Ken Griffin says everyone is misinterpreting the AI revolution — and wishes Zohran and Bernie would ‘read a damn history book for once’

Citadel founder Ken Griffin said many corporate “AI” claims are really machine learning/optimization, but he’s been convinced by a specific agentic system that reproduced and verified finance papers in 2–3 hours versus 6–8 weeks for experts. He also argued markets can hit record highs despite wars due to concrete energy/China-demand shifts, and flagged that a Taiwan blockade of TSMC chips could cut US GDP by ~8% within six months. On policy, he criticized New York’s tax burden versus services and cited Citadel’s and Goldman’s expansion outside New York as a response.

Analysis

The market takeaway is not "AI is everywhere"; it is that verifiable workflow automation is starting to compress the labor content of high-end knowledge work, while most public-market AI narratives remain unproven. That favors firms with dense internal processes and proprietary data, but the payoff shows up first in expense discipline, not revenue acceleration. GS is a modest beneficiary if it can turn this into lower run-rate comp or support headcount leverage over the next 2-4 quarters; the issue is that the stock already discounts elite execution, so the re-rating ceiling is limited unless management quantifies savings.

The bigger second-order effect is competitive: if a small team can do work that previously required a much larger analyst stack, the moat shrinks for research-heavy intermediaries, consultants, and niche asset managers. That is structurally negative for labor-arbitrage business models and should widen dispersion between firms that own data/processes versus those that rent talent. For TSM, the key implication is not next-quarter earnings but persistent geopolitical optionality: the tail risk remains low probability, high severity, and likely keeps a discount embedded in the multiple even if fundamentals stay strong.

Contrarian view: consensus may still be overpaying for broad AI beneficiaries while underappreciating boring automation in financial services and industrial workflow. The article argues that the real near-term gains come from machine learning, optimization, and digitization, which means the winners are likely to be incumbents that can industrialize those tools internally rather than pure-play AI vendors. What would falsify this: no visible expense leverage at GS by the next two reporting cycles, or a sustained de-escalation in Taiwan risk that materially compresses TSM's geopolitical premium.

Key Decisions for Investors

• GS: hold/add on pullbacks over the next 1-3 months only if management commentary points to measurable expense leverage; target a modest relative outperformance, but cap upside expectations because the market already gives GS a quality premium. Fails if the efficiency story never appears in reported comp ratio or non-comp run-rate.
• TSM: do not chase the stock on this article; treat it as a geopolitical hedge rather than a conviction long. If implied volatility is cheap, buy downside protection into any Taiwan/China escalation window; abandon the hedge if policy rhetoric de-escalates or if the premium collapses without follow-through.
• BA: keep on the watchlist as an indirect loser in a Taiwan supply-chain shock scenario, but do not short it solely on this catalyst. Use any semis/geopolitics selloff to reassess, and invalidate the bearish overlay if Taiwan risk fades for a full quarter.
• Avoid buying 'AI narrative' exposure indiscriminately; prefer firms that can prove internal productivity gains in earnings or investor-day disclosures. If a company cannot show lower SG&A or faster cycle times, assume the AI label is mostly marketing.