Is There an AI Bubble? CAPEX, Profitability, Data Centers & Market Risk
A market commentary article examining AI spending, monetization pressure, infrastructure buildouts, valuations, and whether current expectations are running ahead of likely returns.
Yes, it’s another AI bubble post.
Tldr; there is absolutely no way all this CAPEX spending on AI will generate the huge amounts of profits to justify costs in any reasonable time frame. Investors are cowards and will run to save their money at the first sign of a slowdown, be it liquidity crunch, growth slowdown, adoption issues, anything. There is a massive over-expectation for AI profitability.
On March 10th 2000, the NASDAQ peaked, having risen 86% in 1999 alone. By March 15th the NASDAQ was down 9.2%. A month later, on April 12th it closed down 25%, ultimately leading to over 50% down by December. The selling in March began very rapidly without a particular cause, but once it gained steam it did not stop. Pets.com was not making enough money and could not possibly generate enough profit within a human lifetime to justify its valuation. Palm Inc had its IPO days before the peak, one of the largest at that point, peaking at $95 a share and crashing down to $6.50 by June of ‘01. Today the company does not exist. Cisco Systems more or less played the role of Nvidia, and only returned to its peak of 2000 this past December, jumping on yet another hype train. Apple had a single day drop of -50% when it missed an earnings report on 9/29/2000.
Why does this matter? The Dotcom bubble has two very key features that it shares with our proposed AI bubble. Insane P/E ratios and unrealistic time frames for return on money invested (ROI). P/E ratios aren’t the best indicators of whether a stock will go up or down, and many analysts casually refer to over-valuations, though only after the fact, when the euphoria of endless growth wears off. It’s a matter of human psychology; once enough investors realize the gravy train is not “guaranteed” or expected fairly soon, they stampede to bonds. No one likes being a bagholder.
No, I’m not suggesting that MAG7 companies will go under or that the economy will collapse. I am saying that there will need to be a deep correction in the interim period from now until whenever all these data centers come online. More on time frames later. If they don’t figure out how to make it profitable by then, that would just be really funny. The MAG7’s heavy reliance on AI and its own over representation on the SPY/SPX means that if there is a drop, all the average person will see is that AI is nuking their retirement portfolios. The success of AI is holding the American economy hostage, at gun point. More stock market value is dependent on AI’s success than it was on the internet during the .com bubble, plenty more GDP growth was contributed by old economy growth back then. Most of the boom of the market recently has been thanks to debt-based AI spending. Boomer companies like Caterpillar and Generac, are booming because of data center build out contracts.
Let’s look at some P/E ratios. Anything over 20x is typically considered over valued.
S&P 500: 29x (average is typically 19)
https://us500.com/tools/data/sp500-pe-ratio
NASDAQ: 33x (20yr average is 22)
https://worldperatio.com/index/nasdaq-100/
MAG7
Microsoft/MSFT: 25x
Alphabet/GOOG: 28x
Meta/META: 22x
Amazon/AMZN:30x
Tesla/TSLA: 373x (robotics memes and Musk being an AI hypebeast)
Apple/APPL: 33x (Apple isn’t nearly as deep in AI as the others)
Nvidia/NVDA: 37x (these guys are killing it thanks to all the debt spending, won’t be pretty if it slows down)
Tech/AI Related
Nebius/NBIS: 587x
Coreweave/CRWV: -334x (forward P/E)
Vertiv/VRT: 73x
Teradyne/TER: 87x
Broadcom/AVGO: 66x
Datadog/DDOG: 394x (companies similar to DDOG are all insanely overvalued)
Lumentum/LITE: 187x
BE Semiconductor/BESIY: 112x
BBAI/AI (companies like this will just fall through the floor; no earnings)
Most companies will survive a humbling on the market, just as most survived the Dotcom Bubble. Some never returned to their former highs and languished, over-extended fiber optics infrastructure companies were bought up on discount after they bankrupted, and many companies eventually came back. Many of the smaller companies are dependent on continuous subsidies, just like they were back then. If liquidity taps out, they go under.
Sam Altman and the rest of the AI cult have done a fantastic job of selling hype, many people are genuinely convinced that AI is taking over and an endless stream of money needs to be burned to build it out. AGI goalposts keep getting moved around.
https://arstechnica.com/information-technology/2025/01/sam-altman-says-we-are-now-confident-we-know-how-to-build-agi/
https://www.inthacity.com/blog/tech/will-we-see-agi-by-2028-sam-altman-timeline/
Even for menial or repetitive tasks, hallucinations and errors are extremely commonplace. Sure, it is a new technology that needs refinement, though keep in mind, there is currently no proposed technical solution for hallucination. Considering how much money is being spent on it, AI needs to work incredibly well to generate the profits necessary to recoup spending. It absolutely has its uses, but with all the layoffs happening I swear most of them are just off-shoring. It sounds better on paper and maybe companies are hoping AI can help fill in the gaps. There is no way the tens of thousands or more jobs that have been cut as a result of AI are not mostly cost cutting or off-shoring. I’ll admit this is just a theory.
The meat and potatoes; revenue. We all know OpenAI lives on handouts and borrowed time. They already pre-emptively asked the government for a bailout before even truly needing one. Altman is some sort of modern alchemist, his company is not even remotely close to generating a profit and he keeps getting hundreds of billions in funding. I have to respect the guy. I keep bringing up OpenAI since they are planning one of the largest IPOs in human history. Companies like Google, Microsoft, Meta, etc, are able to fund the loans they take out thanks to their other ventures. How long until they begin to actually generate a profit on all these data centers? Real profit from real customers too, not just the circular economy most of us are familiar with. Did anyone else see Satya Nadella complaining that people are not adopting AI fast enough at Davos? (https://www.techspot.com/news/111012-satya-nadella-warns-ai-must-go-mainstream-avoid.html) How are chatbots and LLMs going to make hundreds of billions of dollars and when? Don’t forget, most tokens are currently very heavily subsidized by these companies to facilitate user adoption. Wonder how much they would have to charge without them, as well as enough to make a profit….
Another point, data centers cost a ton, just putting them up. Chips depreciate very quickly, in the grand scheme of things, which means you’re either going to need AI to generate enough profit every 4-5 years to buy new sets of chips, or defer things and take out even more loans. This brings up another very big problem: liquidity. Most of the AI buildouts are paid for with loans. Did you know that ever since the Great Recession banks have become very risk averse to giving large loans? Private equity is covering where traditional lenders are unwilling to. This doesn’t change much in practice, aside from higher borrowing costs. Private Equity is sure to keep giving out loans if they think they will get their money back. The problems will start if smaller companies start losing funding, are unable to meet debt obligations, and end up defaulting on their loans. This has the possibility of triggering a domino effect. Future funding for AI companies may be considered too risky, a specific lender may become “junk status” or collapse as a result, other lenders may get spooked and constrict additional liquidity in an attempt to protect themselves. This isn’t a novel idea, it has happened many times before in stock market history. Liquidity is essential for debt based expansion. Every time it dries up, things get ugly. You’d probably see another tax-payer funded bailout if things got truly ugly. Occupy Silicon Valley?
Let’s talk about all the different ways AI is making money and losing money.
OpenAI is forecasted to accrue $143 billion in negative cumulative free cash flow by 2029.
https://www.emarketer.com/content/openai-forecast-143-billion-loss-raises-stakes-ai-monetization
The company hit $20 billion revenue in 2025, this would be impressive if didn’t have to burn $17 bilion in cash.
https://www.theneuron.ai/explainer-articles/openai-hits-20b-in-revenue-with-a-17b-problem/
Best estimates are cash flow positive by 2030. Not profitable.
Cracks appear to already be forming
The four top hyperscalers, alone, expect to spend $700 billion on AI this year.
Here are some 10-K forms from the big boys
Microsoft is vague about how much money its derived in revenue from AI, as it does not seperately track AI revenue, it only claims that AI increases productivity. No estimated on ROIs are given here.
Amazon also does not separately account for AI revenue, it is attached to seperate business segments
https://companiesmarketcap.com/amazon/sec-reports-10k/0001018724-26-000004/
Google also does not mention ROIs nor does it distinguish revenues from Gemini from its other services.
https://s206.q4cdn.com/479360582/files/doc_financials/2025/q4/GOOG-10-K-2025.pdf
Meta attributes its boom in advertising revenue to AI implementation, but no specifics are given on exactly how much impact AI had and how much money they’ve made on things like AI glasses through RealityLab. Couldn’t have been much, since only $2.2 billion came in through revenue there.
https://d18rn0p25nwr6d.cloudfront.net/CIK-0001326801/451618c0-0616-4018-ad82-17b05b50492d.html
All these 10-K forms are quite optimistic about AI. None of them actually breakdown or even estimate how impactful AI has been with real numbers. How many people subscribed to Microsoft Office since CoPilot was shoehorned into it? How much profit did adding it to Windows generate?
I don’t doubt AWS and Azure have made a killing hosting AI/LLM training, or the other circular economy shenanigans that have taken place. What is AI going to provide to the average consumer that they can’t live without? Consumer and commercial use cases will have to explode to ensure expenditures become profits. It seems odd that AI revenue is not tracked seperately and is instead packaged with existing products. One would think that if there’s something great to talk about, it would take the stage, front and center. AI is “making money” but its unclear how much exactly.
Another big issue is just powering all this crap. The word gigawatts gets thrown around a lot these days, and nuclear power is expected to be here to save us. Did you know that the most recent nuclear power plant system built in America, the Vogtle Electric Generating Plant, took 10-11 years to build, cost around $24 billion, and generates ‘only’ 3.4 gigawatts? There are people out there suggesting we double the electrical capacity of the U.S. to power the AI boom. Maybe you think the 10-11 year process is mostly red tape, which can be pushed along by a motivated government, but do you seriously think that the construction of nuclear power plants should be rushed just to satisfy Wall Street? Keep in mind you’re gonna need a lot of these to power all these data centers all over the country. 2025’s Project Stargate is planned across the country, with 5 different locations and nearly 7 GW of planned power needs. You’d need roughly one nuclear reactor per site to feasibly power it all. 10 GW is the ultimate goal. The logistics for powering all this, even if everything else that was mentioned went over smoothly, are potentially a decade away. Just to really send this point home, the power necessary to actually run all of this is currently non-existent and can, best case scenario, take YEARS to get AI to the peak production capacity, where it is supposed to begin to be powerful enough to run the economy and make record profits.
There is absolutely no way on earth that investors and lenders will continue to burn billions or trillions waiting for the infrastructure to come together, maintaining companies that do not generate any profit. Investors run at the FIRST sign of trouble, there’s no loyalty in this game, expect for profit. Do not buy into the religious beliefs proselytized by Altman and his fellow grifters. We are far, far, far away from AI approaching anything close to “””godlike”””. To the “this isn’t a bubble, its different” people: okay buddy.
Don’t undermine the narrative, say the high priests
https://www.techspot.com/news/110879-jensen-huang-relentless-ai-negativity-hurting-society-has.html
https://fortune.com/2026/03/04/sam-altman-jensen-huang-only-themselves-to-blame-ai-scare/
Oh right, did anyone notice how many AI ads there were during the Superbowl and have been coming out in general? Did you know that during the Superbowl of 2000, huge amounts of ad space was purchased by companies with 0 profitability?
This is a quote from an investor back in April of 2000, before things really slid out of control with tech.
"If the sentiment really turns negative, there is no floor on companies that don't have earnings you can point to".