AI Infrastructure Draws $33.5 Billion — and the IMF Warns of a Correction

As fast as money pours into AI infrastructure, the risk of not earning it back is growing too.

Two stories about AI infrastructure landed on the same day, October 13. Morgan Stanley said U.S. startups working on the power and compute bottlenecks of AI data centers have raised about $33.5 billion this year, and the IMF warned that a correction in AI-linked stocks could shake global financial stability. The first is about where the money is going. The second is about what happens if that money doesn’t pay off.

$33.5 billion for power and compute bottlenecks

According to a Morgan Stanley report, U.S. startups tackling the power and compute bottlenecks of AI data centers had raised about $33.5 billion as of early October. That is nearly double the 2025 total.
Reports say more than 80% of it went to securing power and improving compute efficiency. These include power supply, compute per unit of electricity, and technology that combines power with computing.
The report calls power “the key constraint.” It lists permits, construction delays, grid connections and equipment procurement times as risks, and says a data center’s economics will increasingly depend on where it is built.
The report adds that the next test will move from how much was raised to how much capacity is actually running.

We could not obtain the Morgan Stanley report itself. This section relies on coverage dated October 13.

IMF: an AI correction could spill into financial stress

The IMF released its Global Financial Stability Report (GFSR) on October 13 at its annual meetings in Bangkok. The core message: if AI investment or profits fall short of expectations, AI-linked stocks and credit could be repriced, and that could sharply tighten global financial conditions.

Three triggers: a correction in AI-linked assets, a surge in government bond yields, and a major cyberattack.
Concentration: reports say the seven largest U.S. tech stocks now make up more than 35% of the S&P 500, up from about 12% a decade ago.
Debt: the large cloud companies (hyperscalers) now account for about 10% of global investment-grade corporate bond issuance, up from under 1% in 2024.
Scenario: in a scenario reported by Reuters, a steep drop in AI investment and a sharp fall in asset prices cut global GDP by 1.1% by 2028.
IMF Managing Director Kristalina Georgieva likened today’s global economy to one standing on a single leg: AI.

The IMF made clear this is not a prediction of an imminent crash. The point is that when debt, concentrated holdings and forced selling pile up, even a small setback in AI profitability could become systemic. The IMF also raised its 2026 global growth forecast slightly to 3.1% from 3.0% in July, saying AI infrastructure investment offset higher oil prices from the war in the Middle East.

The two stories are linked

The money is going to data centers, power and compute equipment. Investments like these cost a lot up front and pay back later, and that gap is exactly what the IMF worries about. If investment grows fast while returns arrive late, stock and bond prices could wobble together. Morgan Stanley’s shift in the yardstick from “capital raised” to “capacity deployed” is, ironically, also a sign that the test is getting closer.

What could change for developers

What follows is our estimate and advice based on the reports.

Cloud and GPU prices and capacity. If power is the bottleneck, where a data center is built can change how hard GPUs are to get, and how much they cost, by region. Avoid depending on a single region.
Survival risk for AI startup tools. If investment dries up, cash-hungry AI services may raise prices or cut features. If a core feature leans on one small outside service, look for alternatives ahead of time.
API prices won’t move in one direction. More infrastructure can lower prices through more supply, but profit pressure can push them up. Put a budget cap on usage-based billing.
Stay able to switch. Keep both your model provider and your cloud swappable (isolate the calls, keep your own evaluation set).

What to watch

The Morgan Stanley report itself and capacity-deployed metrics, next-quarter capital-spending guidance from the large cloud companies, and moves in AI-linked stocks and corporate bonds.