Hudson River Trading AI head Iain Dunning says matching fast-evolving AI models to quantitative trading requires strategic GPU scaling

It’s so much fun talking to @iaindunning

He told us the leap from Opus 4.0 to Opus 4.5 was the moment that these tools went from being something impressive to something that could actually right now do the work of a junior employee.

@TheStalwart @iaindunning @tracyalloway @WeAreHRT lol at the magic box reference

@iaindunning We also talked about booming token consumption, quant interviews (and how he can tell when a candidate used a chatbot for interview prep), and whether HRT would ever build its own LLM (à la High Flyer and DeepSeek).

@TheStalwart @iaindunning @tracyalloway @WeAreHRT Near the top Joe asked the guest for a specific example of something 2026 AI enables the firm to do that 2024 AI did not. He didn’t seem to answer this. Anyone else hear it differently?

@Stephen1426397 @TheStalwart @tracyalloway @WeAreHRT I didn't really answer it directly, I don't think I had a good direct answer that I can share without way too much context (and IP sensitivity concerns). Perhaps "simultaneously integrate much more information in one model" could be an answer.

@TheStalwart @iaindunning @tracyalloway @WeAreHRT Tbh the idea of just following the trades output from a black box prediction engine is not new. Many years ago Jim Simons was talking about how this is what they do at Renaissance.

@TheStalwart @iaindunning 4.5 was the eye opening moment.

Many people tried it out over winter break 2025 (shortly after it came out) and had their “oh shit” moment. Things really took off in early 2026 after that.

@TheStalwart @iaindunning @tracyalloway @WeAreHRT @TheStalwart An interesting question is how ROI is being measured internally and how CIOs plan to shift budgets if ROI isn't as healthy as expected 👇

What do you think of this tracker?

@iaindunning @TheStalwart @tracyalloway @WeAreHRT Fair enough, thanks for your reply

@TheStalwart @iaindunning @tracyalloway @WeAreHRT trading firms turning into GPU hunters now

@TheStalwart @iaindunning @tracyalloway @WeAreHRT Compute scarcity makes infrastructure geography decisive. OpenAI's G42 partnership addressed this directly.

@TheStalwart @iaindunning @tracyalloway @WeAreHRT HRT seeing insane leaps in AI capabilities is the real signal here — more than the GPU deployment hunt. Curious if the edge they're referencing is model architecture or just scale.