The sharp selloff in AI leaders may not be the crash itself — it may be the market warning investors have been waiting for.
AI leaders fall as dot-com risk returns

That is the sobering lesson from the dot-com bust, and it matters because the biggest names in today’s artificial-intelligence boom are already taking serious damage. Oracle has fallen more than 60% from last year’s highs, CoreWeave is down more than 70%, and Nvidia, even after years of extraordinary gains, has also slipped more than 60% from its peak before a post-earnings bounce faded.
The reason this matters goes beyond a few battered stock charts. When speculative leaders lose altitude, the question for investors is whether the entire market is simply re-rating, or whether a much bigger unwind is starting beneath the surface. History says those can be very different things. In 2000, the Nasdaq Composite bounced about 35% in the spring before tumbling more than 40% in only about three months. Investors at the time had plenty of reasons to stay bullish: the internet was real, the strongest tech companies were still growing, and dip-buying had worked before. None of that prevented the bear market from becoming far worse.
That is why Dario Perkins of TS Lombard is drawing a direct line between the current AI trade and the first year of the dot-com bust. His argument is not that the boom is already over. It is that the explanations investors use to defend it sound eerily familiar. They point to strong demand for computing power, to shortages of GPUs, to the idea that only the most profitable leaders deserve capital, and to the belief that the Federal Reserve cannot meaningfully hurt companies with this much earnings power. Those were the same arguments investors made in 2000.
Perkins says the most dangerous sign is not exuberance itself, but bottlenecks. In the late stages of the dot-com era, shortages looked like proof that demand was unstoppable. In reality, they were a prelude to a glut. The companies buying all that hardware were often dependent on IPOs, venture funding and junk bonds. Once that funding dried up, orders dried up with it. The leaders kept reporting strong results for a while because backlogs lagged the real economy, and that is exactly what made the turn so painful.
That is the key lesson for investors today. Profitable AI leaders can still be vulnerable if their customers are weak or overleveraged. The market often assumes that if Nvidia, Oracle or Microsoft are still printing money, the cycle must be fine. But the real risk is not just valuation. It is whether the spending boom that supports the earnings boom is sustainable.
There are differences from 2000, of course. TS Lombard says valuations are nowhere near the extremes of the dot-com era, when the Nasdaq’s price-to-earnings ratio topped 80, versus roughly 30 today. That makes a full replay less likely. But lower valuations do not eliminate cyclical risk. They only make the mechanism different. In this cycle, the bigger threat may be that earnings expectations themselves prove too optimistic if financing tightens and data-center demand cools faster than expected.
For broad-market investors, the most important takeaway is that a market can look resilient right up until it does not. In 2000, the S&P 500 was still near record highs six months into the Nasdaq’s slide. The economy and the market did not crack together. The market led the way down, then the economy followed. That is why a wave of AI weakness should not be dismissed just because the rest of the market has not fully rolled over yet.
The recent price action in the major indexes backs up that caution. The S&P 500, Nasdaq 100 and small-cap Russell 2000 have all been choppy, with technical indicators such as the 50-day moving average, RSI readings and MACD showing weakening momentum in the growth-heavy parts of the market. Adalytica’s trade-signal snapshot for the S&P 500 shows neutral sentiment but fear still lingering in awareness, a reminder that investors have not settled into a clean risk-on mood.
None of this means long-term investors should abandon technology. It does mean they should stop treating every pullback as an automatic buying opportunity. The best way to invest through a period like this is still the same: own a diversified portfolio, think in years rather than weeks, and focus on companies with durable cash flow, not just exciting narratives. If the AI buildout is truly a multi-year infrastructure cycle, the winners will eventually be the businesses that can keep investing through volatility, not the ones relying on easy financing to stay in the game.
| Entity | Gains | Losses |
|---|---|---|
| Long-term diversified investors | ▲Better entry points | ▼Short-term volatility |
| AI leaders with strong balance sheets | ▲Survive the cycle | ▼Slower multiple expansion |
| Highly leveraged AI suppliers/customers | ▲Short-term access to demand | ▼Refinancing risk, weaker orders |
| Dip buyers chasing every pullback | ▲Brief rebounds | ▼If earnings estimates roll over |



