AWS re:Invent 2025 - GenAI, Hold the Waste: How H2O Fixed Its Storage Bottleneck (STG204)

AWS re:Invent 2025 - GenAI, Hold the Waste: How H2O Fixed Its Storage Bottleneck (STG204)

Overview

• H2AI is a global leader in enterprise AI, providing both generative and predictive AI capabilities
• H2AI's entire platform runs on Kubernetes, relying heavily on Amazon Elastic Block Store (EBS) for storage
• H2AI faced challenges with EBS storage, including:
  • Over 2 PB of underutilized storage that was growing rapidly
  • Inability to efficiently scale storage up or down

Datafi: The Autonomous Storage Solution

• Datafi is an autonomous storage solution that manages cloud storage automatically for AWS customers
• Key features:
  • Dynamic autoscaling of EBS volumes based on usage
  • Automatic scaling up when data is added, and scaling down when data is deleted
  • No impact on application performance or downtime during deployment
  • Seamless integration with Kubernetes, Terraform, and various Linux environments
• Datafi works by deploying a low-level agent on EC2 servers or Kubernetes clusters, which dynamically adjusts EBS volume destinations without impacting applications
• A SaaS control plane monitors the agents and provides analytics on storage utilization and efficiency

Integrating Datafi with H2AI's Platform

• H2AI faced a few challenges in integrating Datafi:
  1. Ensuring the Datafi agent runs on their existing Bottlerocket infrastructure in EKS
  2. Maintaining the same level of data security and reliability for their customers
  3. Seamlessly integrating with their existing backup solution (Velero)
• H2AI was able to successfully address these challenges and deploy Datafi across their customer environments

Results and Impact

• Before Datafi, H2AI's EBS capacity utilization was only around 25%, meaning they were overpaying for 4x the storage they were actually using
• After deploying Datafi:
  • Capacity utilization increased to around 80%, the target "sweet spot"
  • Total EBS capacity paid for decreased from over 2 PB to less than 1 PB, resulting in significant cost savings
• The Datafi integration was seamless, with zero downtime or disruption to H2AI's customers
• In addition to cost savings, Datafi also improved the performance of H2AI's EBS volumes

Key Takeaways

• Datafi's autonomous storage management capabilities helped H2AI overcome their EBS storage challenges, including:
  • Underutilized and rapidly growing storage
  • Inability to efficiently scale storage up and down
• The integration was smooth, with Datafi seamlessly integrating into H2AI's existing Kubernetes, Terraform, and backup workflows
• The results were significant, including:
  • Increased EBS capacity utilization from 25% to 80%
  • Reduced total EBS capacity from over 2 PB to less than 1 PB, leading to major cost savings
  • Improved performance of EBS volumes
  • All with zero downtime or disruption to H2AI's customers and applications