Peer HRT using Blackwell silicon
NVIDIA highlighted that Hudson River Trading is leveraging Blackwell architecture and Spectrum‑X Ethernet networking to link researchers to large compute pools for data ingestion, model training, simulation and deployment — an AI‑heavy approach to accelerating trading workflows. The post frames another top HFT firm tying advanced silicon and networking closely to its research‑to‑production loop. (x.com/NVIDIAAP/status/2034495078446625273)
HPE’s March 16, 2026 AI Factory announcement lists Hudson River Trading among organizations adopting NVIDIA Blackwell–powered systems as part of HPE’s full‑stack AI offerings for service providers and sovereign deployments. (businesswire.com) NVIDIA’s Spectrum‑X documentation cites up to 1.6× improved AI networking performance and “deterministic” latency and isolation features for large GPU clusters. (nvidia.com) The Spectrum‑X SN5600 switch family supports port speeds up to 800 Gb/s, a capability NVIDIA and partners position to move massive model training traffic across Ethernet fabrics. (tomshardware.com) Hudson River Trading’s AI lab lead Iain Dunning presented “Applying Artificial Intelligence in Quantitative Trading” at NVIDIA GTC, where HRT’s scale was described as trading hundreds of millions of shares a day across more than 200 markets and executing over a million trades daily. (nvidia.com) NVIDIA’s Spectrum‑X architecture is deployed alongside BlueField SuperNICs and advanced RoCE/lossless Ethernet features to offload communications and provide performance isolation between large GPU pools. (firstpasslab.com) Public disclosures show hyperscalers and OEMs — including Microsoft’s Fairwater deployments and public plans from Oracle and Meta — adopting Spectrum‑X and Blackwell platforms, aligning HRT’s reported stack with a wider industry move to Ethernet‑centric AI factories. (convergedigest.com)
