Why AI Still Struggles on Trading Desks Despite Wall Street’s Rapid Adoption
While Wall Street banks widely adopt AI for knowledge tasks, its application in live trading faces significant hurdles due to the complex, numerical nature of market microstructure data. Finance-native AI and robust infrastructure are proving essential to overcome these challenges, as general-purpose models fall short in real-time market dynamics.
Wall Street banks have rapidly integrated AI assistants into research, coding and other knowledge-intensive operations, but their adoption has yet to translate seamlessly into live trading. According to Yianni Gamvros, CEO and co-founder of Paris-based Quantum Signals, general-purpose AI models struggle with the millisecond-level, numerical data that drives intraday markets.
Unlike text-heavy tasks where large language models excel, trading requires real-time analysis of price movements, liquidity, volume and volatility embedded in complex market microstructure data.
The challenge extends beyond model capability to the difficulty of deploying AI reliably in live markets, where systems must process data and generate predictions with both speed and precision. Gamvros argues that finance-native AI models, designed specifically for high-frequency financial time series, are better suited to intraday systematic trading than general-purpose models, although proprietary data, infrastructure and financial expertise remain equally important.
AI can already identify patterns in price movement, liquidity and volatility, but determining trade size, entry and exit points and the underlying causes of those patterns still requires substantial human judgment and quantitative expertise.
As financial institutions experiment with increasingly sophisticated AI systems, the competitive advantage may increasingly depend on the combination of specialised models, proprietary data and robust infrastructure rather than AI alone. Gamvros maintains that newer techniques can overcome some long-standing assumptions about AI’s limited usefulness in markets, including low signal-to-noise ratios and constantly changing market conditions.
With finance-specific foundational models reportedly being used by leading quantitative firms, institutions unable to develop or access comparable capabilities could face a growing technological disadvantage in the race for increasingly sophisticated trading strategies.