Generative AI Service Outages and Price Impact in Cryptocurrency Markets

Abstract

Generative AI has become part of the financial information infrastructure, yet its role in shaping the functioning of capital markets remains poorly understood. We examine 252 service outages identified from the status pages of OpenAI and Anthropic and align their onset times with high-frequency Binance data for Bitcoin and Ethereum.Using a stacked difference-in-differences design, we compare market conditions before and after each outage with those on historical control days matched by weekday and UTC time. Within 15 minutes of outage onset, Amihud illiquidity and the trading-value-scaled price range decline by 15.88\% and 10.87\%, respectively, relative to their counterfactual levels. The decomposition yields point estimates consistent with both narrower price movements and greater trading value. Spot Kyle $\lambda$ falls, while directional order flow and visible depth in perpetual-contract order books do not change significantly. The response is concentrated in ETH, remains robust to alternative specifications and finite-sample inference, and does not differ systematically across providers.