The three warnings point to the same vulnerability: AI infrastructure is increasingly being financed with debt, so disappointing returns or tighter credit could turn a setback in tech shares into a broader market shock. They differ on timing: Burry sees signs of stress already, Dalio says a debt-fuelled bubble may be nearing a breaking point, while Temasek sees an AI-trade unwind as the biggest risk but not an imminent one. These are risk scenarios, not established forecasts of a 2027 crash. 15711

• Burry: He called the market’s mood “denial,” comparing it with periods around the 2000 and 2008 crashes and saying that phase lasted six to nine months. He pointed to stress in private credit and data-center financing, including the risk that AI-linked obligations held through private-equity-owned insurers may be difficult to value or absorb if projects falter. His analogy is a warning, not proof that the same crash will recur. 111316
• Dalio: He described debt-funded AI investment as a “classic bubble” vulnerable to higher rates and demands to turn investments into cash. That concern fits the possibility that even large hyperscalers cannot keep expanding borrowing indefinitely; separately, he said a U.S. debt crisis is possible within three years. 12
• Temasek: CIO Rohit Sipahimalani said an unwind of the AI narrative was the market’s biggest risk, particularly if regulation, safety concerns or inadequate returns for end users undermine it during 2027. His position is caution about a possible unwind, not a call for one immediately. 7

The financing figures explain why those warnings matter. AFP reported that tech-sector borrowing rose from next to nothing in 2024 to about $500 billion in the first nine months of 2026, as Google, Meta, Amazon, Microsoft and others financed chips, servers and data centers; it also reported a Goldman Sachs figure of $1.2 trillion for 2027. But available reports describe that $1.2 trillion figure differently—as borrowing in one account and AI-infrastructure investment in another—so it should not be treated confidently as a forecast of new debt issuance without the underlying Goldman report. 511

Higher borrowing costs, competition with Treasury bonds for investors, and a delayed project would each test whether expected AI revenue can support the buildout. The available evidence supports Treasury-demand concerns and reports a financing or permitting snag involving an Oracle project in New Mexico, but does not adequately establish the question’s precise Meta and data-center rates, Oracle’s $125 billion debt figure, or the project’s ultimate status. 916 Insufficient evidence here also to confirm the stated Bank of England warning, Panmure Liberum’s S&P 500 target of 5,000, or Temasek’s proposed increase from 6% to 15%; those should not be presented as verified facts. The practical test in coming earnings reports is whether companies can show returns and cash generation commensurate with their escalating AI spending—a test, not a predicted collapse. 57

The three warnings point to the same vulnerability: AI infrastructure is increasingly being financed with debt, so disappointing returns or tighter credit could turn a setback in tech shares into a broader market shock. They differ on timing: Burry sees signs of stress already, Dalio says a debt-fuelled bubble may be nearing a breaking point, while Temasek sees an AI-trade unwind as the biggest risk but not an imminent one. These are risk scenarios, not established forecasts of a 2027 crash. 15711

• Burry: He called the market’s mood “denial,” comparing it with periods around the 2000 and 2008 crashes and saying that phase lasted six to nine months. He pointed to stress in private credit and data-center financing, including the risk that AI-linked obligations held through private-equity-owned insurers may be difficult to value or absorb if projects falter. His analogy is a warning, not proof that the same crash will recur. 111316
• Dalio: He described debt-funded AI investment as a “classic bubble” vulnerable to higher rates and demands to turn investments into cash. That concern fits the possibility that even large hyperscalers cannot keep expanding borrowing indefinitely; separately, he said a U.S. debt crisis is possible within three years. 12
• Temasek: CIO Rohit Sipahimalani said an unwind of the AI narrative was the market’s biggest risk, particularly if regulation, safety concerns or inadequate returns for end users undermine it during 2027. His position is caution about a possible unwind, not a call for one immediately. 7

The financing figures explain why those warnings matter. AFP reported that tech-sector borrowing rose from next to nothing in 2024 to about $500 billion in the first nine months of 2026, as Google, Meta, Amazon, Microsoft and others financed chips, servers and data centers; it also reported a Goldman Sachs figure of $1.2 trillion for 2027. But available reports describe that $1.2 trillion figure differently—as borrowing in one account and AI-infrastructure investment in another—so it should not be treated confidently as a forecast of new debt issuance without the underlying Goldman report. 511

Higher borrowing costs, competition with Treasury bonds for investors, and a delayed project would each test whether expected AI revenue can support the buildout. The available evidence supports Treasury-demand concerns and reports a financing or permitting snag involving an Oracle project in New Mexico, but does not adequately establish the question’s precise Meta and data-center rates, Oracle’s $125 billion debt figure, or the project’s ultimate status. 916 Insufficient evidence here also to confirm the stated Bank of England warning, Panmure Liberum’s S&P 500 target of 5,000, or Temasek’s proposed increase from 6% to 15%; those should not be presented as verified facts. The practical test in coming earnings reports is whether companies can show returns and cash generation commensurate with their escalating AI spending—a test, not a predicted collapse. 57