Big Tech AI debt hits $350bn as investors cool on long-dated bonds amid infrastructure spending spree

Big Tech Borrowing Spree Pushes AI Debt to $350bn

The five largest hyperscalers building AI infrastructure have doubled their debt to roughly $350bn over the past five years, with Alphabet, Amazon, Meta, Microsoft and Oracle leading an unprecedented borrowing spree 2

2

. AI-related debt issuance has reached $270bn across currencies this year alone, nearly double the $136bn issued in all of 2025, according to BofA Global Research 1

1

4

4

. This AI-driven capital expenditure boom reflects the industry's massive bet on generative AI and data centers, with combined capital expenditures from the top four US tech firms projected to exceed $650bn in 2026 3

3

.

The strain is becoming visible at the margins. Amazon's free cash flow turned negative in the March quarter, falling to just $1.2bn from $25.9bn a year earlier, driven by a year-over-year increase in property and equipment purchases of $59.3bn 4

4

. Oracle's debt reached approximately 2.5 times its sales, prompting S&P to cut its rating to one notch above junk, citing AI infrastructure expansion spending 2

2

. While these companies remain highly profitable—Google generated roughly $64bn in cash flow last quarter—the interest payments are climbing, with the five hyperscalers paying a combined $10bn on their debt last year, more than double the 2019 figure 2

2

.

Investor Appetite Cools as Amazon Bond Sale Draws Weak Demand

Amazon's $25bn bond sale on July 7 marked a turning point in investor sentiment toward AI-related debt 4

4

. Bank of America called the offering a "surprise," and the company had to sweeten terms by offering 18 to 21 basis points of extra yield on its longest bonds to attract buyers 4

4

. The eight-part sale, with maturities ranging from three to 40 years, drew final orders of roughly $41bn after initially peaking at around $62bn—representing just 2.5 times the bonds on offer, down from 3.2 times in March 3

3

4

4

.

According to BofA, Amazon's overall new-issue performance was the weakest for any hyperscaler since Meta's $30bn bond sale in October 2025 4

4

. The deal brought Amazon's total borrowing to $92bn this year across dollar, euro, and Swiss franc deals, surpassing amounts issued by Alphabet, Meta, and Oracle 4

4

. Amazon told underwriters it does not plan to issue additional debt in 2026, a signal intended to reassure credit markets watching the sector accumulate leverage 3

3

5

5

.

Long-Dated Debt Bears Brunt as Bond Spreads Widen

Investors are pulling back specifically from long-dated AI debt, with bonds maturing in 10 years or more becoming some of the worst performers in the investment-grade market this week 1

1

. The yield on a 30-year SpaceX bond has risen to 7.3%, up from 6.7% when it was sold less than two weeks ago 1

1

. Bonds issued by the top five hyperscalers now yield roughly 0.6 percentage points more than their blue-chip peers with the same ratings and maturities, representing the widest risk premium of any sector in the investment-grade market 1

1

.

Amazon's five-year bonds attracted about 20% more orders than the 30-year portion, with the 30-year bond yielding more than 6.1% on Friday compared to a 4.8% yield on the five-year debt 1

1

. "We prefer taking more near-term risks," said Mariya Entina, portfolio manager at DoubleLine. "Typically when you buy 30-year bonds you want businesses that have a really solid outlook, such as clear returns on investments. There's scepticism over the longer-term profitability of AI capital expenditure" 1

1

.

AI Infrastructure Investments Face Profitability Questions

The money is flowing toward Amazon Web Services, where the company is racing to add capacity for customers training and running large models, along with investments in its own Trainium chips as an alternative to Nvidia hardware 3

3

. Amazon has guided to roughly $200bn in capital expenditures for 2026, the most of any hyperscaler, up sharply from $131bn spent in 2025 3

3

5

5

. CEO Andy Jassy told analysts that AWS has to pay cash for "land, power, buildings, chips, servers, and networking gear" in advance before it can profit six months to two years down the road 4

4

.

Pramod Atluri, a portfolio manager at Capital Group, favours short-dated hyperscaler bonds, noting that "the technology is evolving so rapidly, which makes it a riskier proposition to lend for a long period of time. It's not clear what the industry landscape is going to look like ten years from now" 1

1

. Investors point to risks that future innovation in the fast-moving sector could render current AI infrastructure investments obsolete, while AI's eventual profitability may not match current expectations 1

1

.

European Credit Markets Absorb Tech's Borrowing Overflow

With American tech running short of dollars to borrow, hyperscalers have turned to European credit markets 2

2

. Morgan Stanley expects their euro borrowing to reach €50bn this year, which would make US Big Tech the single largest source of corporate debt in the eurozone, ahead of France 2

2

. Hyperscalers issued no non-dollar bonds in 2024, but by 2026 it has become a core part of their funding strategy 2

2

. Alphabet has borrowed in yen, Canadian dollars, Swiss francs and sterling inside a year, even selling a 100-year bond 2

2

.

When American giants crowd into the same euro debt that European scale-ups and infrastructure funds rely on, the cost of money shifts across the continent 2

2

. John Lloyd, global head of multisector credit at Janus Henderson, said investors needed to offload existing hyperscaler debt to make room for Amazon's new offering, given that many portfolios were already heavily exposed to AI debt. "You're taking obsolescence and technological disruption risks over time. For us to play a new deal, it has got to come with a pretty big concession to entice us" 1

1

.

Market Signals Point to Growing Caution on AI Boom

The pushback from bond investors comes as only Alphabet's stock has beaten the market this year, while Microsoft and Oracle have both fallen more than 20% 2

2

. The share buybacks that once defined these companies have all but stopped as they redirect cash toward AI infrastructure expansion 2

2

. Recent volatility in tech stocks has dampened sentiment, with some investors already carrying large exposure to the tech sector in their equity portfolios, tempering their appetite for further exposure in the debt market, according to Amanda Lynam, chief credit strategist at Goldman Sachs Research 1

1

.

The cautionary tale sits with Intel, which spent years loading up on debt, missed the AI chip boom entirely, and needed a US government bailout and an Nvidia investment to survive 2

2

. "It seems like a lot of demand hype that is very aspirational at this point," said Fitch's Jason Pompeii 2

2

. One research shop estimates the AI debt market could reach $7tn by 2029, making this quietly one of the largest debt bets in corporate history 2

2

. The gamble is no longer just whether AI works—it's whether the revenue arrives before the debt does.