As large artificial intelligence (AI) companies prepare to go public, the hedge fund manager who predicted the 2008 subprime mortgage crisis is intensifying his warnings that capital spending by five major tech companies repeats a risky trend he’s seen before, soon to reach an unprecedented scale.

Michael Burry, whose story was chronicled in the feature film The Big Short, has been cautioning since late 2025 that AI companies are careening toward a potential market-bubble collapse. The question now appears to be not “if,” but “when” this event will occur, he warns in his Substack newsletter “Cassandra Unchained”—named after the Greek priestess who was cursed with the ability to issue truthful prophecies that no one believes.

In his latest analysis, Burry scrutinizes the purchases, leases, third-party debt, the legal entities created to carry that debt, and construction projects undertaken by the five public hyperscalers—Microsoft, Amazon, Alphabet, Meta, and Oracle—calling the aggregate $3 trillion  “scary.”

Burry adds that the companies have concealed these commitments from their balance sheets under controversial accounting regulations.

“A company can sign a non­cancellable 20-year data centre lease, disclose it in a footnote, and carry nothing on the balance sheet until the keys turn and the data centre boots up,” he said.

Burry added that there has not been a comparable level of investment—equal to 2.07% of the United States’ gross domestic product—since the dot-com bubble of the late 1990s.

“I have little doubt the next few quarters will set still higher and higher net investment/GDP marks, possibly even eclipsing that aftermath of the 2000 tech stock peak,” he writes.

“Let’s also remember this past year, the combined earnings of the Big 5 are less than $400 billion,” says Burry. He adds that the group of five has made more than $1.5 trillion in promises to pay for items in the supply chain—purchase commitments that don’t appear on company financials, but are reported in regulatory filings, potentially bringing the total spend to $5 trillion by 2028.

Another of the hyperscalers Burry examined was Oracle, which recently sought to shield itself from contractual obligations for a massive data centre project in New Mexico by invoking the force majeure clause in its contract.

There is also word of growing apprehension on Wall Street as a number of AI infrastructure companies have either halted or postponed their planned initial public offerings (IPOs), The New York Times reported.

Holtec Nuclear, a company that planned to construct small modular nuclear reactors to power AI, initially said it was pausing its IPO plans to raise up to $900 million, but subsequently filed with the Securities Exchange Commission to withdraw completely.

However, immediately after the Times reported last week that another company, SB Energy, a subsidiary of the Japanese conglomerate SoftBank, had delayed its multi-billion-dollar IPO, the company announced it would proceed as planned, reports Investing.com.

It has proposed the largest data centre project in the world in Ohio, leasing compute power to OpenAI and backed by Nvidia, which pledged to invest up to $105 billion to backstop OpenAI’s leases.

Investors had questioned SB Energy’s valuation of at least $50 billion considering it has yet to operationalize a data centre. The company said it anticipated a revenue backlog of $439 billion beginning in 2028 and distributed over 20 years.

Another company that experienced a substantial increase in revenue over the past year due to the acquisition of data centre clients, has slowed down its IPO process. The Times attributed this to “challenges facing data centres,” rising interest rates, and broader economic uncertainty.

Public support for data centre developments has collapsed over the last year. A Heatmap poll by Embold Research found that three-quarters of Americans expressed opposition to local data centre projects, with 60% strongly opposing them.

Other companies anticipating public offerings this fall include Vantage Data Centers, CyrusOne, and Switch, all of which have more experience compared to the newcomers, operating more than 80 data centres among them. Nscale disclosed its financials earlier this month in preparation for a public offering in October, with Anthropic and Microsoft among its clients.

“The scope of what they’re embarking on is just enormous and so they need to tap all the major capital markets,” Tomasz Tunguz, a venture-capital investor at Theory Ventures, told the Times.

Nvidia, the company responsible for the chips used in a significant portion of AI developments, is encountering challenges on multiple fronts. Its shares are trading at their lowest value in over a decade, according to data compiled by Bloomberg.

“The stock has de-rated pretty significantly, which suggests a healthy dose of skepticism that the company’s current earnings power is sustainable,” said Eli Horton, senior portfolio manager for thematic equities and durable growth equities at TCW.

David Russell, global head of market strategy at TradeStation, said some of Nvidia’s biggest customers are developing their own AI chips in-house. Both Meta Platforms and Alphabet (Google) are taking steps to reduce their reliance on Nvidia.

“It is very conceivable its market position will weaken over time, and that means gross margins are more likely to go south than improve, which is a big problem for investors,” Russell suggested.

Nvidia’s environmental impacts are also under scrutiny, with accusations of greenwashing arising from significant increases in emissions, reports EuroNews. The lead author of a new report challenges the company’s claims that AI has a “net climate benefit.”

According to the report published by eight climate, corpo5ate accountability, and digital campaigning groups, emissions from the use of Nvidia’s chips are comparable to those from burning coal. They estimate that operating Nvidia chips sold since 2022 could have generated the equivalent of between four and 21 million tonnes of carbon dioxide emissions in 2025, depending on the electricity used to power them.

The report also states that supply chain emissions have risen by 725% since 2020, going from 1.3 million tonnes of carbon dioxide equivalent in 2020 to 10.7 million tonnes in 2026.

Nvidia’s financial statements have also come under scrutiny from Burry and others who discovered significant write-offs tied to current data centre infrastructure. The company is accused of overstating the lifespan of its graphics processing units (GPUs), the electronic circuits essential to AI computing, in its financial statements, and spreading the depreciation amount over a longer time frame, making earnings appear higher.

Burry said on social media in November that Nvidia’s GPUs have a useful lifespan of two to three years, implying that the company will understate depreciation by $176 billion from 2026 to 2028. In response to the allegations, Nvidia issued a memo stating it could demonstrate that its GPUs last four to six years, reported CNBC News.

Despite the underlying concerns, markets remain hyper-focused on Anthropic’s IPO, an opportunity that shifted from September to mid-October and is now planned for November, following the U.S. mid-term elections, Forbes reports.

Seeking approximately $2 trillion, it would be the largest IPO ever and could set a benchmark for the valuation of Wall Street’s most prominent AI companies, including rival OpenAI which is also slated for an IPO in 2027.

Reuters gained access to Anthropic’s IPO prospectus, which provides public information before investment but also serves as a marketing and sales tool. The document asserts that AI will transform the global economy more profoundly than industrialization, electricity, and the Internet.

Anthropic incurred a loss of $42 billion in 2025 and plans to allocate $518 billion to cloud computing and infrastructure obligations in the future. Reuters wrote that Anthropic’s sale will further determine whether the enthusiasm for the AI industry can withstand increased scrutiny.

Simultaneously, data centres are imposing significant strain on insurers, prompting companies to explore alternate strategies as premiums tied to hyperscale AI projects are projected to rise from $11 billion to $24 billion by 2030, reports The Economist.

Jimmy Keime, a representative of reinsurance giant Swiss Re, said many large AI projects will be insured for less than half their value. “Surety bonds” will play a role in the event that a project’s developer fails to fulfill its obligations.

Some AI firms have opted to create their own “mini insurance companies”, and the premiums they pay to in-house insurers can often be deducted from their taxes, a practice that increased by 8% in 2025, reaching almost $240 billion.