Markets have a way of doing two things at once: they chase the future with enthusiasm, and they punish it when reality refuses to arrive on schedule. That tension is at the heart of today’s live Ask an Expert Q&A, where Lex head John Foley and tech comment editor Elaine Moore tackle a question that has become almost impossible to avoid in boardrooms, trading desks and startup pitch decks alike: has the market lost its mind over AI?
The premise sounds dramatic, but the underlying issue is more precise. Investors aren’t just asking whether artificial intelligence is real. They’re asking whether the speed of adoption, the durability of competitive advantage, and the eventual economics of AI are being priced in faster than the evidence can support. In other words: is the market reacting to momentum—or to expectations that have started to outrun fundamentals?
What makes this conversation timely is that “AI” has stopped being a single bet. It’s now a bundle of bets—on chips, cloud infrastructure, data pipelines, model training and inference, enterprise software integration, cybersecurity, regulation, and labour-market disruption. Each component has its own timeline and its own bottlenecks. When markets price the whole bundle as if it were one coherent story, the risk of mispricing rises sharply.
A useful way to frame the debate is to separate three different questions that often get conflated in headlines. First: is AI improving? Second: is AI being adopted at scale? Third: is AI creating profits that show up in financial statements rather than only in product demos and investor decks? The first question is largely answered by the pace of technical progress. The second is increasingly answered by enterprise pilots turning into deployments. The third is where uncertainty remains—and where valuation becomes most sensitive.
That’s why the discussion focuses on what “AI momentum” looks like beyond the noise. Momentum isn’t just a stock chart or a viral product launch. It’s a pattern of spending, hiring, procurement decisions, and measurable performance improvements that persist across quarters. It’s also the willingness of customers to pay for outcomes rather than novelty. When momentum is genuine, you see it in budgets that don’t evaporate after the initial excitement. When momentum is fragile, you see it in short-lived surges of investment that fail to translate into sustained revenue.
Elaine Moore’s angle, as reflected in the structure of the Q&A, is to examine where excitement collides with fundamentals. That collision tends to happen in predictable places. One is the gap between capability and deployment. A model can be impressive in a controlled environment while still failing to meet the reliability, latency, security, and compliance requirements of real business workflows. Another is the gap between experimentation and integration. Many organisations can run a proof of concept; fewer can integrate AI into systems that govern customer interactions, billing, fraud detection, or medical documentation without unacceptable risk.
There is also the question of cost curves. AI economics are not static. If inference costs fall faster than expected, adoption accelerates and margins improve. If costs remain stubbornly high, companies may delay scaling even if the technology works. Markets often price in a particular trajectory for these cost curves. When the trajectory changes—up or down—the repricing can be swift.
John Foley’s contribution, meanwhile, is to explore how expectations, risk, and incentives shape market behaviour. This is where the “lost its mind” framing becomes less about irrationality and more about incentives. Markets are not designed to wait patiently for consensus. They are designed to price information quickly, and they reward early conviction. In AI, early conviction is often based on partial signals: a new model release, a partnership announcement, a capex plan, a regulatory filing, or a surprising benchmark result. Those signals can be meaningful, but they are rarely sufficient to determine long-term profitability.
The incentive problem is compounded by the fact that AI is a platform race. If you believe the winners will capture disproportionate value, you may accept higher valuations today because you’re effectively buying optionality on dominance. But optionality is expensive. When too many investors buy the same optionality at once, the market becomes vulnerable to any disappointment—especially disappointments that are not catastrophic, but merely “less than perfect.”
This is where risk enters the conversation in a more nuanced way. Risk in AI isn’t only about whether the technology works. It’s also about whether the business model works, whether the regulatory environment tightens, whether data access becomes constrained, whether energy and supply chains become binding constraints, and whether competition compresses margins. Even if AI adoption grows, the distribution of profits may shift away from the companies that currently look best positioned.
One unique take that emerges from the framing of the Q&A is the idea that the market may not be pricing AI incorrectly—it may be pricing the wrong unit of analysis. Investors often treat “AI” as a monolith, but the market is actually pricing multiple industries simultaneously. Chips and infrastructure can experience different demand cycles than enterprise software. Cybersecurity spend can behave differently than marketing spend. Consumer AI features can scale differently than regulated-industry deployments. When the market moves as if all these segments share the same timeline, it creates pockets of overvaluation and undervaluation.
That’s why the discussion emphasises “what to watch next” rather than assuming a single outcome. The next phase of AI markets is likely to be defined by evidence of durable adoption and by the emergence of clearer winners in specific categories. For example, the market will pay close attention to whether AI spending shifts from experimental budgets to operational budgets. It will also watch for signs that AI is reducing costs or increasing revenue in ways that are measurable and repeatable.
Another signal that matters is the quality of customer retention. Many AI products can win initial trials. The harder test is whether customers keep paying after the novelty fades and after internal teams learn the real constraints. Retention is a proxy for whether AI is becoming embedded in workflows rather than sitting on top of them. If retention improves, it suggests that the technology is delivering value that survives scrutiny. If retention stalls, it suggests that the value proposition may be weaker than the market assumed.
The Q&A also implicitly raises the question of how much of the current market enthusiasm is driven by narrative versus fundamentals. Narrative matters because it coordinates capital. But narrative can become self-reinforcing: rising stock prices attract more capital, which funds more development, which produces more announcements, which sustains the narrative. This feedback loop can be rational in the short term, but it can also mask the point at which the market’s expectations become disconnected from the pace of monetisation.
In that context, “has the market lost its mind?” can be interpreted as: has the market moved from pricing a plausible future to pricing a near-certain future? There’s a difference. Pricing a plausible future means you accept uncertainty but believe the probability-weighted outcome is attractive. Pricing a near-certain future means you assume the hardest parts—deployment, cost reduction, regulation, and competitive dynamics—will go smoothly. The more the market behaves as if the future is certain, the more sensitive it becomes to any friction.
Friction is inevitable. Even the most successful AI deployments face constraints: data quality issues, integration complexity, governance requirements, and the need for human oversight in high-stakes contexts. The market may initially underweight these frictions because they are less exciting than breakthroughs. Over time, however, frictions show up in timelines and margins. When they do, the market can reprice quickly.
There’s also a broader macro dimension. AI spending competes with other forms of investment. If interest rates remain high or if economic growth slows, companies may become more selective. They might still invest in AI, but they may prioritise use cases with clearer ROI. That selection process can create a divergence between companies that sell “AI as a feature” and those that sell “AI as a measurable improvement.” The former can suffer when budgets tighten; the latter can hold up better.
Policy and regulation add another layer of uncertainty. AI governance is evolving, and compliance costs can be significant. Regulation can also reshape competitive advantage by determining what kinds of data practices are allowed, what transparency is required, and how liability is assigned. Markets may price in a regulatory path that turns out to be more restrictive than expected. Alternatively, markets may overestimate the speed of regulatory clarity. Either way, the timing matters.
The Q&A’s emphasis on incentives is particularly relevant here. Companies have incentives to present AI progress as inevitable. Investors have incentives to reward momentum. Analysts have incentives to maintain coverage and narrative coherence. Meanwhile, customers have incentives to avoid being early adopters of risky systems. These incentives don’t guarantee mispricing, but they do create conditions where optimism can spread faster than evidence.
So what would “AI momentum” look like if the market is right to be excited? It would look like a steady increase in the number of deployments that move from pilot to production, accompanied by evidence that performance and reliability improve over time. It would look like cost reductions that are not just theoretical but reflected in unit economics. It would look like enterprise contracts that expand rather than churn. It would look like supply chains and energy constraints being managed rather than becoming bottlenecks that cap growth.
And what would “AI hype outpacing market pricing” look like? It could look like valuations that assume rapid monetisation while the industry is still wrestling with integration and governance. It could look like revenue growth that is strong but not strong enough to justify the implied future margins. It could look like capex plans that are ambitious but not matched by demand that can absorb the capacity. It could look like a wave of product launches that doesn’t translate into sustained customer spending.
But there’s another possibility that the Q&A format invites viewers to consider: the market might be ahead of itself in some areas while still being correct in others. AI markets can be uneven. Some segments may be priced too richly, while others remain undervalued because they are less visible. Infrastructure and tooling can be less glamorous than consumer-facing models, yet they can be essential to scaling. Similarly, companies that focus on data