CoreWeave Signs Hudson River Trading Cloud Agreement

HRT has used machine learning, neural-network models, and high-performance computing in quantitative research and trading, according to CoreWeave. The announced platform centers on NVIDIA's Vera Rubin NVL72 infrastructure.

The technical requirements named in the announcement are notable for quantitative research systems. Training experiments can require frequent movement between data storage, feature pipelines, distributed training jobs, validation workloads, and simulation environments. Low latency alone does not determine research throughput, but predictable data movement and stable cluster operations can affect how quickly researchers can run and compare experiments.

Lee said, "As we scale our AI and machine learning research, the AI platform we build on matters as much as the models we build." He added that giving researchers access to newer AI technology enables a research team to push further and faster.

The transaction gives CoreWeave another publicly reported customer relationship in financial services, a market where firms increasingly deploy accelerated computing for model training, backtesting, simulation, and data-intensive research. TipRanks notes that Jane Street, an HRT competitor, previously agreed to spend $6 billion on CoreWeave data center capacity and invested $1 billion in the cloud provider.

TipRanks also reports that Microsoft previously represented more than 70% of CoreWeave sales, and frames the HRT contract as part of broader expansion beyond large technology customers. Those customer-concentration figures and the commercial value of the HRT agreement were not included in CoreWeave's August 20 press release.

For ML infrastructure teams, comparable financial-services deployments illustrate how AI demand extends beyond generative AI applications. High-performance model development in trading can combine deep learning with large-scale historical datasets, repeated experiments, and latency-sensitive production constraints. Industry experience with such workloads indicates that accelerator availability is only one component of usable capacity; network performance, storage throughput, scheduling reliability, and data-governance controls also determine end-to-end research velocity.

CoreWeave chief revenue officer Jon Jones called financial services "one of the most demanding proving grounds for AI" in the company's announcement. The disclosed agreement does not specify HRT's workload mix, capacity commitment, deployment schedule, or the models it intends to train.