Panmure Liberum warns the AI bubble could burst within two years

Panmure Liberum has put a number on it for the market: two years. According to analysts at the British brokerage, that is the horizon within which the bubble around AI companies could burst. Joachim Klement, managing director and analyst at Panmure Liberum, said so on Bloomberg Open Interest. He laid out the risks in AI stocks and discussed the likelihood of a severe market crash that this very sector would set off.

The publicly available part of the Bloomberg piece is short. The teaser gives no specific index levels, no estimates of the decline, and no list of companies. So below we separate what was actually said from the context that helps gauge how much weight the warning carries.

What the statement actually says

There are few key facts, but they are unambiguous:

• Author of the forecast: Joachim Klement, Panmure Liberum.
• Horizon: no more than two years.
• Scenario: not a correction, but a severe crash with AI as the trigger.
• Venue: Bloomberg Open Interest, Bloomberg Tech, October 9, 2026.
• Target of the criticism: stocks of AI-related companies. The teaser does not disclose details on names or valuation multiples.

Klement has long been known as a skeptic on market valuations and investors' behavioral errors. That does not make his forecast right, but it explains the tone: he views the AI sector through the lens of historical cycles, not the quality of the models.

Why the conversation inevitably goes back to 2000

The dot-com analogy suggests itself. Back then, investors called the technology correctly: the internet did transform the economy. But they paid for future revenue long before it existed. The Nasdaq index lost about 78% from its March 2000 peak to its October 2002 low. Telecom companies built fiber infrastructure with room to spare, and demand caught up with it only years later.

Today's AI cycle is similar in one respect: the main burden has fallen on capital expenditure. Data centers, accelerators, and energy contracts are being paid for now, while revenue from LLM-based and agent-based products is growing more slowly than costs. The two-year horizon coincides with the point when customers will have to show a return on these investments.

Where the system could break

If a bubble exists, it rests on a few assumptions. First: demand for compute will keep growing at the same pace. Second: customers will agree to pay noticeably more for AI features than they pay now. Third: credit markets will keep financing capacity buildouts on acceptable terms. If any one of the three links fails, the damage runs down the chain: first chip and hardware suppliers, then data-center operators, then startups living on venture money.

Concentration is a separate risk. When a small number of companies drives the performance of broad indexes, a slump in the AI sector stops being one sector's problem. It hits pension funds and index ETFs, that is, people who never made a conscious bet on AI.

How to read such forecasts in practice

A single forecast is not a fact. Analysts have predicted bubbles before, and their timing has often been wrong. It is more useful to watch verifiable indicators than loud headlines.

For teams building products on AI:

• Plan for a scenario in which funding gets more expensive or disappears for 12–18 months. Check whether your runway is enough.
• Do not tie your unit economics to today's subsidized prices for APIs and cloud GPUs.
• Keep your architecture portable across model and cloud providers.

For investors and executives: watch the ratio of capex to revenue at the large cloud providers, the terms of debt financing for data centers, and the pace of real AI adoption in customers' business processes. If these three indicators diverge, the Panmure Liberum warning becomes more relevant. If they converge, skeptics will have fewer arguments.