Ray Day NYC Spotlights AI Scaling from Coinbase, Discord, Torc
Jessie A Ellis Jun 10, 2026 16:31
Ray Day NYC featured Coinbase, Discord, and Torc Robotics sharing how Ray boosted AI workloads. Highlights include Torc's 90% GPU utilization.
Ray Day NYC brought together engineers from leading companies like Coinbase, Discord, and Torc Robotics to showcase how they’re leveraging the open-source Ray framework and Anyscale to scale AI workloads. Held at One Liberty Plaza, the event highlighted real-world production challenges and how Ray unlocked new efficiencies, including Torc Robotics achieving 90% GPU utilization, a dramatic leap from its prior 30-40%.
Torc Robotics: Scaling Multimodal AI for Autonomous Trucks
Neil Wadhvana, ML Ops Tech Lead at Torc Robotics, outlined the stakes for autonomous trucking—targeting a $200 billion U.S. long-haul market by 2030 amid a 160,000 driver shortage. Torc consolidated its fragmented multimodal data processing stack into a single Python-based engine powered by Ray. This shift allowed independently scaled CPU and GPU node pools, boosting GPU utilization to 90%. Training time per epoch dropped from 20 minutes to 5, delivering a 4x improvement without additional hardware. Their pipelines now handle up to 38 TB of training data with ease. For those interested, Torc is hosting a webinar diving deeper into these advancements.
Discord: From Open-Source Ray to Anyscale
Discord’s Senior Software Engineer Serrana Aguirregaray detailed how the company scaled its ML platform from open-source Ray to Anyscale. Initially used to train recommendation models for 90+ million daily active users, Discord's adoption of Ray expanded rapidly across teams handling ads, safety, and content understanding. However, managing multiple Kubernetes-based Ray clusters became a bottleneck. Transitioning to Anyscale resolved these issues with a unified control plane and declarative configs, enabling more focus on model development. Discord’s first deep learning use case improved ad rankings by over 200%.
Coinbase: Scaling Financial Risk Analysis
Coinbase’s Aman Choudhary shared how the company uses Ray and Anyscale to process thousands of batch ML jobs daily for financial risk prediction. After moving from SageMaker to Ray in 2023, Coinbase cut costs by 20% and slashed iteration times from hours to seconds. Subsequent optimizations with Anyscale enabled the team to scale from 3,000 degraded jobs to over 10,000 stable daily jobs, significantly improving operational efficiency.
Market Context and Relevance
The innovations on display at Ray Day NYC underscore the growing importance of scalable AI infrastructure as industries integrate machine learning into critical workflows. These developments also align with broader trends in blockchain and crypto, where decentralized platforms like Raydium (RAY) are leveraging advancements in distributed computing to push tokenized real-world asset adoption.
Raydium, a Solana-based decentralized exchange, recently surpassed $1 trillion in cumulative trading volume and reported over $2 billion in tokenized equity trading volume. While the RAY token saw a modest 2.81% price increase today to $0.5820, the ecosystem faces challenges, including a $1.34 million exploit of an old liquidity pool. The convergence of AI scalability and blockchain technology could present significant opportunities for platforms like Raydium to enhance their infrastructure.
What’s Next
Ray Day NYC is part of a broader “Ray on the Road” initiative by Anyscale, culminating in the Ray Summit in San Francisco on August 24. The event will feature keynotes, workshops, and product roadmap previews. With Ray’s growing traction across industries, the summit is poised to set the stage for the next wave of AI scalability.
Image source: Shutterstock.jpg)