After AI Absorbs Capital, Don't Just Watch for Bubbles

Last night, I exported the capital expenditure metrics of several listed cloud providers into Excel and got a very awkward table. Models, data centers, power, networks—every row is rising; financing costs for manufacturing, software services, healthcare, and education haven't come down either. I spent three weeks using LSEG and capex data to put "AI boom" and "other industries lacking funds" on the same chart for the first time. It feels less like a pure bubble and more like the entry point for capital has shifted.

Last week I wrote about Uber cutting middle management, discussing organizational hierarchy. Today's table shows another side: Money hasn't disappeared; it's just been sucked away by more certain stories.

Capital isn't magically increasing; it's just changing entry points

According to that Project Syndicate article, the AI investment boom drives up demand for capital while savings decline, and interest rates remain high. This is crucial. We used to say "the market has money," but money doesn't fall from the sky—it has a cost. If AI companies are willing to bear higher financing costs because they believe models, computing power, and token calls will form new moats, other industries can't compete on the same field. Looking at the data, the issue isn't "is there a bubble," but "who gets the money first, and who turns it into cash flow first."

David Rosenberg's judgment is also noteworthy. He believes the AI boom has held back a recession but is "draining momentum from other business capital expenditures." If this statement holds true, it's bad news for industry analysts. Because macro resilience might stem from one sector over-absorbing capital, rather than the overall economy becoming healthier.

The core variable in this race is cost of capital.

I listed three easily confused points:

• High AI capex doesn't mean all AI companies are profitable. Much of the money goes first to GPUs, power, data centers, and depreciation.
• Difficulty in financing for other industries doesn't mean they lack demand. More likely, banks and investors demand higher returns or are simply waiting for the interest rate path to clarify.
• Falling inference costs for large models don't mean enterprise-side costs are falling. Computing power, storage, networking, compliance, and organizational restructuring will squeeze out the savings.

I tried Claude, ChatGPT, Gemini, and Grok for a week, and ran DeepSeek V4 Pro for two weeks. The models themselves are indeed improving fast. But when it comes to enterprise procurement, what often blocks the budget isn't "can it generate," but "is there a process to handle what's generated." So capex doesn't automatically translate to productivity.

The real question is who gets squeezed out

That SSGA article says that in the past, everyone competed for who could grab the largest share of AI capex. Now the question has changed: With rising financing costs, can this capex cycle hold up? This shift is very real. Because capital markets' patience usually goes to the top tier first, then the second tier, and finally to any company touching AI. Once debt maturities, electricity prices, depreciation, and interest rates all rise together, the story becomes hard to tell.

CNBC mentioned that high energy prices and surging AI capex are making the Fed and Kevin Warsh uncomfortable. There's a very realistic conflict here: On one hand, computing expansion drives investment; on the other, electricity prices and financing costs push inflationary pressure. Policy can't just look at the AI line. The more AI resembles infrastructure, the closer it gets to being a utility. Utilities aren't valued on concepts; they rely on cash flow, balance sheets, regulation, and delivery.

So don't just ask "Is AI a bubble?" Ask instead: Which industries will be squeezed out, which jobs will be suppressed, and which capital expenditures will become sunk costs.

I recall my previous concerns about Agent security. Mechanisms like context isolation and concurrent leakage sound technically heavy, but they're commercially heavy too. If enterprises dare not hand core processes to Agents, capex will stall at the pilot stage. Pilots spend R&D budgets; scaling spends financing budgets. The difference is huge. When financing budgets tighten, the first things cut are often projects without clear recovery paths.

Bubble judgments depend on interest rates and cash flow

That ECB article provided a framework I found very useful: Uncertainty spreads from a single sector to the whole economy. Initially, people thought AI was just one sector, but later realized it affects power, real estate, employment, fiscal policy, and financial stability. At that point, asset price volatility isn't just tech stock volatility, but a change in overall risk appetite.

History warns us. The lesson from the Dot-com era wasn't that technology was useless, but that the installation phase was over-capitalized. Technology changing the economy and investors making money are not the same thing. AI might follow this pattern now: Data centers are built, models get stronger, applications emerge, but profits are eaten by depreciation, financing, energy, and competition.

The WEF Chief Economists Survey mentioned that if AI-related assets suddenly reprice, the impact scope could be significant. I'm not surprised by this claim. In the capex table exported from LSEG, the most glaring thing isn't growth, but concentration. A few companies take away funds, talent, power, and credit lines all at once. Concentration itself improves efficiency but creates fragility.

Bubbles don't necessarily burst because the model fails; they might burst because money gets expensive.

My current judgment leans cautious: The AI boom will continue, but it will raise the capital threshold for the rest of the economy. In the next one or two years, those who survive won't be the companies best at telling model stories, but those who can align inference costs, customer budgets, delivery cycles, and cash flow.

The protagonist of this article is capital, not models. No matter how fast the models are, they must first pass the hurdle of interest rates.

---

📌 This article is compiled from Hacker News. Original source: https://publicemails.com/blog/display/659255/Will-the-AI-Boom-Price-the-Rest-of-the-Economy-Out-of-Capital

All rights reserved by the original authors. This is a compilation and independent analysis based on public reports.