An internal report shows that the US Department of the Treasury has begun to assess the risks of the AI bubble.

According to an exclusive report by NOTUS on July 6, within the U.S. Department of the Treasury, a report on artificial intelligence market risks has been completed for weeks and is awaiting official approval. The document, drafted by career analysts, can be summed up in one core conclusion: The AI market is repeating the 2000 dot-com bubble, and if it bursts, the impact will ripple through the entire economic system.

(Note: NOTUS is a news outlet focused on U.S. politics and policy reporting, founded in 2024 by a group of veteran journalists, known for providing in-depth, exclusive coverage from inside Washington.)

Yet while the report awaited signature, Treasury Secretary Scott Bessent stood on a stage in New York praising tech giants for pouring $750 billion into AI buildout this year. He told other leaders at the recent G7 meeting that the biggest risk from AI is not safety, not jobs—it is China outpacing the United States.

This stark disconnect in the policy logic around AI is deeply revealing.

01

What the Report Says

According to the internal document obtained by NOTUS, career Treasury analysts mapped out a clear causal chain: AI firms are embedded far more deeply into the U.S. economic structure than dot-com companies ever were. They rely on more complex financing instruments and far larger infrastructure investments. If financial conditions worsen, productivity targets are missed, or bottlenecks in supply chains or power grids choke growth, the entire system will be hit.

The report’s assessment is highly specific: an AI downturn would trigger a "shockwave across the entire economic ecosystem." The stock market, private credit markets, firms financing data center construction, cloud service providers, chip manufacturers, and utility companies—all lie in the transmission path.

The analysts specifically note that the AI industry is vastly more concentrated than the dot-com era, with a handful of firms controlling the vast majority of market share. These giants are also highly interconnected and deeply cross-cutting across different markets. If investment dries up or demand slows, the ripple effect will not stop at the tech sector.

The most alarming assessment lies deep in the report: unlike the dot-com bubble, which was driven largely by retail investors, this current AI boom sees heavy participation from institutional investors. A retail bubble bursting harms stock markets and household wealth; an institutional bubble bursting threatens the stability of the entire financial system. Large banks, hedge funds, and private credit firms are the primary backers of this high-stakes AI bet. Risk has not been diversified—it has been concentrated, into a handful of too-big-to-fail institutions.

Of course, the report does not assert that a bubble is guaranteed to burst. According to NOTUS, analysts acknowledge that AI companies are fundamentally different from dot-com-era firms: they are more mature, some are already profitable, and their balance sheets are far healthier.

The report even offers an optimistic out: if productivity growth expectations are met and commercialization paths are proven viable, the so-called bubble might never burst. But the flip side of that statement is this: the stability of the entire financial system now rests on the premise that AI must deliver on its promises.

02

What Was Said on the Stage

When Bessent spoke in New York, he presented a different narrative. Citing the productivity miracle of the internet era, he asked the audience: Can we at least achieve that? Can we go even further?

He did not discuss bubbles, risk exposure, or market concentration. He only talked about speed.

This tone cannot simply be dismissed as the two-faced nature of politicians. The U.S. government’s AI strategy rests on a single pillar: AI is a race that cannot be lost. The cost of losing to China outweighs the cost of any bubble bursting.

In the shadow of that pillar, all discussions about risk, leverage, and concentration have been relegated to secondary issues. Just as the Manhattan Project would not pause to debate whether the atomic bomb could harm its creators, or the space program would not stop to calculate the return on investment for a moon landing. Under the logic of the race, risk is reduced to a temporary pain to be endured before the mission is accomplished.

A Treasury spokesperson issued a near-contradictory statement on the matter. On one hand, he told NOTUS that the report is "unvetted" and "does not represent the Treasury’s policy or views." On the other hand, he reaffirmed the department’s official position: "Artificial intelligence will be a key driver of America’s new golden age." The subtext is clear: we acknowledge that analysis exists in a drawer, but we choose not to look at it—because a higher priority overrides it.

If we seek a historical parallel for this choice, the closest one is not the dot-com bubble, but the U.S. housing market on the eve of the 2008 financial crisis. Around 2005, analysts inside the Treasury and the Federal Reserve had already warned about the systemic risks of subprime mortgages. Those reports were equally clear, their transmission chains equally complete—and they too were locked away in drawers.

The prevailing narrative back then was identical: home prices always rise, financial innovation has diversified risk, this time it’s different. The difference is that before the subprime crisis, at least some officials were openly discussing risks and pushing for regulation. Today, even those voices have almost entirely vanished.

03

Lifeline or Strangulation Rope?

The outside world is not unaware. Both the Bank of England and the managing director of the International Monetary Fund have publicly expressed concerns over AI overvaluation and its systemic risks to the economy.

Within the U.S., Capitol Hill is also taking action. According to NOTUS, Senator Elizabeth Warren, senior member of the Senate Banking Committee, introduced a bill last month requiring financial firms to disclose non-public data related to AI buildout to the Treasury, and mandating the Treasury to submit a special report on the financial system’s full exposure to AI-related risks. Warren’s words cut to the core: "AI and big tech companies are increasingly relying on opaque forms of debt and balance sheet magic to underpin trillions of dollars in AI construction."

This bill itself is an indictment of the status quo. The market does not know the true leverage ratios of AI companies, how much cross-collateralization is hidden in data center financing structures, or how many private credit contracts would trigger default clauses the moment computing power rental prices decline. The Treasury—the only body with the power to force disclosure of that information—has locked the report in a drawer, while stepping onto stages to declare that America must stay ahead.

NVIDIA’s GPUs, cloud providers’ data centers, startup valuations, and big tech capital expenditures—all these asset prices are tied together by a single rope: expectations.

If AI companies deliver on their promise of productivity growth, that rope becomes a lifeline, pulling all assets upward. If they fail to deliver, it becomes a noose: the fall of one asset will, through that same rope, drag down every other asset connected to it.

This article is compiled from public sources and is for information exchange only, and does not constitute any investment advice.

This article is from the WeChat public account "Silicon Starlight", authored by Xing Yao, and published with authorization by 36Kr.