Bridgewater Associates partners with YellowDog to operate at a scale of compute that fundamentally changes the economics and velocity of quantitative research.
Executive Summary
As Bridgewater’s quant teams built increasingly large and complex ML models, infrastructure quickly became the constraint. Working closely with YellowDog gave research teams “on tap” access to parallel compute across multiple AWS regions, powered by dynamically sourced, cost-efficient Spot capacity at scale.
High-scale workload execution is embedded directly into Bridgewater’s research workflows, giving each team the autonomy to build and run complex models without requiring infrastructure teams to pre-provision resources or manually intervene. Production clusters now regularly surge beyond 35,000 nodes so results are achieved faster.
- Increased speed of research — by running massively in parallel across tens of thousands of nodes, YellowDog delivers more than 7× faster research velocity.
- Unlocking new possibilities — increased compute scale helps drive the creation of new models and ideas.
- Cost-efficient scale — consistently high Spot utilisation across multiple AWS regions delivers over 70% cost savings versus on-demand pricing.
YellowDog Enables High-Scale Global Compute
Quantitative research is, at its core, about maximizing compute efficiency to run more experiments, test more hypotheses, and act on signals faster than the competition. The ability to access and orchestrate compute at scale – within acceptable cost envelopes – is a decisive factor in research velocity.
Bridgewater selected YellowDog because true performance at scale demands intelligent, global workload management. YellowDog helped increase Bridgewater’s compute ceiling while maintaining close to 100% Spot utilisation, expanding the potential for new ideas more quickly and cost-efficiently.
Intelligent Global Orchestration At Production Scale
The production evidence is unambiguous. Bridgewater’s infrastructure team seamlessly spins up nodes on a massive scale across multiple AWS regions, executing large-scale workloads with the most compute-intensive portions completing in less than 3 hours, representing more than 7× faster research velocity.
Multi-Region Spot Strategy
By dynamically distributing cloud worker nodes across multiple AWS regions, YellowDog enables Bridgewater to “follow the moon,” drawing from regions where Spot capacity is most available and cost-efficient at any given time. This geographic agility ensures that supply constraints in a single region do not force a fallback to on-demand pricing. The result: Bridgewater achieves 99.7% Spot utilisation with an estimated 70% cost savings compared with on-demand pricing.
What Performance At Scale Looks Like in Production
Orchestration & Infrastructure That Accelerates Alpha
YellowDog enables Bridgewater to operate a compute engine that would be unachievable using conventional infrastructure. By dynamically drawing from multi-region Spot capacity across dozens of instance types, Bridgewater’s quant teams execute workloads that deliver results faster.
The numbers are clear: production runs regularly reaching 35,000+ nodes; validated 7× faster research velocity; consistently high Spot utilisation across multiple AWS regions; and an estimated 70% savings versus on-demand pricing.
