The head of AI at Hudson River Trading suggests the industry is moving past needing human-understandable narratives for trading strategies. If a model, after rigorous backtesting, finds a pattern that works, it's traded, even if the logic is incomprehensible or feels like a "loss of control" to humans.
The head of AI at Hudson River Trading highlights a practical barrier to creating a financial market for compute. For serious training, the minimum "lot size" is thousands of GPUs, not a small, fungible unit. This makes it difficult to standardize a contract and create liquidity, unlike commodities with smaller, interchangeable units.
A top practitioner at Hudson River Trading clarifies that securing GPUs isn't the primary challenge. The real bottleneck is finding available data center capacity and power at short lead times. Even if chips are available for delivery, the complete "solution" of a powered, operational site is scarce and fiercely competitive.
With AI models capably handling implementation, Hudson River Trading is shifting its hiring focus. The firm can now hire "theorists" or "dreamers" who excel at ideation but may lack coding skills. The ability to clearly articulate ideas and prompts to an AI has become a highly valued skill in itself.
The head of AI at Hudson River Trading describes an incredibly competitive market for GPU capacity. Providers offer newly available leases that require a commitment to multi-year contracts for thousands of GPUs by the end of the day. This high-stakes, high-speed environment means buyers cannot be picky about location or terms.
In the high-stakes world of securing GPU capacity, counterparty risk is a major factor for both sides. Data center providers scrutinize the financial stability of tenants like Hudson River Trading (asking about bond ratings), while HRT in turn analyzes providers' credit default swaps to hedge against their potential failure.
The use of large language models for research and coding has introduced a significant new operational cost. At Hudson River Trading, individual AI researchers can spend between $100 and $1,000 per day on API tokens. This creates a "token rich" vs "token poor" dynamic, potentially accelerating the gap between well-funded teams and others.