Backblaze, Inc. the high-performance capacity storage layer for AI, will exhibit and present as a sponsor at AI Infra Summit 2026, Sept. 15–17 at Santa Clara Convention Center. AI infrastructure teams are facing existential challenge as datasets outgrow what premium flash can affordably hold.
Backblaze AI workflow expert Kamren Mahaney will explain framework for architecting AI data plane to achieve strong storage economics while maintaining performance needed to keep GPUs productive at scale.
Expert Speaking Session: Architecting the AI Data Plane: Pack, Place, Prefetch will be presented Tuesday, Sept. 15 at 1:00 p.m. PT in Expo Theater 1.
Mahaney's session will cover three techniques: packing many small acquisition files into training-friendly shards to reduce request overhead, placing durable data and active working sets across object storage and flash, prefetching to keep data moving ahead of GPU consumption. Attendees will leave with practical framework for designing storage and data-movement architectures that keep compute fed while allowing data to scale efficiently.
"Most teams hit the same wall. They start by putting everything on flash because that's what keeps GPUs busy, and then the dataset grows and the math stops working," said Mahaney. "But tiering only helps if the data moves well. Most pipelines are pulling millions of small files individually, and that costs you throughput on every epoch. By packing them into training-friendly shards, placing the working set close to compute, and prefetching ahead of consumption, you change what your pipeline can do without changing what you spend."
Validated with WEKA
Event comes one week after Backblaze announcement of collaboration with WEKA, AI data and memory infrastructure company, giving customers validated way to keep performance-sensitive workloads on WEKA NeuralMesh while retaining large datasets, checkpoints, and outputs in Backblaze B2. Certification of B2 Cloud Storage for NeuralMesh is underway.
About Backblaze
Backblaze is high-performance capacity storage layer for AI. Built over two decades, company has turned hardware, software, and operational innovation into platform that delivers performance and economics AI era demands — without lock-in. Today, more than 500,000 customers trust Backblaze to move and store data powering businesses.
