At 1pm (BST) today, Lex head John Foley and tech comment editor Elaine Moore will take questions in a live Ask an Expert Q&A on a question that has become impossible to avoid: has the market lost its mind over AI?
It’s a deceptively simple prompt. “AI” is now shorthand for everything from chip demand and cloud capex to software redesign, labour-market anxiety, regulatory risk and the next wave of consumer products. Yet the market’s behaviour often looks less like a careful assessment of those moving parts and more like a single, self-reinforcing narrative—one that can swing from euphoria to alarm within weeks. The result is a landscape where investors are trying to price not only technology, but also timing, adoption, productivity gains, and the political permission structure that determines how quickly deployment can scale.
The live discussion is likely to focus on what’s actually driving sentiment right now, why markets react so sharply to AI headlines, and whether today’s valuations reflect near-term progress or longer-term potential that may be arriving faster—or slower—than many expect. But the deeper issue is whether the market is responding to fundamentals or to a kind of collective forecasting error: a tendency to extrapolate the most visible successes while underweighting the frictions that determine real-world impact.
To understand why this question has traction, it helps to look at how AI has moved from being a research story to a capital allocation story. In the early phase, the debate was about capability: could models do useful work? Then it shifted to infrastructure: could companies build and afford the compute required to run them at scale? Now the conversation is increasingly about deployment and economics: can AI deliver measurable returns, and if so, how quickly?
That shift matters because markets don’t just price “AI exists.” They price the path from existence to revenue. And that path is rarely linear. It runs through procurement cycles, integration into existing workflows, data governance, security reviews, model evaluation, and—perhaps most importantly—change management inside organisations that have to decide what to automate, what to augment, and what to leave alone.
When those steps go smoothly, the market tends to reward the winners quickly. When they don’t, the correction can be brutal. That’s one reason AI sentiment can feel extreme: the upside is concentrated in a handful of companies that appear to control key bottlenecks, while the downside is dispersed across everyone else who is still trying to prove that the technology can be made profitable at scale.
One of the most persistent drivers of AI market momentum is the perception of scarcity. Chips, advanced packaging, high-bandwidth memory, power capacity, and the ability to secure supply contracts have all become part of the AI investment thesis. Even when demand is strong, the market often treats capacity constraints as a near-permanent feature rather than a temporary phase. That can create a feedback loop: strong demand leads to aggressive capex; aggressive capex leads to expectations of sustained demand; sustained expectations justify further capex. In such an environment, valuations can detach from near-term earnings because investors are effectively underwriting a multi-year industrial ramp.
But scarcity is not the same as permanence. Supply chains adapt. Competitors build. Efficiency improves. New architectures reduce the compute intensity required for certain tasks. And regulation can change the economics of deployment. The market’s challenge is that these adjustments are hard to forecast precisely, and small changes in assumptions can produce large valuation swings.
This is where the “lost its mind” framing becomes both useful and misleading. Useful, because it captures the sense that expectations are sometimes too optimistic. Misleading, because markets are not irrational; they are forward-looking. The question is whether the forward-looking component is anchored to realistic scenarios or to narratives that are difficult to falsify quickly.
Consider how AI headlines function as information shocks. A breakthrough in model performance, a new product launch, a partnership announcement, or a regulatory decision can all move sentiment. Yet the market often reacts as if each headline is evidence of immediate monetisation. In reality, many announcements are about capability or distribution rather than confirmed economics. A model that performs better in benchmarks may not translate into lower costs or higher conversion rates for a specific business line. A new tool may be impressive, but adoption can be slow if it requires workflow redesign or if customers are cautious about compliance and liability.
The gap between “demonstration” and “deployment” is where optimism can outrun reality. Investors may interpret early signals—pilot programmes, user growth, or internal productivity claims—as proof that the technology is already delivering returns. But pilots are not the same as scaled operations. They often involve generous conditions, limited scope, and teams that are highly motivated to make the experiment succeed. Scaling introduces friction: integration costs, monitoring requirements, model drift, customer support burdens, and the need for robust evaluation frameworks.
Elaine Moore’s perspective as a tech comment editor is likely to emphasise this distinction. Tech narratives tend to compress time. They treat the future as if it is already here, because the most visible progress is often in the lab or in product demos. Markets, however, must price the entire system: the cost of running models, the cost of maintaining them, the cost of ensuring they behave safely, and the cost of convincing users to trust them.
John Foley’s focus on markets and policy may bring another dimension: the role of regulation and geopolitics in shaping the pace of adoption. AI is not merely a technological trend; it is a strategic asset. That means export controls, data localisation rules, procurement policies, and national security concerns can all affect which companies can deploy what, where, and when. Even if the technology is ready, the permission structure may lag.
This is one reason the market can swing sharply. When policy signals appear supportive, investors may assume deployment will accelerate. When policy signals appear restrictive, investors may assume a slowdown. Yet policy is often incremental and uneven. Companies may find workarounds, or regulators may carve out exceptions. The market’s difficulty is that it wants binary answers—accelerate or stall—while the real world tends to deliver partial approvals, phased compliance requirements, and sector-specific rules.
Another factor behind the “AI mania” feeling is the way AI has become a general-purpose theme that investors use to justify exposure to multiple sectors at once. In practice, AI is not one market; it is many markets. Some are infrastructure-heavy, some are software-heavy, some are services-heavy, and some are consumer-facing. Each has different adoption curves, different competitive dynamics, and different margins.
When investors treat AI as a single trade, they can overestimate correlation. A company that benefits from AI infrastructure demand may not benefit from AI application demand in the same way. A firm that sells tools to developers may see demand rise even if enterprise buyers are cautious. Conversely, a consumer app may grow quickly but struggle to monetise if users churn or if acquisition costs remain high.
This is why a unique take on the current moment is to ask not “Is AI overhyped?” but “Which part of the AI value chain is priced correctly, and which part is priced as if it will skip the hard parts?”
The hard parts are not glamorous, but they are decisive. For AI to deliver sustained economic value, organisations need reliable outputs, predictable costs, and governance that stands up to scrutiny. They need to integrate AI into processes without creating new failure modes. They need to measure performance in ways that matter to the business, not just in ways that look good in a demo.
There is also the question of productivity versus substitution. Some AI use cases genuinely augment human work and reduce time-to-completion. Others primarily substitute for tasks that were previously done by humans, which can create cost savings but also triggers political and reputational risks. The market may be tempted to focus on the productivity upside while underweighting the social and legal consequences of rapid substitution.
Even when substitution is economically attractive, it can be constrained by labour laws, union negotiations, and public backlash. That doesn’t mean AI won’t be adopted; it means adoption may be slower or more selective than the most bullish forecasts imply.
Then there is the issue of competition and differentiation. AI capabilities are improving rapidly, but differentiation is not guaranteed. If multiple providers offer similar performance, the market may compress margins. Companies that win may do so not because their models are always best, but because they have distribution, proprietary data, workflow integration, or superior customer relationships. Those advantages can be durable, but they are harder to identify early.
This is where valuations can become fragile. If investors assume that “best model” equals “best business,” they may overpay for capability. If the market later realises that monetisation depends on distribution and integration, the re-rating can be swift.
So has the market lost its mind? A more precise answer is that the market is pricing a set of assumptions that are unusually sensitive to timing. AI is moving fast enough that investors feel justified in being aggressive. But it is also complex enough that the path from capability to cash flow is uncertain. When uncertainty is high and the market is confident, valuations can look detached. When uncertainty is high and confidence drops, valuations can correct sharply.
The most interesting angle for today’s Q&A is likely to be how to separate “expectations” from “evidence.” Evidence includes measurable adoption, cost reductions that persist, and revenue growth that holds up under scrutiny. Expectations include narratives about transformation that may be true in principle but not yet proven in practice.
Investors often learn this lesson repeatedly in technology cycles. The difference with AI is that the cycle is happening across multiple layers simultaneously: models, chips, cloud services, enterprise software, and consumer interfaces. That makes it easier for optimism to spread, because progress in one layer can be interpreted as progress in all layers.
Yet the market’s own behaviour suggests it knows something is uncertain. The volatility around AI-related announcements is not just excitement; it is the market constantly updating its beliefs about what will happen next. The question is whether those updates are grounded in fundamentals or driven by momentum.
A useful way to think