U.S. technology giants are piling into the bond market to fund artificial intelligence expansion, while off-balance-sheet obligations tied to that buildout are also swelling.
Bloomberg reported on Aug. 13 that Advanced Micro Devices, or AMD, raised $4.75 billion in the largest U.S. dollar bond offering in the company’s history. The deal added to a broader debt boom driven by AI spending.
AI companies are tapping debt markets at scale
AMD is far from alone. This year, companies across the AI supply chain, from chipmakers to cloud providers, have turned aggressively to bond investors.
Nvidia sold $25 billion in bonds in June and drew about $85 billion in orders, more than three times the size of the offering. Foreign media reported in July that Amazon was planning to raise at least $25 billion. In early August, Alphabet, Google’s parent company, also sold $25 billion in bonds, with orders reaching $115 billion.
Reuters reported on Aug. 14, citing data from the London Stock Exchange Group, that Alphabet, Amazon and Meta have issued nearly $220 billion in bonds so far this year. That is more than double the $108 billion they sold in all of 2025.
On the surface, the borrowing spree looks like a financing race for AI infrastructure and computing capacity. Large-model training continues to push compute demand higher, and broader deployment of AI agents is lifting inference demand. High-end GPUs remain in short supply. At the same time, debt burdens are getting heavier.
Real yields have moved toward multi-year highs
The surge in issuance is also feeding into real yields. Reuters said U.S. 30-year real yields, measured using inflation-linked bonds, are near an 18-year high at about 3%. Real yields on 10-year debt in the U.K. and Germany are also hovering near the highest levels seen in more than a decade.
Investors and analysts told Reuters that, with governments still spending heavily, a jump in borrowing by AI hyperscalers has been one of the main forces pushing real yields higher. Buyers are demanding more return to keep absorbing the large amount of debt coming to market.
Real yield refers to the inflation-adjusted return bond investors require. It is also a key gauge of the true borrowing cost faced by governments and companies, and is typically shaped by expectations for growth, interest rates and money supply and demand.
Goldman Sachs and Morgan Stanley put off-balance-sheet commitments near $2 trillion
The bonds already on the balance sheet do not tell the full story.
An Aug. 11 article published by BigGoFinance, citing recent research from Goldman Sachs and Morgan Stanley, said AI cloud leaders including Alphabet, Microsoft, Amazon and Meta have built up close to $2 trillion in off-balance-sheet financial commitments through not-yet-started lease agreements and procurement contracts.
The article said, 「If returns on AI data center investments fall short of expectations, these hidden bombs buried in the notes to financial statements could deliver a material shock to credit markets.」
Goldman Sachs estimated that major cloud operators have about $1.5 trillion in off-balance-sheet lease commitments, with around $1 trillion not yet commenced. Under U.S. GAAP, those obligations are disclosed in the notes to financial statements rather than recognized as lease liabilities and right-of-use assets.
Morgan Stanley estimated that, as of the end of the first quarter, Alphabet, Microsoft, Amazon, Nvidia and Oracle had combined procurement commitments of $982 billion for chips, equipment, power and computing capacity. Put together, the two estimates approach $2 trillion.
Meta structure for Hyperion data center drew industry attention
The report traced part of this off-balance-sheet financing trend to a structured deal built for Meta’s Hyperion data center in Louisiana.
Meta and alternative asset manager Blue Owl formed a joint venture called Beignet, with Blue Owl responsible for developing and holding the data center assets. Meta owns a 20% stake in Beignet but committed to lease the facility for at least 20 years. That long-term lease support allowed Beignet to sell a record $27 billion amortizing bond deal.
Although Meta effectively carries the repayment obligation, the debt does not sit directly on Meta’s own balance sheet. That lets the company expand AI infrastructure at scale without sharply lifting reported leverage. The report said the structure has since inspired similar leasing arrangements across the industry.
Credit metrics still look sound, but markets are repricing risk
By traditional credit measures, these cloud and AI companies still appear financially solid. Morgan Stanley data showed their average net leverage ratio at just 0.5x, below the 0.8x average for the technology sector and the 1.8x average for U.S. non-financial corporates. Their total cash holdings also exceed reported debt.
More optimistic investors argue that strong cash flow from core businesses provides a cushion, and that if data center investments underperform, equity holders would take the first hit rather than bondholders.
Even so, credit markets have started to reassess the risk. Beignet bonds, for example, are yielding 6.95%, up from about 5.65% when they were issued last autumn. Part of that move reflects the broader rise in U.S. Treasury yields, but it also shows investors asking for more compensation.
Prices for credit default swaps, or CDS, which track default risk on debt, have also been rising. Debt insurance costs for Oracle, Nvidia, Meta and Alphabet have each climbed to record highs.
This article was sourced from the WeChat public account Zhongxin Jingwei (ID: jwview) and written by Luo Kun.