AI’s $518 Billion Gamble: Is the Next Financial Crash Already Here? | Tom Bilyeu Reacts

The central financial risk is that AI companies are committing to enormous spending before their revenue can reliably cover it. The host accepts that AI may prove transformative, but argues that the boom can still end in a damaging bust: borrowers either repay debt or default, while inflation can also erode its real value as a “soft default.” He rejects the idea that the boom is safe simply because major technology companies are profitable, noting that the AI company discussed has substantial operating losses. The dot-com era’s roughly 90% drawdown, he says, shows how much capital investors can lose.

Reuters reportedly disclosed $518 billion in spending commitments tied to that company. The host says its revenue grew twelvefold to nearly $4.6 billion, but considers its roughly $8 billion operating loss the more relevant measure of its current funding gap. He distinguishes that deficit from a reported $42 billion loss in 2025, which he says included capital expenditures and non-cash effects tied to potential share buybacks and mark-to-market accounting. His concern is that the commitments may be a best-case cost estimate: equipment competition could raise prices, and slower revenue or delayed projects could add hundreds of billions to borrowing needs.

The speakers illustrate the exposure with roughly $500 billion in commitments, $100 billion in equity and $50 billion in revenue, leaving about $350 billion to finance through debt. For Anthropic, one speaker cites more than $80 billion raised in equity, a valuation of about $950 billion at its last equity raise and a possible $2 trillion valuation sought in another raise. He also cites about $5 billion in revenue, $40 billion in losses and half a trillion dollars in planned spending. These figures make the outcome highly dependent on rapid revenue growth: if spending stays high while revenue disappoints, companies may struggle to make payments, and lenders could tighten credit or demand different terms.

The episode describes financing that spreads risk across companies and investors rather than leaving it with one borrower. Customers borrow to buy chips and other products from suppliers such as NVIDIA, Broadcom and Google, sometimes pledging the equipment as collateral. The speaker says suppliers may guarantee part of a shortfall or support equipment residual values, while asset managers, shadow banks and private-credit firms arrange loans through structures such as SPVs. In one Anthropic deal, he says Broadcom provides a payment backstop, Morgan Stanley arranged the financing, and Apollo and Blackstone are involved. These arrangements rely on chips retaining enough resale value, even though equipment can depreciate quickly.

The speaker compares this web of guarantees, collateral and distributed borrowing with Nortel and Lucent’s customer financing during the dot-com era, when borrowed money helped customers buy suppliers’ equipment. He also invokes 2006–2008, but says the comparison concerns structures such as SPVs, not a prediction that the financial crisis will repeat. The danger is that rapid growth assumptions can fail together: Nortel and Lucent’s projections weakened their customers and contributed to their own collapse. He cites early distress in private credit, including bad loans at Blue Owl and BlackRock, and says post-2008 regulation shifted some risky lending rather than eliminating demand for it.

A credit pullback could damage the AI boom without anyone deliberately triggering a collapse, particularly as rising ten-year interest rates add funding pressure. The speaker warns that obligations could rise from $518 billion to $618 billion next month and $1 trillion by mid-next year, but stresses that such projections depend on events going right. He also says continued AI borrowing could conflict with efforts to curb inflation and with the Treasury’s need to attract buyers for U.S. debt. The speakers disagree over whether an early bust would clear out bad investments or prolong economic pain; the host favors gradual deflation and cautions against both having no AI exposure and going all-in.

Long-term profitability remains possible if AI revenue overtakes spending, but the path is uncertain. Cheaper or open-weight models could limit demand for frontier systems and weaken companies’ pricing power; if products become interchangeable, value may shift toward infrastructure providers. The speaker says open-source competition, including China’s use of open-weight models, is already pressuring revenue strategies, while regulation could determine who is allowed to compete and where returns accrue. The host sees infrastructure as a potentially safer investment with less upside, but says investors should understand and limit their exposure rather than assume the sector cannot fail.