Hudson River Trading (HRT), a leading market maker known for deploying artificial intelligence technology, detailed its experiences and challenges with AI at a Bloomberg live show on June 5, 2026, held at New York's City Winery [1, 2].
Iain Dunning, HRT's head of AI, discussed the firm's significant use of tokens within its AI systems and the associated expenses [1, 2]. Employees at HRT consume substantial amounts of these tokens, highlighting the resource intensity of their AI operations [1, 2].
Dunning also raised concerns about memory and computing bottlenecks that impact AI deployment at HRT, pointing to the high costs and technical hurdles the firm faces in scaling these systems [1, 2].
In response to these challenges, Hudson River Trading is actively considering developing its own semiconductor chips to better handle the specific workload demands of AI software [1, 2]. This move aims to optimize performance and potentially reduce reliance on commercially available hardware.
The discussion also touched on phenomena described as AI-induced delirium, emphasizing the complex behavior and unexpected outcomes encountered when operating advanced AI models [1, 2].
These insights follow initial conversations with Dunning about HRT's AI use that began in late 2025, about seven months before the public Bloomberg event [1, 2]. The firm is continuing its efforts to innovate around AI infrastructure and efficiency.
Hudson River Trading’s next updates on hardware development and AI deployment strategies will likely come as it actively explores bespoke chip designs to support its evolving AI needs [1, 2].
