AI boom risks global financial crash, warn central bankers

Original Article ↗ Hacker News Discussion ↗

Central bank warnings & historical context

  • Some readers look for past central-bank warnings (e.g., 1999, 2007 BIS reports) and feel the 2026 AI warning is unusually explicit.
  • Others note we’d also need to know how often such warnings were issued without a major crash, and that warnings can become “self-defeating prophecies” if they reduce risk-taking.

Scale and structure of the AI boom

  • BIS excerpts highlight >$1T in AI capex by a handful of hyperscalers in 2025–26, partly debt-financed, with uncertain returns.
  • Risks cited: over-investment driven by winner-take-most expectations, electricity and chip bottlenecks, and the potential for a synchronized equity/credit repricing and investment bust.

Bubble dynamics and systemic risk

  • Many see a clear AI bubble; debate is whether it ends in a sharp crash or long stagnation.
  • Some argue impact is muted because AI is mostly equity-financed; others worry about knock-on effects via suppliers, credit markets, and retail investors if/when these firms IPO.
  • There’s recurring cynicism that, in a downturn, elites will be bailed out and scapegoats will again be immigrants and the poor.

Labor, capitalism, and long-run compatibility

  • One view: if AI eventually automates most labor, capitalism—premised on labor having positive value—breaks down.
  • Others emphasize shorter-run dangers: rapid white‑collar displacement shrinking consumer incomes and demand, versus a scenario where AI underdelivers and wipes out speculative investment.

Value, rents, and “parasitism”

  • Heated debate over whether frontier AI firms are “parasites” extracting rents:
    • Critics argue they privatize gains from “plundering the commons” (training on others’ work, sometimes via pirated data) and may destroy meaningful intellectual labor.
    • Defenders counter that training has been ruled fair use so far and compare AI to past productivity technologies (tractors, fertilizers), which also caused disruption but created value.
  • Some distinguish genuine value creation from rent-seeking even within the same business models.

Capabilities, SaaS, and productivity

  • Several practitioners report LLMs “one‑shotting” or rapidly iterating SaaS-style apps, cutting costs and time-to-build. Others dismiss these as non–production-grade, “vibe-coded” systems accruing hidden tech debt.
  • There’s concern that ultra-cheap software shrinks parts of the economy: replacing a $XX,XXX/year SaaS subscription with $50 of API calls may not be offset by new hiring or new products.
  • Skeptics argue software engineering isn’t “cracked”: models struggle with requirements, domain understanding, and long-term maintainability.

Sectoral extensions: therapy, bullshit jobs, and last mile

  • Some foresee insurers pushing AI “therapists” as a cheaper first line; others see this as dangerous given known failure modes and weak training data.
  • Discussion of “bullshit jobs”: automation often doesn’t reduce headcount; organizations like having humans attached to decisions.
  • Several argue AI’s gains are currently concentrated in SWE; other fields require heavy “last mile” domain encoding, which may not scale with current LLM architectures.

Opportunity cost and alternative uses of capital

  • A major thread laments $2T+ in AI-related valuation/investment versus funding for infrastructure, clean energy, housing, education, or social support.
  • Counterpoints:
    • This is mostly private capital seeking returns, not tax money.
    • Governments already spend trillions on social programs and infrastructure; the problem may be structural inefficiency and misaligned incentives, not sheer dollars.
  • Proposals include demurrage-based money to reduce hoarding, and large-scale investments with high social ROI (active transport, nuclear, solar, housing) as better macro bets than an AI arms race.

Personal finance reactions

  • Some participants describe shifting heavily into cash/T‑bills, expecting bad news as AI, energy shocks, and existing financial vulnerabilities interact.
  • Others caution timing is uncertain; AI exuberance might be politically sustained for years before any reckoning.