Hudson River Trading is a useful reminder that AI adoption is no longer just about access to better models. The harder question is whether heavy usage is producing enough value to justify the cost.
Hudson River Trading is not the kind of firm that can afford to treat AI as a novelty. In a business built on speed, models and engineering discipline, every tool has to prove itself against the work it is supposed to improve.
That makes the firm a good lens for a broader problem now landing on founders, operators and finance teams. AI looks simple from the outside because the interface is simple. Behind it sit tokens, compute, vendor contracts, security decisions, hardware planning and the less visible cost of asking people to review what machines produce.
Bloomberg's Odd Lots spoke with Iain Dunning, head of AI research at Hudson River Trading, in an episode published on Oct. 31, 2025. The discussion was not about a crypto token burn or an on-chain event. It was about how a major market maker thinks about artificial intelligence inside a business where technical advantage has to be measured, not assumed.
HRT describes itself as a multi-asset class quantitative trading firm that provides liquidity across global markets. That matters because this is not casual software adoption. The firm's work depends on low-latency systems, mathematical research and a culture where vague productivity claims are unlikely to travel far. If AI helps research, coding, operations or trading infrastructure, the value can be tested. If it simply creates more activity, it becomes another cost center.
For many companies, the first phase of AI adoption was deliberately loose. Give employees access to the tools. Encourage experiments. See where usage appears. That made sense when the bills were small and the goal was to learn quickly.
That period is ending. Token usage is becoming a budget line, and in some organizations it is becoming the clearest signal of how deeply AI has entered daily work. The problem is that usage is not the same thing as leverage. A team can generate drafts, summaries, code suggestions and meeting notes all day without changing the economics of the business.
That is the lesson for entrepreneurs. Adoption means people are using AI. Leverage means AI is improving output, margins, speed or decision quality. Those are not interchangeable, and confusing them is how companies end up mistaking noise for progress.
The Infrastructure Question Is Getting Closer
The next issue is infrastructure. Once AI moves beyond casual use, companies have to think about model choice, latency, data controls, vendor dependence and whether the most expensive model is actually the right one for the task.
For a firm like HRT, that question is natural. Trading companies have long understood that infrastructure is not just support. It is part of the strategy. Small delays can matter. Weak systems can change outcomes. If AI becomes part of the workflow, then the cost and performance of that AI stack become strategic as well.
Most startups will not build custom AI infrastructure, and they should not pretend they are Hudson River Trading. But they can borrow the discipline. A small model may be enough for classification or routing. A frontier model may be justified for complex coding, research or analysis. A human may still be the right answer when judgment, accountability or customer trust matters most.
The useful question is not whether a company is using the most powerful model available. It is whether the company is matching the tool to the job and measuring what comes back. That is where AI strategy starts to look less like experimentation and more like operating management.
Why This Matters Beyond Trading
HRT's example matters because AI is moving from novelty into operations. The moment a company starts asking what its token burn looks like, the conversation has changed. It is no longer about whether the technology is impressive. It is about whether the business can absorb the cost and convert it into durable advantage.
That is especially relevant for startups, where waste can hide behind excitement. A founder can justify almost any AI bill by calling it experimentation. For a while, that may even be true. But experimentation needs a clock. At some point, the company has to know which workflows improved, which tools saved time, which costs scaled too quickly and which habits were just dressed up as innovation.
The best companies will not retreat from AI because usage becomes expensive. They will become more precise. They will track where AI removes bottlenecks, where it creates new review burdens and where employees are using it because the company rewarded activity instead of outcomes.
That is the practical takeaway from Hudson River Trading's AI discussion. The firms that benefit most will not be the ones that spend the loudest. They will be the ones that understand the economics early, measure the work honestly and tie AI consumption to real business progress. Watch that shift closely, because it is where the AI story becomes less about hype and more about discipline.
Also read: Mira Murati is bringing Thinking Machines Lab out of stealth • Monako Glass puts AI coding agents on a developer's face • Apex shows investors are moving deeper into the space supply chain
