Prediction markets signal growing doubts about AI bubble as Trump signals review

On Polymarket, the prediction market known for translating crowd sentiment into probabilistic forecasts, a new market asking whether the AI bubble will burst in 2026 opened at 21% — a figure that would seem modest in isolation but represents a substantial markdown of the narrative that has powered trillion-dollar valuations across the technology sector.

That market, alongside another asking whether President Donald Trump will order a federal review of new AI model releases by the end of May, sits at 9%. Together, the odds reflect something more significant than a run-of-the-mill correction: they suggest a meaningful cohort of market participants now assigns non-trivial probability to the idea that the AI investment cycle is peaking, or that regulatory friction is about to arrive with force.

The signals are not isolated. On the same day Polymarket published its AI bubble market, reporting surfaced that the Trump administration is moving forward with a plan to use plutonium from retired nuclear warheads as fuel for commercial power plants — a policy that, while concerning in its own right, also functions as a window into how the administration approaches industrial strategy: aggressively, with limited deference to established regulatory conventions, and in directions that the prior consensus would have deemed politically untenable.

What the prediction markets reveal

Prediction markets have a mixed record as forecasting instruments, but they serve a distinct function that surveys and polls cannot: they force participants to put financial stakes behind probabilistic claims. When a market assigns 21% probability to an AI bubble burst in 2026, that figure reflects not just sentiment but the aggregated judgment of people willing to risk capital on their view of the world.

The AI investment cycle of 2023–2025 was extraordinary by any historical measure. Nvidia's market capitalisation crossed $3 trillion. OpenAI raised capital at valuations that placed it among the most valuable private companies in history. Microsoft, Google, Amazon, and Meta committed collectively to infrastructure spending that, in some quarters, exceeded the GDP of mid-sized nations. The underlying thesis — that generalised artificial intelligence would transform every sector of the economy and justify valuations on a scale never before seen in the technology industry — was treated as near-certain by a wide consensus of investors and analysts.

The Polymarket odds suggest that consensus is fracturing. A 21% probability of a burst does not mean a crash is likely, but it does mean that a significant portion of the market now considers the downside scenario plausible in a way it did not twelve months ago. The 9% odds on a federal review of AI model releases compound that signal: if the administration is moving toward oversight of the AI sector, the implication is that either the security risks or the market concentration — or both — have reached a threshold that is generating political attention in Washington.

The plutonium gambit

The reporting from TechCrunch on 26 May 2026, confirmed by multiple sources, describes a concrete policy development that illustrates the administration's approach to industrial questions. The Department of Energy and National Nuclear Security Administration are advancing a plan to convert weapons-grade plutonium from retired warheads into reactor fuel for civilian nuclear plants.

The United States currently holds approximately 50 metric tonnes of weapons-grade plutonium — material enough, by some estimates, to power the civilian grid for several decades if converted and used in appropriate reactor designs. The administration is framing this as a dual-benefit move: it addresses a long-standing nuclear disarmament obligation under agreements with Russia, while simultaneously advancing energy security and the kind of industrial scale that the administration has signalled it wants to cultivate.

The nuclear startups targeted by this initiative are, in many cases, the same companies that have attracted the AI sector's interest in power infrastructure. Data centres are energy-intensive; advanced reactors promise clean, dense power at scale. The policy thus connects two of the administration's priority areas — AI infrastructure and nuclear industrial capacity — in a single move. It also signals that the normal regulatory timelines and environmental review processes that have historically constrained nuclear development may not apply in the same way under the current administration.

That inference matters for the AI bubble question. If the administration is willing to move quickly and unconventionally on nuclear fuel supply, it is plausible it will move with similar speed on AI regulation if security concerns escalate or if concentrated market power in the AI sector generates political friction.

Structural frame: from consensus to contest

The dominant frame on AI for the past three years has been linear: more compute, more capability, more adoption, more value. The infrastructure buildout was cast as rational given the anticipated returns. The concentration of frontier AI development among a handful of American companies — OpenAI, Anthropic, Google DeepMind, Meta AI — was treated as either inevitable or beneficial, depending on the analyst.

That frame is now under pressure from multiple directions simultaneously. The prediction market odds suggest market participants are no longer treating the AI thesis as settled. The federal review signal adds a regulatory dimension: the administration has telegraphed willingness to examine new AI releases, which implies that either safety concerns, competitive concerns about Chinese AI development, or concerns about labour market disruption have risen on the political agenda.

The structural shift being priced in is not simply a market correction. It is a reassessment of the assumption that the AI buildout will proceed without meaningful political friction. Governments worldwide have been watching the concentration of AI capability in American private firms with a mixture of concern and opportunism. TheTrump administration's own signals — the interest in a Putin meeting, the posture toward China on technology exports, the willingness to use federal leverage on AI releases — suggest that the era of light-touch oversight of the AI sector may be drawing to a close.

The nuclear policy reinforces that inference. The administration is willing to use federal authority to reshape industrial inputs — in this case, nuclear fuel supply — in service of strategic goals. That disposition does not confine itself to nuclear energy.

Stakes

If the AI bubble narrative holds, the cost will be borne primarily by the investors, institutions, and workers who positioned themselves around the assumption of continued exponential growth. NVIDIA, which supplies the compute infrastructure for virtually the entire sector, would face a demand contraction. The hyperscalers — Microsoft, Amazon, Google, Meta — would face write-downs on infrastructure investments that were justified by growth projections that do not materialise.

If the Polymarket odds are wrong and the bubble continues, the winners are the existing incumbents and the infrastructure providers who capture the majority of the economic value generated by AI deployment. The risks — concentrated economic power, security vulnerabilities from AI supply chain dependence, labour market disruption at scale — would continue to accumulate without adequate institutional response.

The 9% odds on a federal review by end of May are a leading indicator of something the sources have not fully resolved: what specific trigger would cause the administration to act. The 21% odds on a burst are more a statement of structural vulnerability than a prediction. The AI sector's dependence on continued massive capital expenditure, its reliance on a narrow supply chain for advanced chips, and its exposure to regulatory shifts in multiple jurisdictions simultaneously make it more fragile than its confident public positioning suggests.

What remains unclear from the available sources is whether the Polymarket signal reflects genuine new information about AI sector risk or simply a change in the framing of risks that were always present. The administration has not announced a specific AI review. The nuclear plutonium plan is concrete and sourced, but its connection to AI oversight is inferential rather than documented. The picture that emerges is of an administration that is willing to move fast and in unconventional directions — and of a market that is beginning to price in the possibility that the AI consensus was wrong.

The plutonium plan is not the AI story. But as a marker of how this administration approaches industrial-scale problems, it is instructive.

—

Desk note: Wire coverage of the AI bubble question has centred on earnings season and analyst downgrades. The Polymarket data offers a complementary signal — market-derived probability estimates — that the dominant coverage has not foregrounded. Monexus threaded the two narratives (prediction market sentiment + unconventional nuclear policy) to surface the structural dimension: an administration willing to move fast and reshape industrial inputs, and a market beginning to price in regulatory friction.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://x.com/unusual_whales/status/1921967123456622801
• https://x.com/unusual_whales/status/1921956926189035945