The AI Bank Run Has Started

Overview

The video argues that panic in AI-related financing has shifted from public markets into the private credit system—creating an “AI bank run” dynamic that could destabilize funding for data center buildouts.

Oracle / Blue Owl: debt stress tied to “electricity not ready” risk

• The presenter claims Oracle (positioned as a major AI data center operator) is running out of money and struggling to meet debt obligations.
• Oracle reportedly issued a force majeure notice to Blue Owl Capital (a private credit firm) tied to a data center under construction in southern New Mexico near Las Cruces.
• Core allegation: the data center is unlikely to have power connected in time because local opposition delayed permitting—potentially pushing the connection out by years.
• The presenter emphasizes the scale of the power requirement (2.45 GW, compared to powering millions of homes) and argues Oracle either:
  • didn’t adequately mitigate the power/connectivity risk, or
  • is attempting to avoid/limit payments.
• Despite the electricity delay, the presenter argues Oracle may still owe ongoing lease payments (reduced but still substantial) to cover construction costs and lender interests.
• The presenter also notes that lenders (including BNP Paribas and Goldman Sachs) expect interest payments, while Oracle’s credit quality is described as near “junk” territory—raising downgrade risk if projects fail to deliver revenue.

Private credit “bank run” dynamics: withdrawals rising, liquidity constrained

• The presenter claims investor flight is accelerating in private credit funds that finance AI/data centers.
• Reported cases include:
  • Ares Strategic Income Fund: withdrawal requests reportedly rose to 13.1% of shares in Q3 (from 14.4% in the prior quarter), which the presenter says is not an improving trend. Any temporary improvement may be due to limited payouts that reduce headline withdrawals.
• The presenter asserts typical quarterly withdrawals are usually ~1–3%, but AI-driven stress has pushed requests to ~13–15%.
• Mechanism described: funds impose withdrawal limits because they lack sufficient cash—similar to bank runs where there isn’t enough liquidity “in the till.”
• The presenter cites similar pressure at Blackstone’s private credit fund, saying it shrank for multiple quarters and imposed withdrawal limits due to high redemption demand.
• The video contrasts this with rising Treasury yields (>5%), arguing that higher safe returns make it easier for investors to abandon illiquid private credit funds.

“Two-trillion-dollar problem” and off-balance-sheet liability timing

• The video claims a broader structural issue is looming: future payment obligations for data center projects may be large and may not fully appear in financial statements until facilities are operational.
• The presenter argues these liabilities are often accounted for only after completion, meaning accounting/reporting may temporarily understate the eventual scale of cash outflows.
• Framing: liabilities rise over time, but private credit growth has stopped and funding is no longer plentiful.

IPO timing and “force majeure” language connected to slowing/pausing AI spend

• The presenter connects financing stress to wider AI industry debates.
• In the video’s telling, major AI executives (including figures linked to Nvidia / Anthropic / OpenAI, and Elon Musk) have reportedly called for the U.S. government to intervene to slow AI development.
• The presenter jokes this could become another force majeure rationale to delay spending.
• Commercialization is presented as an accelerant: as AI products scale, spending commitments may continue even if customer demand and willingness to pay lag.

Market reality: models are replaceable; customer monetization assumptions may be wrong

• The presenter claims AI customers—especially enterprise users—show little loyalty to specific model providers, switching quickly between options (e.g., OpenAI/Anthropic/Google) via API keys and tooling.
• The video compares LLM providers to utilities/mobile operators: users pay for performance, not long-term attachment to a specific provider.
• It argues rapid model switching undermines long-term pricing power that some “trillions” valuation narratives assume.

Valuation and amortization pressure: long leases/depreciation extend cash and P&L burden

• The video argues hyperscalers and AI infrastructure investors extended financial amortization schedules (including data center lifespans and GPU depreciation) by:
  • reducing near-term hit timing, while
  • extending how long costs flow through depreciation and profit/loss—potentially worsening earnings pressure if demand drops or AI becomes cheaper (“race to zero”).

Overall conclusion

The presenter concludes that the combination of:

1. Data center execution risks (power delays)
2. Private credit withdrawal pressure
3. Higher alternative yields pulling money away
4. Off-timing / underrepresentation of liabilities
5. Uncertain end-customer willingness to pay

is creating cascading liquidity and solvency risk—starting in private lending and potentially spreading outward through the AI buildout financing chain.

Presenters or contributors

• Sasha (the sole on-screen presenter/voice in the provided subtitles)