So Is This a Bubble or Isn't It?

I have heard and read this question more times than I can count over the last two years. Every market selloff, every Nvidia earnings beat, every AI startup raising at a valuation that makes no obvious sense — it comes back. Is this a bubble? Are we living through the dot-com crash in slow motion? Should I get out? And here is the honest answer: you are asking the wrong question.

The railroads were real. Most of the schemes were not.

In the 1840s, Britain went through something called Railway Mania. Between 1844 and 1847, Parliament received applications for over 600 railway schemes. Promoters raised enormous sums of capital from the public. Newspapers ran breathless coverage of a technology that was going to change everything — and it was going to change everything, that part was true. The railroad was genuinely transformative infrastructure. It remade commerce, shrank distances, and built the industrial backbone of a nation.

Most of those 600 schemes were never built. Their promoters collected fees and disappeared. Investors who couldn't tell the difference between a real railroad and a credible-looking certificate representing one were wiped out. The railways that did get built survived, consolidated, and became exactly what the promoters promised they would be. Two outcomes, same mania, same moment in history.

The United States did it again in the 1870s. The transcontinental railroad was real. The Credit Mobilier scandal — where insiders billed the construction company at wildly inflated rates while ordinary investors held worthless paper — was also real. Both existed simultaneously. That is always how it works.

The question is never whether the technology is real. The question is which part of it you are actually holding.

What a real bubble actually looks like

Before we can talk about whether AI is one, we need to calibrate on what a real bubble looks like in the data.

The dot-com peak in March 2000 is the most useful reference. At its height, the Nasdaq traded at a price-to-earnings ratio of around 200. That is not a typo. For every dollar of earnings these companies were producing, investors were paying two hundred dollars. The Nasdaq-100 specifically — the index of the largest tech companies — sat at a forward P/E of roughly 60x. Only about 14% of tech companies that had gone public were profitable at all. Pets.com had run a Super Bowl ad. Webvan had raised $375 million and built a national grocery delivery infrastructure before it had proven a single market. The ratio of narrative to receipts was essentially infinite.

Today, the same index trades at roughly 26x forward earnings. That is high by historical standards, no argument. But it is less than half the multiple investors were paying in 2000 — and this time, the earnings are real. The companies holding up the Nasdaq today — Alphabet, Microsoft, Apple, Meta, Amazon — are generating billions in actual profit per quarter. These are not Pets.com. The comparison collapses the moment you look at the balance sheets.

That does not mean everything is fine. It means the bubble framing, applied to the whole of AI, is imprecise in a way that costs you good decisions.

The receipts

[WIDGET GOES HERE — dot-com vs. AI leaders comparison table]

The Argument — Data CheckThen vs. Now: What the Numbers Actually SaidDot-com peak (2000) versus AI-era leaders (2025–2026). Same question, very different balance sheets.

This is where I want to be specific, because "trust me, AI is different" is exactly the kind of thing people say at the top of every cycle. So let's look at what the companies actually reported to the SEC.

Alphabet ended Q4 2025 with Google Cloud revenue up 48% year-over-year to $17.7 billion — in a single quarter. That growth is explicitly attributed in their filing to enterprise AI infrastructure and AI Solutions demand. Cloud's annual revenue run rate has crossed $50 billion. AI Overviews now serves 1.5 billion users per month, and those queries are monetized. This is not a projection. It is in the 10-Q.

Microsoft disclosed at the close of 2024 that its AI business had already surpassed a $13 billion annual revenue run rate, up 175% year-over-year. By early 2026, that figure had grown to $37 billion annualized — up 123% again — with over 20 million paid Microsoft 365 Copilot seats. Azure, which runs much of the AI infrastructure underneath the industry, surpassed $75 billion in annual revenue. Satya Nadella did not say AI was going to be important. He reported the number.

Snowflake runs on a consumption model — meaning customers pay for what they use, and usage tied to AI workloads shows up directly in revenue. Product revenue hit $1.23 billion in Q4 fiscal 2026, up 30% year-over-year, with remaining performance obligations growing 42% to $9.77 billion. Over 9,100 customer accounts are now using AI products. On May 28, 2026 — the day this piece published — Snowflake reported its strongest sequential dollar growth in company history, with shares surging nearly 40%. The market was not reacting to a story. It was reacting to a number.

Palantir built its Artificial Intelligence Platform — AIP — and U.S. commercial revenue tied to it grew 121% year-over-year in Q3 2025. They closed 204 deals worth over a million dollars in a single quarter. Total contract value hit a record $2.76 billion, up 151% year-over-year. Their Rule of 40 score — a combined measure of growth and profitability used across the software industry — came in at 114. The Rule of 40 benchmark for a healthy software company is, as the name suggests, 40.

None of this means these stocks are cheap. None of it means they cannot fall. What it means is that the revenue is not imaginary.

The trust-me-bros

Here is where the railroad parallel earns its keep.

In every transformative technology cycle, the genuine infrastructure attracts a second layer: companies that raise capital on the idea without having a business attached to it. They are not necessarily fraudulent. Some of them genuinely believe the pitch. But belief and a business model are different things, and in the early years of any technology wave, that difference is easy to obscure.

Since 2022, the market has seen a wave of companies add "AI" to their product descriptions, their press releases, their investor decks, and in some cases their actual names — without disclosing any AI-attributable revenue, any proprietary model, any data moat, or any credible path to profitability that depends on AI specifically rather than on conventional software they were already selling. The pitch is often some version of: AI is going to transform everything, we are in that space, therefore we will benefit. That is not a business plan. That is a location on a map.

Parliament approved over 600 railway schemes in the 1840s because promoters understood that if the underlying technology was real, the schemes didn't need to be. Investors would fund proximity to the idea. That dynamic did not end with the Victorians.

The distinction matters because these two layers — the real infrastructure companies and the proximity plays — are often discussed as a single thing called "AI." When the proximity plays correct, people write pieces about the AI bubble. When the infrastructure companies keep compounding, people write pieces saying AI is different. Both are happening. They are just happening to different companies.

The actual answer

The technology is real. The railroads got built. The internet connected the world. And AI, specifically large language models and the infrastructure that runs them, is already embedded in enterprise workflows at a scale that is now measurable in quarterly earnings.

The bubble — if you want to use that word — is in the fringe. It is in the companies whose AI strategy consists entirely of an API wrapper and a press release. It is in the valuations assigned to businesses that have told a good story without producing a revenue line that corresponds to it. That fringe will correct. Some of it already has.

But the question "is AI a bubble?" treats a technology layer and a speculation layer as the same thing. They are not. The dot-com bubble did not prove that the internet was a bad idea. It proved that most of the companies claiming to benefit from the internet were not actually building anything. Twenty years later, the internet runs everything. The companies that survived — Amazon, Google — did so because they had the receipts.

Before you decide whether you are in or out of AI, find out what you actually own. Not what sector it is in. Not what the press release says. What does the company report as AI revenue, in an actual filing, to the actual SEC? If that number does not exist, you are not invested in AI. You are invested in the hope that someone else will figure out how to make AI pay before the money runs out. That is a different bet entirely, and you should know you are making it.

This is not investment advice. The companies referenced are used as illustrative examples of how AI revenue is and is not disclosed in public filings. Do your own research before making any investment decisions.

A note on the data.

Dot-com era valuation figures sourced from Goldman Sachs Research (via Marcus by Goldman Sachs), Nasdaq.com and Wikipedia's Dot-com bubble entry. Nasdaq-100 forward P/E at the 2000 peak sourced from IntuitionLabs AI/Dot-Com comparison analysis citing VisualCapitalist. Current P/E comparison sourced from the same IntuitionLabs analysis and Nasdaq.com. Alphabet revenue figures from SEC Form 8-K filings, Q4 2024 and Q3/Q4 2025. Microsoft revenue and AI ARR from SEC Form 8-K filings, Q2 FY2025 and Q3 FY2026 earnings releases. Snowflake product revenue from Snowflake press releases for Q4 FY2025 and Q4 FY2026. Palantir AIP and U.S. commercial revenue from SEC Form 8-K filings, Q3 2025 earnings release. All figures cited reflect the most recent available earnings as of publication date.