Why the AI Crash Will Start in Private Credit, Not Big Tech Stocks
Behind closed doors at the Treasury Department and the Financial Stability Oversight Council, the government’s risk apparatus is quietly shifting its gaze. A leaked internal draft, paralleled by legislative demands from Senators Elizabeth Warren and Richard Blumenthal, reveals that Washington is no longer treating artificial intelligence as a simple equity-market valuation debate. Instead, policymakers are directing the Office of Financial Research to map exposure across a sprawling, non-bank lending apparatus: chipmakers, data centers, hyperscalers, cloud providers, and model developers.
Across the Atlantic, the Bank of England has reached the same conclusion. British regulators warn that AI infrastructure will devour trillions of dollars in debt-financed investment over five years, and that the sector has become so outsized it drove U.S. GDP growth in the first half of 2025. A sharp asset drawdown, the Bank notes, would now impair macroeconomic stability through investment contractions and wealth effects.
The regulatory risk perimeter has formally expanded from public stock tickers to private debt plumbing.
The Three-Balance-Sheet Collision
This supervisory anxiety stems from the convergence of three balance sheets: Big Tech’s capital expenditure accounts, private credit’s maturity-transformation funds, and the state’s industrial policy.
The hidden architecture of this buildout is a direct descendant of post-2008 regulatory arbitrage. When capital mandates forced banks out of speculative lending, credit creation migrated into opaque asset-backed structures, infrastructure vehicles, and bespoke debt markets. Today, AI data center developers are exploiting those exact shadow channels—borrowing against long-duration, physically fixed assets to finance demand curves that remain largely theoretical.
Markets initially priced these ventures with software economics: infinite scalability, high margins, and asset-light balance sheets. The reality is closer to late-1990s telecom. The industry is locked in a brutal capital-expenditure arms race defined by massive upfront infrastructure costs, rapid hardware depreciation, capacity hoarding, and uncertain end-user willingness to pay. This cycle is not a dot-com replay; it is dot-com equity froth stacked on top of telecom infrastructure overbuild, wrapped in pre-2008 structured credit opacity.
The State’s Moral Hazard
Why hasn't Washington pulled the emergency brake? Because the Treasury is trapped in a policy contradiction. Through its AI Innovation Series, the department champions rapid AI adoption as essential for competitiveness, with Secretary Scott Bessent framing the failure to adopt productivity-enhancing technology as a standalone risk.
This creates an unstable equilibrium. The state desperately needs the productivity boom and equity wealth to sustain national-security and economic narratives, yet its supervisory arms fret over a financial-stability shock. This is not hypocrisy; it is the hallmark of strategic bubbles. By the time regulators obtain clean exposure data, industry incentives will have already solidified around disguising leverage, smoothing valuations, and lobbying against hard constraints.
Where the Plumbing Breaks
The critical realization for institutional investors is that when this cycle resets, the detonation will not occur in Microsoft or Nvidia. Mega-cap equities are merely the billboard; private credit is the plumbing, data centers are the collateral, power contracts are the choke point, and enterprise cash flow is the ultimate solvency test.
The entire financial architecture rests on a single vulnerability: compute demand must remain supply-constrained at premium prices long enough to amortize the debt. But private credit’s greatest selling point—negotiated, non-mark-to-market valuations—is its systemic Achilles’ heel. Losses are delayed, net asset values are smoothed, and covenants are amended in the shadows.
If inference efficiency improves, open-source pricing collapses, or enterprise adoption stalls, the capital stack breaks long before the broad productivity thesis is proven. The trigger will be a marginal data-center developer or GPU leasing platform failing to refinance, forcing simultaneous write-downs across private portfolios and triggering LP redemptions. Only then will hyperscalers trim capital expenditures, validating the equity sell-off.
Crucially, a credit reset will not democratize AI; it will entrench an oligopoly. The dominant balance sheets will simply sweep up distressed compute, talent, and scarce power rights at penny-on-the-dollar valuations. The institutional playbook is clear: avoid duration-mismatched infrastructure debt, and go long on unavoidable bottlenecks—power grids, secured cloud distribution, and profitable workflow software.
not investment advice