Is AI making markets more or less efficient?

Bernstein argues AI makes markets more efficient by processing earnings/regulatory filings faster, narrowing information gaps and reducing earnings surprises. However, it warns AI-driven signals and AI-generated misinformation can amplify volatility and worsen dislocations—citing the Aug 2024 yen carry trade unwind and brief U.S. equity moves from synthetic misinformation. Net effect: lower average inefficiency but higher tail risk, implying more severe reversals during stress and potential reflexivity/crowding.

Analysis

AI-driven research compression is a mixed P&L story for the sell side: it lowers the value of differentiated stock-picking, but it should increase turnover, faster repositioning, and demand for execution/derivatives services. That makes large diversified brokers and market-makers relatively better positioned than boutique research franchises, while under-researched small caps lose the most because their valuation premia were largely an information-rent. The first-order effect is lower dispersion in calm markets; the second-order effect is more violent factor reversals when the same models are crowded into the same names.

The real opportunity is not “AI makes markets efficient,” but that it makes exits more correlated. If macro data, policy headlines, or a synthetic-information scare hits, model convergence can turn a routine de-risking into a gap move over days rather than weeks. That argues for owning convexity when realized volatility is cheap, because the regime risk is asymmetric: upside is incremental efficiency, downside is a sudden liquidity vacuum.

For the tickers here, BAC looks relatively insulated versus smaller brokers because it can monetize higher churn and options flow even if research alpha erodes. SNDK is the cleaner medium-term beneficiary if AI capex keeps pulling through storage demand, but that thesis is cyclical and can be broken quickly by NAND pricing or a capex pause. SMNEY should be treated as a watchlist item rather than a conviction long if it is an emerging-market proxy, since broader AI coverage should compress—not expand—its informational edge over 6-18 months.

Key Decisions for Investors

• Long BAC vs short a more research-dependent broker/asset manager basket for 1-3 months; thesis is that model crowding lifts trading activity while commoditizing analyst alpha. Falsify if BAC trading revenue guidance softens or equity volatility collapses.
• Buy 3-month SPY put spreads as a low-cost tail hedge; the edge is not direction but convexity if synchronized AI-driven positioning unwinds suddenly. Exit if VIX stays pinned and cross-asset dispersion keeps falling.
• Add SNDK on weakness for a 6-12 month AI-storage exposure, but size it as a cyclical growth position, not a pure factor trade. Invalidate if NAND pricing rolls over or hyperscaler capex decelerates.
• Avoid or underweight EM small-cap proxies for now; greater AI coverage should compress the historical valuation premium over the next 6-18 months. Reassess only if idiosyncratic earnings start beating despite tighter information spread.
• If BAC underperforms on a market-wide volatility spike, rotate from outright long to a call-spread structure to keep exposure to higher flow activity while capping downside.