BlackRock’s Larry Fink predicts AI bankruptcies: ‘That’s capitalism’

Hyperscaler capex is forecast at $650B over the next 12 months, ~70% above the $380B invested in 2025, with some estimates projecting trillions over 3–5 years. BlackRock CEO Larry Fink says ‘one or two’ AI-related bankruptcies are inevitable but encourages continued heavy investment to keep the U.S. competitive with China. Evercore flags risk of companies going cash-flow negative and rising corporate debt levels; Amazon, Alphabet, Meta, Microsoft and Oracle issued $121B of bonds in 2025 versus a $28B five-year average, and Oracle issued $26B in 2025 with plans for $45–50B this year. These dynamics are sector-moving—supporting aggressive capex and competition while raising credit and valuation risk for Big Tech.

Analysis

The AI capex wave will bifurcate winners between differentiated software/contract plays and commoditized infrastructure providers. Firms that can convert raw compute into sticky, high-margin enterprise contracts (software + integration) will see persistent ROIC advantages even if raw compute pricing falls 20-40% over a multi-year buildout; conversely, pure-play hardware and open-cloud resellers face margin pressure and cyclical capacity write-down risk. This will compress voluntary resale markets and elevate the value of long-term service agreements, pushing more economic rent to software/IP owners than raw hyperscale builders.

A key near-term vulnerability is credit and liquidity sensitivity: a 75–125bp move higher in global risk-free rates within 6–12 months materially raises hurdle rates on multi-year AI infrastructure bets and can flip firms from EBITDA-positive to cash flow negative during the build phase. Regulatory shocks (export controls or antitrust actions) or a plateau in AI monetization (revenue per chip saturates) are credible catalysts to rerate capex-heavy names within quarters, while demand normalization and model-efficiency gains are 1–3 year mitigants. Monitoring corporate bond spreads and capex-to-free-cash-flow trajectories monthly will be the fastest early-warning signal.

The market consensus underestimates the duration of overcapacity in raw compute and overestimates uniform upside across hyperscalers: concentration risk means a handful will capture most long-term economic surplus, leaving the rest refinancing-dependent. That implies asymmetric trade opportunities: long differentiated enterprise cloud/software providers and select financial-advisory/underwriting beneficiaries, short levered hyperscalers or buy-tail protection on their credit profiles. Position sizing should explicitly account for idiosyncratic policy and yield shocks that can compress or invert expected returns in 3–12 months.

Key Decisions for Investors

• Long ORCL (12–18 months): buy ORCL equity or a 12–18 month call spread to capture durable enterprise cloud contract leverage. Target +30% upside if enterprise renewals and margin expansion continue; downside capped ~15% if growth disappoints. Hedge with 30–50% notional protection via puts if funding markets widen.
• Pair trade — Short AMZN / Long ORCL (6–12 months): short AMZN equity or buy a 6–12 month put spread financed by short-dated calls, paired 1:1 with a long ORCL position. Expect ORCL to outperform by 15–25% as software capture outpaces raw infra spending; risk if Amazon proves faster at monetization (loss scenario ~20%).
• Credit/IB exposure — Long EVR (6–12 months): accumulate EVR to capture higher advisory and underwriting flow from elevated bond issuance; target +25% upside with limited downside if underwriting volumes normalize. Use position as a levered play on sustained capital markets activity rather than direct AI monetization.
• Tactical hedge — Buy 1–2 year protection on large hyperscaler credit (or buy put spreads on MSFT/AMZN 9–12 month expiries): cost should be treated as insurance against a >75bp rate shock or a near-term monetization miss. This protects equity exposure across the portfolio and offers skewed payoff if a refinancing/credit event occurs within 12 months.