Here are our thoughts on why the much-feared AI bubble won’t pop anytime soon.

This is a genuine paradigm shift — but demand is outstripping capacity, not the reverse

Jamie Dimon’s crew confirms what we all suspect: AI represents a very genuine paradigm shift, comparable to railroads in the 1800s or the internet in the 1990s. Generative AI is advancing rapidly, with agentic models potentially reaching human-level performance by mid-2026. US tech giants are projected to triple their annual capital expenditures from 2023 levels, potentially exceeding $500 billion by 2026. Real cash will be spent—that means the providers of the products that are essential to AI will have material demand, and will create material revenues.

Historically, genuine paradigm shifts do coincide with investment bubbles. The railroad boom created massive overcapacity. The dot-com era left millions of miles of unused fiber optic cable buried underground, literally dark fiber that never carried a single byte of data.

So what’s different this time? JP Morgan argues that unlike those past booms featuring unjustified overcapacity, the current AI infrastructure build-out is actually struggling to keep up with demand. Data center vacancy rates have hit a record low of 1.6%. Components remain scarce. The investment is chasing actual need rather than speculative future demand.

The paradigm is real. The question is whether demand will remain real once the technology matures and the initial gold rush subsides. And JP Morgan predicts that won’t happen in 2026. I like to sum it up in a pretty simple formula: supply > demand = bubble.

Cheap credit is not flowing indiscriminately — it’s being directed to blue chips

Credit is becoming more available. Oracle’s oversubscribed debt deal for AI investment is one recent example. The Fed’s rate-cutting cycle only sweetens the financing environment. In a typical bubble, this is where alarm bells start ringing.

But there is indeed a vital nuance: this capital is flowing primarily to the largest, most creditworthy tech companies with fortress balance sheets. They contrast this sharply with bubble eras fueled by cheap, speculative credit to marginal players— think subprime mortgages handed out like central bank liquidity during a crisis.

The current financing environment is characterized by fairly disciplined markets willing to fund blue-chip companies, not a broad-based decay in lending standards that would indiscriminately inflate asset prices. When Oracle wants to borrow billions for AI infrastructure, investors are comfortable because Oracle has decades of cash generation behind it.

The lending standards are holding. For now.

The hyperscalers are using operating cash, not financial alchemy

This is perhaps JP Morgan’s strongest point. In the housing bubble, banks were lending 100%+ of home values with zero down payments. In the dot-com boom, companies with no revenue were taking on massive debt to build infrastructure. The pattern is always the same: leverage piles up, risk accumulates, and when the market turns, the whole structure collapses.

The major AI players are funding their historic capital expenditures from their own substantial operating cash flows, which still exceed their spending and dividends. They’re not mortgaging their future— they’re spending money they actually have. While there are emerging curiosities like GPU-collateralized debt (yes, that’s apparently a thing now), the foundation remains cash, not debt.

This is the Michael Burry test. When credit quality decays and financial engineering replaces fundamentals, that’s when the smart money starts shorting. JP Morgan’s analysis suggests we haven’t reached that point. The dangerous leverage simply isn’t there yet.

Valuation “froth” lives largely in private markets, not public markets

Now we’re getting to the interesting part— the valuation gap.

JP Morgan identifies this as the clearest sign of “froth,” but crucially, only in private markets. AI startups command valuation premiums ranging from 56% to 230% over non-AI companies at later funding stages. Certainly bubble territory.

But the public market story is completely different. The AI rally in public equities has been driven entirely by explosive earnings growth. Nvidia’s earnings grew 20-fold over five years, actually outpacing its 14-fold stock increase. Forward price-to-earnings multiples for public AI stocks have declined. The public market rally appears anchored in fundamental profits, not speculation.

CHECK OUT OUR PODCAST:

This creates a fascinating dichotomy. Public market investors are getting paid. Private market investors are paying for promises.

Here’s where I think the original analysis flags a potential problem: companies like Anthropic and OpenAI have plans to go public, potentially as soon as 2026. When overleveraged, non-revenue (or barely-revenue) companies transition from private to public markets, that froth gets repriced. And the repricing can be very violent.

Moreover, the intertwining of private and public markets has never been tighter. We’ve seen the headlines about circular financing arrangements— big tech companies investing in AI startups that then turn around and spend that money on cloud services from those same big tech companies. Financial engineering in private markets has a way of leaking into public markets. Always has, always will.

JP Morgan was quick to dismiss private market problems as contained. I’d suggest otherwise.

Hype is high — but so is skepticism, which is a stabilizer

Finally, JP Morgan argues that market exuberance is being tempered by widespread skepticism. Yes, recent IPOs have performed well. Yes, there’s hype. But there’s also constant talk of bubbles, comparisons to dot-com, and even Sam Altman himself saying AI is a bubble.

This psychological dynamic supposedly acts as a counterbalancing force. When everyone is calling something a bubble, when fear and skepticism permeate discussions, that fear decreases greed in the AI trade. It’s harder for speculation to run completely wild when every other article warns about the impending crash.

This is the most fragile of their five arguments. Bubble participants have always known they were in bubbles. The difference is they thought they could get out before everyone else. “I know it’s overvalued, but I can sell before the crash” has been the mantra of every bubble since markets began.

The question isn’t whether people recognize the risk. I always say; anyone in the world recognizes risk. Timing the risk is where wealth is made.

—J&E