AI infrastructure “burns money” to reach 5.5 trillion US dollars Xiaomo: the bond market has the ability to absorb the wave of debt issuance, and tech giants can still increase leverage

The Zhitong Finance App learned that as tech giants set off a wave of debt issuance to build AI data centers, the market is beginning to worry about whether the US investment-grade bond market can absorb the increasing supply of debt. However, Stephanie Aliaga, a global market strategist at J.P. Morgan Asset Management, believes that the current leverage level of hyperscale cloud computing companies is still low, and strong demand for AI computing power also supports future cash flow, so the bond market is fully capable of absorbing new issuances. J.P. Morgan estimates that the six largest hyperscale cloud vendors could even add about $1.5 trillion more in debt from the current level without putting significant pressure on their financial situation.

Currently, the bonds of the six largest cloud computing companies already account for about 5% of the US investment-grade bond index, which is double that of two years ago. As AI infrastructure investments continue to expand, the influence of these tech giants in the global bond market is rapidly rising.

“We think they will continue to issue debt,” Aliaga said in an interview on Tuesday.

The six AI giants' share of bonds has doubled in two years and can still borrow another 1.5 trillion US dollars

AI data center construction requires huge capital investment, so technology companies are increasingly raising long-term capital through the bond market. Currently, the six largest hyperscale cloud vendors already account for about 5% of the US investment-grade bond index, which is double the level of two years ago. The rapidly growing supply of bonds has also made some investors start to worry about whether technology companies will put more and more supply pressure on the investment-grade bond market.

However, J.P. Morgan Asset Management believes that these companies' balance sheets still have plenty of room to expand. Aliaga pointed out that compared with the overall investment-grade bond market, the current leverage ratios of the top six hyperscale cloud vendors are still significantly lower. J.P. Morgan estimates that it would be easier for these companies to add about $1.5 trillion in additional debt to their current level.

The bank expects that as AI infrastructure investment continues to expand, large technology companies will continue to use the bond market for financing in the future. Aliaga said debt itself is not a bad thing. Debt can actually be a very attractive way to finance hyperscale cloud vendors that are building data centers that can be used for 5, 10, or more years.

US bond yields rose to a high level since 2008, and the wave of tech giants debt raised concerns about “insufficient buyers”

Aliaga's comments come at a time when the global bond market is under clear pressure. US government bond yields have risen to their highest level since 2008 due to market concerns about continued high US inflation and raised expectations for further interest rate hikes by major central banks. At the same time, the boom in AI infrastructure construction is driving large technology companies to speed up the issuance of investment-grade bonds.

Government financing needs are expanding at the same time as technology companies' financing needs, which has also raised investors' concerns about whether the bond market has enough purchases to absorb the new supply.

But Aliaga believes this concern may have been exaggerated. She said, “We think the market is fully capable of absorbing these additional bond issues. If anything, it might actually help the AI boom become more sustainable.”

In other words, if the bond market can continue to provide long-term and relatively stable financing channels for large technology companies, AI data center construction will not have to rely entirely on the company's own cash flow, thus helping to extend the current AI capital expenditure cycle.

Anthropic computing power contracts have exceeded 175 billion US dollars, and the scale of AI infrastructure investment continues to expand

AI computing power is still one of the most scarce resources in the global technology industry. Take the artificial intelligence company Anthropic as an example. The company has promised to sign more than 175 billion US dollars of cloud computing power contracts, highlighting the huge scale of AI companies' demand for computing infrastructure.

As large technology companies, AI laboratories, and cloud computing companies compete for GPU, server, storage, network, and data center resources, capital requirements for the entire AI industry chain are rapidly expanding.

This demand for financing not only pushes technology companies to increase bond issuance, but also begins to compete with the financing needs of the US government and other sovereign issuers for capital in the global fixed income market.

However, Aliaga believes that AI demand itself is also providing important support for these debts.

Currently, the operating cash flow of hyperscale cloud vendors can roughly cover their capital expenses. What is more noteworthy, however, is that the backlog of customer contracts already signed by the three largest hyperscale cloud vendors is growing faster than capital expenditure.

Aliaga believes this is a positive sign because it shows that the company's huge investment in AI infrastructure is being supported by more and more future customer needs, thereby increasing the likelihood that these projects will eventually receive a return on investment.

AI infrastructure investment may reach 5.5 trillion US dollars by 2030

J.P. Morgan predicts that total global investment in AI infrastructure could reach a staggering $5.5 trillion by 2030. This huge demand for capital means that even if the world's largest technology companies have strong cash generation capabilities, it is difficult to cover the entire investment by relying on their own operating cash flow alone.

Aliaga said that hyperscale cloud vendor's own cash flow can only cover a portion of the $5.5 trillion investment requirement, so debt financing and other forms of external capital will play an increasingly important role in the future.

In addition to the open bond market, alternative capital such as private credit may also be an important source of funding for AI infrastructure. This means that the AI investment boom is likely to expand more and more from the technology stock market to the bond and private credit markets in the future.

For investors, evaluating the AI capital expenditure cycle will not only determine how much money technology companies are willing to invest; they also need to pay attention to whether the global capital market can continue to provide sufficient financing for these projects.

AI capital expenditure will eventually slow, and profit margins and free cash flow are expected to improve

Aliaga believes that the huge capital expenses of today's hyperscale cloud vendors will not continue to grow at such a rapid rate forever. As AI infrastructure is gradually built, capital expenditure growth will eventually slow down. At that time, the pressure on profit margins and free cash flow of large technology companies is expected to ease.

The eventual deceleration of capital expenditure “should bring some relief to profit margins and free cash flow,” she said. However, at least in the foreseeable future, it will still be difficult to completely resolve the problem of the tight supply of AI computing power.

Therefore, the real critical question in the next phase of the AI investment cycle may not be whether there is demand, but rather who can take the lead in solving the bottleneck in computing power supply and when these additional production capacities can actually be launched.

Aliaga specifically pointed out that memory supply bottlenecks are one of the important limitations facing the current expansion of AI infrastructure. For investors, in the future, they need to focus on determining which companies can take the lead in breaking through supply restrictions on key components such as memory and put new AI computing power into the market.