Every few decades, markets find a story too powerful to ignore. Railways, the internet, crypto…and now AI. The overarching question investors keep circling back to today isn’t whether AI matters, but whether the market’s pricing already assumes the “AI revolution” has solved everything.

Given the valuations attached to AI-linked names, there’s no way to avoid the issue. Benchmarks are increasingly shaped by the market’s faith in this technology’s potential; even those who try to evade the theme are judged against it.

You might expect the usual rebuttal: charts showing runaway multiples, references to tulipomania and dot-com implosions, and the claim that another bubble has formed. That’s the default reaction of anyone who believes that popular, hard-to-value ideas are usually overpriced. It’s a tempting heuristic, and it might prove right, but we shouldn’t rely on reflex.

The truth is likely a bit more complex. From an investor’s perspective, AI sits uncomfortably at a crossroads between legitimate transformation and speculative storytelling. The only obvious mistake would be assuming the winners are clear and the outcome is predetermined. That’s the message the market is sending today; it will almost certainly be wrong.

At the same time, the antithetical doom-and-gloom view that this is all smoke and mirrors also feels too lazy. As Azeem Azhar and Nathan Warren note in “Is AI a Bubble?”, the evidence doesn’t yet support that claim. They write from what we’d consider the overly optimistic side of the debate, viewing the current phase as more constructive than speculative, but with a rigor that merits engagement, even if their confidence in the durability of this boom feels premature. Their analysis frames AI through five gauges of economic strain, industry strain, revenue growth, valuation heat, and funding quality; taking these elements together, they suggest a capital-intensive boom, not a speculative bubble (at least not yet).

Economic strain, for instance, remains manageable. AI infrastructure investment accounts for roughly 1% of US GDP, comparable to the telecom buildout of the late 1990s but well below the 4% peaks of past railway booms. Revenue growth, meanwhile, has been extraordinary, doubling annually across the sector, with estimates ranging from $60 billion today to as much as $1 trillion by 2028. And unlike the dot-com era, funding quality remains high: hyper-scalers such as Microsoft, Amazon, and Alphabet are still able to finance much of their AI buildout from operating cash flows rather than debt, though that balance may shift as capital demands accelerate.

So, what’s the catch? Valuation and structure.

A growing number of reports suggest that tech giants are engaged in a capital arms race to corner infrastructure capacity and box out competition, a strategy that may prove short-sighted. We’ve already seen that state-of-the-art results can be achieved with far less compute and energy, as DeepSeek has recently demonstrated, calling into question the long-term viability of this capital-heavy moat. Yet the optimistic camp tends to overlook key structural risks. Rapid technological leapfrogging could render today’s trillion-dollar GPU and data-center investments obsolete within a few years, especially as models become more efficient and require less compute for inference. Meanwhile, mounting regulatory pressure on data use, digital likeness, and content authenticity could slow monetization just as capital intensity peaks. Add to that the creeping, insidious problem of “AI slop” (low-quality, AI-generated content feeding itself, resulting in a flywheel of ever-sloppier slop) and the ecosystem starts to look more fragile than self-correcting.

And just because today’s prices haven’t yet reached the absurdity of past cycles doesn’t make them reasonable. The market may not be manic, but it is crowded, and crowded trades can break without warning. What’s more, the circular nature of current capex, with Big Tech firms effectively buying each other’s compute, cloud, and GPUs, creates a closed system that feeds on its own momentum. Think ouroboros, the snake eating its own tail. This mutual reinforcement keeps the AI narrative alive but also blurs the line between real, broad-based demand and self-sustaining hype.

Still, nuance matters. It’s easy to be cynical; pessimism can be more comfortable and intellectually tidy. But optimism deserves attention too, provided it’s grounded in evidence rather than faith. Too many bullish takes lean on overconfident forecasts and “just trust me” logic. However, the better question isn’t whether AI is over- or under-valued, it’s where we sit on the continuum between durable transformation and unsustainable exuberance.

Yet even the most balanced optimism may underestimate how vulnerable this exuberance is to geopolitical realities. AI’s power extends well beyond productivity; it can sway minds and influence civic behavior, putting it directly in the sights of governments worldwide. Nations that once tolerated American dominance in smartphones, advertising, and social media are unlikely to remain passive this time. The risk of AI monopolization could trigger nationalistic pushback and regulatory firewalls. Meanwhile, the global trend toward influence gating, such as age restrictions on social platforms, suggests growing discomfort with technologies that shape thought and behavior. If AI is seen as a threat to sovereignty or employment, resistance could escalate quickly. And if labor displacement continues unchecked, calls to regulate the pace of development will only grow louder. Ironically, the same tech elite inspired by Dune may soon find themselves facing a Butlerian-style revolt.

For now, the runway for AI enthusiasm remains long. The narrative continues to attract capital and the public’s imagination, with markets still rewarding belief. That has clear implications for positioning and strategy: we can’t afford to ignore the story, but we also can’t afford to believe it blindly.

Discipline, not dogma, should guide exposure. AI may not yet be a bubble, but it’s undeniably a belief system. And belief, in markets, always has a half-life. The ultimate winners will be those who stay invested with intent, anchored to real adoption and demonstrable use cases, tangible cash flows, and a clear view of which investments can still justify their cost when the hype cycle resets—especially across the layers of infrastructure where value ultimately settles: not with those building capacity, but with those turning it into utility—and attention inevitably shifts elsewhere.