A Ceiling That Was Holding Back Research
As Bridgewater’s quant teams built increasingly complex ML models, infrastructure became the bottleneck. The scale their models demanded simply wasn’t achievable at the level of speed needed to get results faster.
The requirement was clear: massive parallel compute, on tap, at near-total Spot pricing, with no pre-provisioning and no manual intervention from infrastructure teams.
What YellowDog Delivers In Production
The numbers speak for themselves:
- 35,000+ nodes in a typical production run – across multiple AWS regions simultaneously
- 7× faster research velocity – the most compute-intensive workloads complete in approximately 3 hours
- 99.7% Spot utilisation – using dynamic orchestration across AWS regions to access preferred instances types for example, Arm Graviton, whilst never forcing a fallback to on-demand pricing
- 70% cost reduction versus on-demand
Researchers access compute and submit workloads directly within existing workflows, so infrastructure teams can better plan capacity and manage cost, whilst giving quant research teams the freedom to go faster.
In Bridgewater’s Own Words
“What differentiates YellowDog from any other provider is the efficiency and speed with which it achieves scale. It automatically orchestrates compute so our research teams can run extremely large workloads without the complexity and cost of building that capability ourselves.”
“YellowDog enables us to continuously push compute scale further. Using global cloud Spot availability has enabled us to use compute efficiently — which is key to sustaining our research velocity and competitive performance.”
— Artem Kazakov, Lead Architect at AIA Labs, Bridgewater Associates
Why This Matters Beyond Bridgewater
The Bridgewater story is a signal for the wider industry. As AI/ML models grow more complex, the firms that can run more experiments per day at the scale that makes new research viable will become the most competitive. The ability to access massive parallel compute dynamically, across regions, at near-total Spot pricing, without building and maintaining that capability in-house is no longer a complex infrastructure challenge. It is a strategic imperative, that is available to any firm with the quant research capabilities to harness it.
YellowDog exists to make this new world of high-scale, cost-efficient compute accessible to every research team that needs it, without the overhead of building and maintaining that orchestration capability in-house.
Why The YellowDog Team Matters
At YellowDog, the platform didn’t emerge in isolation: it’s the product of a deeply committed team that has pushed the boundaries of what high-scale workload orchestration can achieve. Today, it powers one of the fastest and most scalable workload orchestration platforms on the planet. But just as important is how that team shows its genuine dedication to service, support, and partnership. The same passion that drives innovation in the platform is what ensures our customers can rely on it in production, at scale, when it matters most.
Read The Full Case Study
The Bridgewater case study covers the full story: the challenge, the solution, the production evidence, and what performance at extreme scale looks like in practice, including how YellowDog’s platform with multi-region Spot enables Bridgewater to “follow the moon” for lowest-cost capacity at any given time.
Read the Bridgewater Associates Case Study.
—
YellowDog will be at Future Alpha in New York City March 31 – April 1. If you are planning to attend come meet us at Stand B6.
Contact us here to discuss your workload requirements.
