Artificial intelligence has moved from the margins of technology to the centre of finance, strategy and public debate with a speed that still feels hard to calibrate. One day it’s a breakthrough that will remake productivity; the next it’s a bubble, a distraction, or a threat to jobs and governance. In markets, those competing narratives don’t just coexist—they trade places, sometimes within weeks. That volatility is exactly why the Financial Times’ Lex team is hosting a live Ask an Expert Q&A: “Has the market lost its mind over AI?” On Thursday July 30 at 1pm (BST), John Foley, head of Lex, and tech comment editor Elaine Moore will take questions from readers about what is hype, what is momentum, and what is genuinely changing the outlook for businesses, markets and technology.
The question sounds rhetorical, but it isn’t. It reflects a real tension investors and executives are wrestling with: how to separate the measurable from the merely imaginable. AI is not one thing. It is a stack of capabilities—models, data pipelines, compute infrastructure, software integration, and organisational change—each with different timelines and different bottlenecks. When markets price “AI” as if it were a single event, they can end up rewarding companies for expectations that are only loosely connected to near-term cash flows. Yet dismissing AI as hype can be equally dangerous, because the technology’s impact often arrives through compounding improvements rather than a single moment of adoption.
To understand why the debate has become so heated, it helps to look at how AI has been financed and marketed. The last few years have seen a surge in investment across chips, cloud services, data centres, enterprise software and AI-native startups. Much of this spending is rational: training frontier models requires enormous compute, and inference at scale is becoming a cost and engineering challenge in its own right. But the market’s reaction has also been shaped by a familiar pattern in technology cycles. Early winners can capture attention and capital long before their business models are fully proven. Meanwhile, laggards can be punished even when they are quietly building the foundations—data governance, model evaluation, security controls, and workflow redesign—that determine whether AI delivers value beyond demos.
This is where the “lost its mind” framing becomes useful. It doesn’t necessarily mean prices are irrational in a simple sense. It means the market may be over-weighting certain signals—such as model capability benchmarks, headline partnerships, or rapid revenue growth—while under-weighting other factors that decide whether AI becomes durable economic advantage. Those factors include distribution, switching costs, regulatory compliance, and the ability to integrate AI into existing processes without creating new operational risks.
One of the most important questions for the experts on July 30 will likely be how to think about valuation in an environment where the product cycle is moving faster than the accounting cycle. Traditional metrics—earnings, margins, free cash flow—are slow to reflect changes in capability and adoption. AI, by contrast, can improve quickly, and the competitive landscape can shift rapidly as new models and tools reduce the cost of producing useful outputs. That creates a mismatch: investors may be pricing future efficiency gains and new revenue streams before they show up in financial statements, while companies may be investing heavily now without yet having the commercial proof to match the spending.
But there is another layer: AI is not only a technology; it is a platform for other technologies. The same underlying models can be embedded into customer service, coding assistants, marketing workflows, fraud detection, logistics planning, and scientific research. That breadth makes AI feel like a universal solvent. Yet universality is not the same as uniform impact. Different industries have different data quality, different regulatory constraints, and different tolerance for errors. A model that performs well in one context can fail in another—not because the model is “bad,” but because the surrounding system is not designed to manage uncertainty, handle edge cases, or ensure accountability.
That distinction matters for markets. If AI adoption is uneven, then the winners will not simply be those with the best models. They will be those who can translate model performance into reliable outcomes. That translation requires engineering discipline: monitoring, evaluation, human-in-the-loop design, and governance. It also requires organisational change: training staff, redesigning workflows, and aligning incentives so that AI is used where it improves decisions rather than where it merely produces content.
Elaine Moore’s perspective as a tech comment editor will likely bring the policy and societal dimension into focus as well. AI is being deployed in environments where legal liability, privacy, and security are not optional. The market may be moving faster than regulators, but regulators are not standing still. Rules around data usage, transparency, copyright, and risk management are increasingly shaping what companies can do and how quickly they can scale. Even when regulation does not directly stop deployment, it can raise costs and slow adoption—especially for sectors such as healthcare, finance, education and government procurement.
John Foley’s Lex lens, meanwhile, tends to emphasise the strategic and economic consequences: who captures value, how competition evolves, and what happens when expectations outrun execution. In AI, the competitive dynamics are unusual because the cost structure is changing. As models become more efficient and as tooling improves, the marginal cost of generating useful outputs can fall. But the fixed costs—compute, data acquisition, infrastructure, and talent—remain significant. That combination can create a “winner-takes-more” dynamic, where scale and access to resources matter disproportionately. At the same time, open-source ecosystems and commoditisation of components can reduce barriers for some use cases, making it harder for any single company to monopolise the entire stack.
So is the market losing its mind? The answer may depend on which part of the market you’re looking at. In some segments, exuberance is clearly visible: valuations that assume rapid adoption across many industries, or business models that rely on optimistic conversion rates from pilots to production. In other segments, the market may be underpricing the complexity of implementation. Companies that appear to be “behind” on headlines can still be building the systems that make AI safe and profitable. Conversely, companies that look “ahead” can struggle to monetise because their products are not integrated into the workflows where customers actually spend money.
A unique feature of this moment is that AI is simultaneously a consumer-facing experience and an enterprise infrastructure. Consumer adoption can be fast, but it doesn’t automatically translate into enterprise value. Enterprises care about reliability, auditability, and integration with legacy systems. They also care about procurement cycles and vendor risk. That means the path from impressive demos to recurring revenue is often longer than investors expect. Yet the market can still reward early momentum if it believes that once integration hurdles are cleared, scaling will be swift.
Another factor driving market psychology is the pace of capability improvement. When models get better quickly, it becomes tempting to extrapolate. But extrapolation is risky because improvements are not linear. There are diminishing returns, and there are new bottlenecks: data availability, compute constraints, energy costs, and the need for robust evaluation. Moreover, “better” can mean different things depending on the metric. A model that scores higher on benchmark tests may still produce unacceptable outputs in real-world settings if it lacks grounding, fails to follow instructions consistently, or cannot handle domain-specific knowledge without retrieval and fine-tuning.
This is why the “hype versus momentum” distinction is central. Momentum is not the same as hype, but it can be mistaken for it. Momentum refers to observable progress: deployments expanding, costs falling, integration improving, and measurable productivity gains emerging. Hype refers to narratives that outpace evidence. The challenge is that evidence in AI often arrives indirectly. Productivity gains might show up as reduced cycle times, fewer support tickets, faster coding iterations, or improved decision accuracy. Those outcomes can be hard to attribute cleanly, especially when multiple initiatives are running at once.
In a live Q&A format, readers will likely ask how to evaluate claims of AI-driven transformation. One practical approach is to focus on unit economics and operational metrics rather than broad promises. For example: what is the cost per successful task? What is the error rate and how is it measured? How often does the system require human correction? What is the time-to-resolution compared with baseline? How does performance vary across customer segments and edge cases? These questions are less glamorous than model benchmarks, but they are the ones that determine whether AI becomes a sustainable business.
There is also the question of labour and the future of work, which has become inseparable from AI’s market story. The fear is displacement; the hope is augmentation. The reality is likely to be both, depending on job categories and how quickly organisations redesign workflows. AI can automate parts of tasks—drafting, summarising, searching, translating, and assisting with routine analysis. But it can also create new roles: prompt engineers, AI governance specialists, model evaluators, and workflow designers. The net effect on employment may be less about total job counts and more about task composition and skill requirements. Markets may be underestimating the time needed for workforce adaptation, training, and change management—yet they may also be overestimating the resistance to adoption, because many firms are already facing competitive pressure to reduce costs and improve responsiveness.
Another angle that will resonate with investors is the geopolitical and supply-chain dimension. AI compute depends on hardware, energy, and logistics. Data centres are capital-intensive, and the availability of power can become a binding constraint. That means AI’s growth is not only a software story; it is a physical infrastructure story. When markets price AI as if compute is unlimited, they can ignore the reality that capacity expansions take time and face permitting, grid constraints and construction delays. This can create periods where demand outstrips supply, pushing up costs and affecting margins. It can also create opportunities for companies positioned to deliver infrastructure efficiently.
At the same time, the market’s attention is not evenly distributed. Some investors chase the most visible AI narratives—frontier model development, flashy consumer apps, or high-profile partnerships—while overlooking the unglamorous layers