As tech giants ramp up borrowing to build AI data centers, some are questioning whether the U.S. investment-grade bond market can keep pace with rising supply. But Stephanie Aliaga, global market strategist at JPMorgan Asset Management, argues that hyperscalers still carry relatively low leverage, and robust AI computing demand supports future cash flows, meaning the bond market is well-positioned to absorb new issuance. JPMorgan estimates the six largest hyperscalers could take on roughly $1.5 trillion in additional debt from current levels without materially straining their financial profiles.
Hyperscaler bonds have grown to about 5% of the U.S. investment-grade bond index, up from roughly half that two years ago. As AI infrastructure spending expands, these tech giants are rapidly gaining influence in global fixed-income markets. "We think they are going to keep issuing debt," Aliaga said in an interview on Tuesday.
AI giants double their bond market share in two years, still room for $1.5 trillion more
Data center construction requires enormous capital outlays, prompting tech companies to increasingly rely on bond markets for long-term funding. The six major hyperscalers now represent about 5% of the U.S. investment-grade index, double their share from two years ago. The surge in issuance has led some investors to worry about growing supply pressure on investment-grade credit. But JPMorgan Asset Management believes these balance sheets still have ample room to expand. Aliaga noted that the hyperscalers' leverage remains significantly lower than the broader investment-grade universe. JPMorgan estimates they could add roughly $1.5 trillion in debt with relative ease.
The firm expects big tech to keep tapping debt markets as AI infrastructure investment continues to climb. Aliaga said debt is not inherently problematic. For hyperscalers building data centers expected to operate for five, ten, or even more years, debt can actually be a highly attractive financing option.
Treasury yields at highest since 2008, tech debt wave raises "buyer shortage" questions
Aliaga's remarks come as global bond markets face notable stress. U.S. government bond yields have risen to their highest level since 2008 on concerns about persistent inflation and expectations for further rate hikes by major central banks. Meanwhile, the AI infrastructure boom has accelerated investment-grade issuance from big technology firms. With both government and corporate financing needs expanding, some investors worry whether the bond market has enough buying power to digest the new supply.
Aliaga argues those concerns may be overdone. "We think the market is fully capable of absorbing these new bond offerings. If anything, this may actually help make the AI boom more sustainable," she said. In other words, if bond markets can provide tech giants with steady, long-term funding, AI data center projects do not have to rely entirely on company cash flows, which could extend the current AI capex cycle.
Anthropic's compute contracts exceed $175 billion, AI infrastructure investment keeps expanding
AI computing power remains one of the most constrained resources in the global tech sector. Anthropic, for example, has committed to cloud computing contracts worth more than $175 billion, underscoring the massive scale of AI infrastructure demand. As large tech companies, AI labs, and cloud providers compete for GPU servers, storage, networking, and data center capacity, capital needs across the AI supply chain are expanding rapidly.
This financing demand is not only pushing tech firms to increase bond issuance but also competing with U.S. government and other sovereign borrowers for global fixed-income capital. However, Aliaga maintains that AI demand itself provides important support for these debts. Operating cash flows at hyperscalers currently cover their capital expenditures. More notably, the backlog of signed customer contracts at the three largest hyperscalers is growing faster than capital spending. Aliaga views this as a positive signal because it suggests that heavy investment in AI infrastructure is increasingly backed by future customer commitments, improving the odds of healthy returns on these projects.
AI infrastructure investment could reach $5.5 trillion by 2030
JPMorgan projects that global AI infrastructure investment could total as much as $5.5 trillion by 2030. Such an enormous capital requirement means that even the world's largest tech companies, despite strong cash generation, cannot fund it all from internal cash flows alone. Aliaga says hyperscaler cash flows can only cover a portion of that $5.5 trillion, making debt financing and other external capital sources increasingly important.
Beyond public bond markets, alternative capital such as private credit could play a significant role in AI infrastructure funding. That means the AI investment wave may increasingly extend beyond tech equities into bond and private credit markets. For investors, assessing the AI capex cycle is no longer just about how much tech companies are willing to spend, but also whether global capital markets can consistently provide sufficient financing for these projects.
AI capex will eventually slow, easing pressure on margins and free cash flow
Aliaga believes the extraordinary capital spending at hyperscalers will not continue at this pace forever. As AI infrastructure gradually matures, capex growth will eventually moderate, relieving pressure on margins and free cash flow at major tech firms. A slowdown in capital spending "should provide some relief to margins and free cash flow," she said.
Still, in the foreseeable future, AI compute supply constraints remain far from resolved. The next critical question for the AI investment cycle may not be whether demand exists, but who can overcome compute supply bottlenecks first and when new capacity can actually come online. Aliaga specifically highlighted memory supply constraints as one of the key limitations facing AI infrastructure expansion. For investors, the key focus ahead will be identifying companies that can break through supply limitations on memory and other critical components, and bring new AI computing capacity to market in a timely manner.
Comments