Michael Burry's AI-bubble warning tests Wall Street's momentum trade

Michael Burry said the AI-led stock rally was starting to resemble the final months of the dot-com bubble just as the Philadelphia Semiconductor Index extended a 70 per cent rise from the end of March. His warning lands as more than another famous-bear soundbite because of that timing. The issue is not whether demand for chips is real. It is. The issue is whether price action is starting to run ahead of the part of the story investors can still defend with numbers.

In a CNBC interview, Burry said that for any stocks going parabolic investors should “reduce positions almost entirely.” He also told CNBC separately that the market felt like “the last months of the 1999-2000 bubble.” It is framing that shifts the conversation past quarterly results and toward market structure — concentration, crowding and what happens when a winning trade prices perfection.

Reuters’ account of AI leadership narrowing is useful here. Earlier in the cycle, AI enthusiasm lifted a wider set of equities. More recently, Reuters described a market that was no longer lifting all boats — it was sinking the ships that missed the AI narrative. Strong demand. Stretched positioning. Those are not the same thing, but markets often trade as if they are.

What makes the dismissal argument weak is that other seasoned investors hear a similar rhythm. In the same CNBC report that carried Burry’s 1999 comparison, macro investor Paul Tudor Jones said the current environment felt similar to 1999. None of this proves a crash is imminent. It does suggest the rally has reached the point where experienced money managers talk less about upside scenarios and more about the cost of being the last buyer.

Late-stage rallies tend to follow a pattern. Early on, investors buy the theme because earnings can still surprise on the upside. Later, they buy because not owning the theme starts to feel professionally dangerous. The distinction is small on a strong day and obvious on a weak one. The first version of the trade tolerates uneven execution. The second punishes it.

Why the comparison keeps landing

Dot-com comparisons keep resurfacing because the numbers argue both sides. Bulls can point to a semiconductor industry that still has tangible demand behind it. Reuters reported in February that global chip sales were expected to reach $1 trillion in 2026, citing an industry-group forecast. That is not 1999-style vapourware. It is a real revenue base, reinforced by data-centre spending, AI server orders and the supply-chain discipline that did not exist during the first internet bubble.

For the bear case to work, the technology does not need to be fake. Markets just have to overpay for a growth story they have already absorbed. Investopedia’s summary of the move highlighted a Bespoke note showing that only 33 per cent of chip stocks were trading above their 50-day moving average even after the group had surged. That is an awkward detail for anyone arguing the rally is uniformly healthy. A market can post a spectacular index gain while breadth underneath it starts to fray.

Breadth matters here because narrow leadership flatters the indices while hiding an increasingly selective tape. If only the largest names are still pulling in fresh money, the trade stops being a sector call and starts becoming a conviction test on a handful of balance sheets. Volatility feels low at that point — right up until it does not.

Burry’s warning is less a call to short every AI name and more a stress test for the momentum trade. If investors are buying because earnings estimates still need to move higher, that thesis can survive volatility. If they are buying because the chart has turned vertical, the thesis is thinner than it looks. Not because revenue has evaporated — because the marginal buyer may not care which part of the semiconductor stack is generating the cash.

What would prove him wrong

The counterargument is simple: 2026 still has a fundamentals story that 2000 eventually lost. The Bloomberg report on Burry’s warning tied the latest jump to semiconductor shares and to a market still willing to pay for AI-linked growth. If that growth keeps arriving in guidance, margins and order books, investors can argue that elevated multiples are expensive rather than absurd. Real cash flow. Real capex. Real end demand. That is a sturdier base than the story stocks that defined the old bubble’s weakest edge.

What it leaves investors with is a checklist shorter than bubble rhetoric implies: earnings revisions, backlog and the market’s reaction to merely good results. If estimates keep rising and second-tier beneficiaries start participating again, the trade can broaden rather than snap. If misses get punished harder and only the biggest names hold up, Burry’s complaint about parabolic behaviour starts to look less like theatre and more like tape-reading.

Still, the lasting lesson from Reuters’ “sinking ships” framing is that this rally may already be behaving like a mature thematic trade. Leadership can stay narrow for longer than sceptics expect. It can also become less forgiving. When too much performance is concentrated in a few bellwethers, a good quarter is no longer enough; the market starts demanding blowouts, cleaner guidance and immediate proof that the AI build-out is still accelerating. At that point a narrative stops broadening and starts policing itself.

Burry’s warning lands as a reminder about conditions, not as prophecy. He is not arguing every semiconductor stock is a fraud. He is saying that when price turns parabolic, risk management matters more than storytelling. For Wall Street, the real comparison with 1999 is not the technology. It is the behaviour around it. When investors begin paying up for the certainty that a winning trade will keep winning, the trade is no longer just about fundamentals.