Uploaded September 2026 | Updated November 2026, 1 day ago
What happens to storage architecture when AI workloads move from terabytes to petabytes and throughput requirements move toward terabits per second?
Backblaze CTO Dan Spraggins examines the engineering consequences of AI-scale data and the role of HDD-based object storage throughout the AI data lifecycle.
The session explores SSD and HDD infrastructure tradeoffs and how capacity storage supports ingest, curation, training, checkpoints, inference, logs, and archive. Dan also looks at the engineering foundation behind Backblaze’s platform, including 5+ exabytes of customer data under management, more than 350,000 drives, and infrastructure designed for eleven nines of durability.
Real-world examples from Hume AI, Decart, and CoreWeave illustrate high-throughput ingestion, multi-cloud AI training, and large-scale AI cloud storage.
What happens to storage architecture when AI workloads move from terabytes to petabytes and throughput requirements move toward terabits per second?
Backblaze CTO Dan Spraggins examines the engineering consequences of AI-scale data and the role of HDD-based object storage throughout the AI data lifecycle.
The session explores SSD and HDD infrastructure tradeoffs and how capacity storage supports ingest, curation, training, checkpoints, inference, logs, and archive. Dan also looks at the engineering foundation behind Backblaze’s platform, including 5+ exabytes of customer data under management, more than 350,000 drives, and infrastructure designed for eleven nines of durability.
Real-world examples from Hume AI, Decart, and CoreWeave illustrate high-throughput ingestion, multi-cloud AI training, and large-scale AI cloud storage.










