If you ask traders what they're rubbing against the most at the moment, it's rarely strategy and rarely timeframe. It's the question of whether what's happening in equity markets right now is a replay of 1999 or something fundamentally different. And depending on whom you listen to, you get a very confident answer in either direction.
We pay attention to the AI-bubble-vs-dotcom question for two reasons. First, because if equity valuations pop, crypto goes with them — we've seen that pattern three times in a row by now. Second, because history rarely plays the exact same tune, but often enough rhymes with existing patterns that it's worth listening.
In this post we look at what really echoes 1999 about today's situation — and where the picture is fundamentally different. Without falling into either camp.
What echoes 1999
If you've read the last twelve months of headlines, you know the list already. The biggest parallels, in order of visibility:
Market concentration. The seven largest tech stocks make up over thirty-five percent of the S&P 500. That's a higher concentration than at the dotcom peak in March 2000. Listen to the index, and you're really listening to a handful of companies — and that handful is largely living off the same story.
Valuations without profits. OpenAI is at a valuation of seven hundred and thirty billion dollars. Profitability isn't expected before 2030. The valuation corresponds to roughly fifty-six times current revenue. These are numbers that would feel right at home in the 1999 vocabulary.
Circular deals. OpenAI buys compute from Microsoft, Microsoft invests in OpenAI, OpenAI has supply deals with Nvidia, Nvidia invests in OpenAI, Oracle has a major contract with OpenAI. Draw the arrows on a whiteboard and you get a closed loop. Similar structures existed in 1999 in the telecom and internet-hardware world — the question then as now: who's actually buying value creation here, and who's buying their own revenue further down the chain?
A CapEx wave that's historic in scale. The major cloud providers — Microsoft, Google, Amazon, Meta, Oracle — together plan to invest around six hundred and fifty to seven hundred billion dollars in AI infrastructure in 2026. That's roughly a sixty to sixty-seven percent increase year over year. Outside of wartime, the US economy has never seen an investment program of this size.
A study that raises eyebrows. In early 2025, MIT examined how many companies actually generate measurable returns from their generative AI projects. The result: ninety-five percent come in at zero return. That's not the same as "AI doesn't deliver" — many of these projects are pilots, and pilots rarely produce direct revenue. But it's a number that's hard to ignore while at the same time sixty-one percent of global venture-capital funding in 2025 flowed into AI companies.
These five points together paint a picture that genuinely echoes the weeks before the March 2000 peak. If you stopped here, the conclusion would be obvious: this ends like that.
But in four places the picture is fundamentally different
We think the parallels above are real, but they paint an incomplete picture. In four places today's situation differs structurally from the year 2000 — and these differences aren't cosmetic, they're fundamental.
First: The companies actually make money
In March 2000, Cisco was trading at a price-earnings ratio of around two hundred. Pets.com had no profit. Webvan had no profit. The valuations of the time's market leaders were almost entirely justified by a story about the future.
Nvidia, in fiscal year 2026, did two hundred and fifteen point nine billion dollars in revenue with net income over one hundred and twenty billion. The P/E ratio sits at around fifty-seven. Microsoft, Google, Meta — all companies whose AI business is profitable today, not five years from now.
The Nasdaq-100 is currently trading at a P/E of around thirty-three. In March 2000, the same metric stood at around sixty. That's a factor-of-two difference, and it's the single most important number in this whole debate.
Does that mean valuations are "fair"? No. Does it mean a sharp correction can't happen? Also no. But it means: if AI stocks fall forty percent, that's a painful correction, but the companies underneath are still cash machines. When dotcom stocks fell forty percent in 2000, the entire business model was often gone.
Second: The investments are paid from profits
This is the point we ourselves found most underrated while researching.
In 2000, the internet pioneers' CapEx wave was largely funded through debt and equity issuance. The CapEx-to-free-cashflow ratio across the broad Russell 3000 index sat at around four — meaning companies were spending four times what they earned. The rest they borrowed or raised in capital markets.
Today, that ratio is below one. Microsoft, Google, Amazon and Meta spend money on their AI infrastructure that they themselves earn. Semiconductor manufacturers reinvest about sixty percent of their distributable profits into research, development and capacity expansion — that's a high ratio, but it's a ratio that comes from ongoing earnings, not from debt markets.
Why this matters: a bubble inflated with debt pops differently than an investment wave funded from equity. In the first case, the crash drags balance sheets down and pulls the banking system in with it. In the second case, companies can simply scale back their investments without their existence being in question. That's an entirely different kind of shock.
Third: The infrastructure is being used
In the dotcom era, US telecoms laid fiber-optic cable at a pace that massively exceeded contemporary demand. Estimates suggest around ninety-seven percent of that fiber sat unused for years — the famous "dark fiber." Telecom companies had built capacity nobody was buying.
In the AI era, it's the opposite. Nvidia ships its high-end chips as fast as they can be produced, and waiting lists stretch quarters into the future. Microsoft, Google and AWS sell compute they don't yet have at the time of sale — the data centers are under construction, the customers are waiting. GPUs aren't just being used, they're being thermally stressed because they run around the clock. There's no such thing as "dark GPUs."
Does that mean demand will exceed supply forever? Certainly not. When the investment wave peaks and the capacities come online, the picture can flip — some analysts expect exactly that for 2027 or 2028. But right now the situation is different from the telecom overcapacity of 1999.
Fourth: Monetary policy points in the opposite direction
In 2000, the US Federal Reserve was raising rates aggressively to cool an overheated economy. Greenspan had hiked the federal funds rate from four point seven five percent in mid-1999 to six point five zero percent in May 2000 — a tightening of one hundred and seventy-five basis points within eleven months. That was one of the main triggers for the bubble popping: money got more expensive, discounted values of future profits fell, valuations broke.
Today, the Fed is in a cutting cycle. The federal funds rate currently sits at three point seven five to four percent, with consensus expecting further moves down. That's exactly the opposite of the conditions under which the dotcom bubble burst. Money is getting cheaper, not more expensive.
On top of that, US money market funds currently hold over eight trillion dollars — a historic record. That capital is hunting for yield and will, once money market rates drop below a critical threshold, rotate into risk assets. This liquidity reserve didn't exist in 2000. Back then the economy was more fragile on the debt side, and there was no "dry powder on the sidelines" of this magnitude.
Where real risks still sit
We don't want to run this comparison so that the conclusion reads "everything's different, everything's safe." There are three places where, despite all the fundamental differences from the dotcom era, things still feel uneasy.
First: CapEx fatigue. In recent quarters a pattern has become visible that wasn't there two years ago. When a cloud provider announces a massive increase in its AI investments, the stock no longer reacts with enthusiasm but with selling pressure. When Alphabet announced in early 2026 that it would spend one hundred and seventy-five to one hundred and eighty-five billion dollars on AI infrastructure in 2026, the stock dropped immediately. That's a changed market reaction. Investors are now asking whether these investments will show up in profits, and they're no longer prepared to take that on faith.
Second: The gap between investment and return. The ninety-five percent zero-return number from the MIT study is a snapshot. If that number doesn't improve meaningfully in two to three years, the reality test will arrive with full force. Then the cloud providers will be sitting on hundreds of thousands of GPUs whose utilization has to be paid for, while the paying end-customers are missing. That's exactly the pattern that sent the telecom overcapacity into the crash back then.
Third: Concentration risk. When the seven largest companies carry the market and one of those seven takes a hit, it drags the whole index. In January 2025, the appearance of DeepSeek — a Chinese AI model that achieved comparable results with significantly less compute — erased around six hundred billion dollars in market value in a single day. Such efficiency shocks can become a cascade in a concentrated market. That kind of concentration risk didn't exist in 1999 in this form — Cisco and Microsoft were large, but not as dominant as Nvidia is today.
The lesson from history here
At the Backtesting Arena we follow a simple discipline: before we trust a thesis, we rebuild it and check it against data. Macro comparisons are harder than trading strategies — we can't backtest the dotcom-vs-AI comparison, because each comparison only exists once. But we can look at the conditions that triggered the crash back then and check whether those conditions hold today.
If we line up the four triggers of the March 2000 crash — overheated valuations without profits, debt-financed overcapacity, monetary tightening, breakdown of trust in the future narrative — then today two of them clearly aren't present (profits are there, financing comes from internal cash flow), one is pointing in the opposite direction (the central bank is cutting, not raising), and one is open and potentially important (the narrative is alive, but it has shown first cracks — CapEx fatigue, efficiency shocks).
Three of four triggers aren't there. One is there, but not yet ripe. That's not "nothing can happen." That's "something can happen, but it would look different from 2000."
The most likely scenario given the data: a longer, late-stage expansion with elevated volatility and sharper sector rotations. Such phases can run longer than skeptics expect — the market can talk itself into its own optimism for quite a while. And they often end more abruptly than optimists think — usually not because the P/E got too high, but because the "E" suddenly stops growing.
The one or two variables we're watching closely in this context:
- CapEx-to-revenue conversion starting in 2027. If the six hundred to seven hundred billion dollars spent don't show up in corresponding revenue growth, the picture shifts materially.
- Stock reactions to future CapEx announcements. If the pattern intensifies — markets greeting CapEx increases with growing selling pressure — that's a clear early signal that confidence is tipping.
- Utilization of the new data centers coming online in 2026 and 2027. If utilization rates drop, there's a second wave of "dark GPU" concerns. That would be the direct echo-parallel to 1999, just shifted in time.
What you can take away
We don't give investment advice here — we offer five observations we ourselves take from this comparison, as workshop notes, not as a recipe:
- Blanket "It's like 2000" or "It's nothing like 2000" statements are both too simple. The first ignores structural differences, the second ignores structural parallels. Reality sits between them.
- Concentration is the biggest single risk. If your portfolio is implicitly a bet on three to five companies, it's a bet on three to five companies, no matter how many ETFs you hold.
- Debt-financed speculation is today less an equity-market thing and more a crypto thing. The high leverage right now doesn't sit at Microsoft and Nvidia, it sits in trader accounts at crypto exchanges. That shifts the crash risk into a different market — and that market is correlated with equities.
- The central bank is on your side until it isn't. As long as the cutting cycle runs, the environment stays supportive. The moment the Fed has to tighten again — because inflation comes back, for instance — the whole picture shifts.
- Watch the conversion gap. If AI investments don't translate into revenue and profit over the next two years, the narrative wobbles. That's the one variable that's really decisive for the overall picture — everything else is a side condition.
History rhymes, but it doesn't repeat. Whoever understood 2000 has an advantage in reading today's situation — but no algorithm that tells them what happens next.
Further reading:
→ Strategy Insights — what the data shows right now → Previous post: 7,000 Backtests Insights
Backtesting Arena · Study the Past — Improve your Future.