BNP Paribas’s full-year supply forecast.

The pressure is showing up in new-issue pricing rather than in a broad loss of access to capital. Reuters reported that Alphabet’s bond offering earlier this month was well received but still needed a concession of roughly 10 to 15 basis points compared with the company’s existing bonds. The distinction matters: investors are still buying, but they are no longer treating repeated AI-related borrowing as something that can be absorbed at almost any price.

Portfolio construction creates another constraint. Choi told Reuters that many pension and insurance investors limit exposure to a single corporate issuer to roughly 2% to 3% of assets. Those limits can become binding when the same companies issue tens of billions of dollars of bonds in a short period, particularly at longer maturities that are attractive to insurers and pension funds matching long-dated liabilities. The market therefore can reach a practical capacity limit before investors develop serious doubts about an issuer’s solvency.

Amazon’s bond sales show how quickly long-dated funding has scaled

Amazon provides one of the clearest examples of the change. The company raised $37 billion in a U.S. dollar bond offering in March, then returned in July with another $25 billion sale. The July financing was spread across eight tranches, including fixed-rate notes maturing in 2029, 2031, 2033, 2036, 2046, 2056 and 2066, plus a floating-rate note due in 2029.

The pricing term sheet filed with the SEC shows how the required compensation rose with maturity. Amazon’s 2046 notes priced at 100 basis points over the benchmark Treasury, the 2056 notes at 110 basis points over and the 2066 notes at 125 basis points over. Their coupons were 6.00%, 6.10% and 6.25%, respectively. Those terms do not by themselves prove that investors were reluctant, but they illustrate the cost of asking the market to absorb large amounts of long-duration corporate debt.

Amazon’s borrowing has arrived alongside a sharp increase in infrastructure spending. In its second-quarter filing, the company reported $96.3 billion of cash capital expenditures for the first six months of 2026, up from $55.6 billion a year earlier. Amazon said the spending primarily reflected technology infrastructure, with the majority supporting AWS growth, as well as additional fulfillment capacity. AWS property and equipment, net, rose to $263.75 billion at June 30 from $190.06 billion at the end of 2025.

The effect is visible in cash flow as well. Amazon reported trailing-12-month operating cash flow of $161.4 billion, but purchases of property and equipment, net of proceeds and incentives, reached $169.0 billion over the same period. That pushed its reported free cash flow measure to an outflow of $7.6 billion. The company still generates enormous operating cash flow, but the scale of infrastructure spending helps explain why even highly profitable technology groups are making much heavier use of bond markets.

AI investment is changing the financing profile of megacap technology

The same pattern extends beyond Amazon. Meta reported $31.08 billion of capital expenditures, including principal payments on finance leases, in the second quarter and narrowed its 2026 capital expenditure outlook to $130 billion to $145 billion. It also reported $83.66 billion of long-term debt at the end of June. Alphabet said earlier this year that it expected 2026 capital expenditures of $175 billion to $185 billion, with most of the spending directed toward technical infrastructure such as servers, data centers and networking equipment.

That combination of high spending and strong credit quality makes the current bond-market debate different from a traditional credit scare. Investors are not primarily asking whether the largest technology companies can service their debt. The harder portfolio question is how much exposure they should carry to the same handful of issuers, especially when those positions add long duration and new supply keeps arriving.

The result is a market in which funding remains available, but pricing discipline is becoming more visible. Heavy Treasury issuance and broader corporate borrowing mean AI companies are competing for investor capital at the same time that their own financing requirements are rising. If hyperscalers continue to issue at the current pace, wider spreads and larger concessions would raise the marginal cost of building data centers even without any deterioration in underlying credit ratings.

BNP Paribas’s $250 billion full-year forecast now sits only about $30 billion above the August 10 total. The next large hyperscaler offerings will provide a clearer test of whether investors can absorb the remaining supply near current spreads or insist on still more yield to finance the next stage of the AI infrastructure buildout.