After record AI investments, investors have sharply slammed the brakes: what put them on edge
Back in the summer, companies were borrowing tens of billions of dollars for data centers, chips, and other artificial intelligence (AI) infrastructure. Now the flow of money has dramatically shrunk: global debt financing tied to AI fell from a record $113bn in June to $23bn in September.
The shift is especially noticeable in the US. In September, issuance of new investment-grade bonds to fund AI projects virtually came to a halt — after extremely active borrowing in previous months.
Simply put, companies receive billions from investors now and promise to return them later with interest. And if earlier the market was eager to give money to almost any big AI-related plan, investors now increasingly ask: when will these enormous outlays start paying off?
From $113bn to $23bn in just a few months
According to Morgan Stanley figures cited by the Financial Times, in September companies worldwide raised about $23bn of new debt tied to developing AI.
That is roughly 80% below the June peak, when such funding reached a record $113bn.
The calculations include both public bonds and private debt deals.
That said, in the first nine months of 2026 global AI financing through debt already reached about $466bn. So the September figures, for now, point not to the disappearance of money but to a sharp slowdown after a record first half of the year.
Even this summer investors were ready to give AI giants hundreds of billions
In July Kurs already reported that the largest tech companies had sharply increased borrowing for AI development.
Alphabet, Amazon, Meta, Oracle, Nvidia, and SpaceX placed about $244bn in bonds back then. Even at that stage investors began demanding a higher yield for the risk.
The new figures reveal the next stage of this story: after a record flow of borrowing, the market for the first time visibly hit the brakes.
What put investors on edge
One of the reasons is fairly simple: many companies had borrowed huge sums in advance and do not yet need new funding at the same pace.
But it is not just that.
The cost of borrowed money has risen, and AI projects themselves are becoming ever more expensive and complex.
Construction of data centers can be hindered by power shortages, delays in grid connections, shortages of certain components, and pushback from local authorities or residents to new facilities.
The larger the initial bill becomes, the more important for investors is the key question: how much money the project can truly generate after launch.
AI needs money long before profits appear
Modern AI infrastructure requires enormous outlays even before it starts running.
Companies need processors, servers, cooling systems, optical components, data transmission networks, and new power plants or connections to the grid.
A data center can cost billions of dollars, and it must pay for itself after launch — through demand for computing power and AI services.
Therefore investors are starting to distinguish more carefully between projects by large profitable corporations and riskier plans by companies that still need to prove that future revenue will justify the construction costs.
Why this matters not only to bondholders
The speed of the current AI race directly depends on the cost of funding.
If raising tens of billions of dollars becomes more expensive or difficult, companies are forced to pick more carefully which data centers to build first, which projects to delay, and how much of their own money to put in.
That can affect the pace of building new infrastructure, demand for chips, electricity, and equipment — and thus the speed of expansion of AI services themselves.
That is why the debt market has become something of an indicator of how long the investment boom can continue at the same pace.
But it is still too early to call the end of the AI boom
Morgan Stanley does not view the September drop as a sign that industry funding is drying up.
Rather, it looks like a pause after an unusually active period, when many companies had secured funding in advance.
New major deals are already being prepared in the market, so the amount of raised capital may grow again.
However, investors' attitudes are becoming noticeably more selective: after hundreds of billions of dollars of new debt, it is no longer enough for them to simply hear that the next project is tied to artificial intelligence.
Now it is ever more important to understand who exactly is borrowing the money, what it will be spent on, and when the project will start paying it back.
A new phase is beginning for AI
Initially the main constraint on the AI boom was seen as a shortage of chips. Then electricity, sites for data centers, and the ability to build the needed infrastructure quickly were added.
Now another constraint is emerging — investors' willingness to keep funding this race with such huge sums.
The drop from $113bn to $23bn in a few months does not yet signal a crisis. But it reveals an important turn: after a period when money flowed almost non-stop behind AI boom promises, investors are starting to count much more carefully how much these promises cost and when they might pay off.
Based on materials from: Financial Times, Morgan Stanley.