Backblaze and Weka Simplify Data Management Across the AI Lifecycle

Backblaze Inc., a high-performance capacity storage layer for AI, and Weka, an AI data and memory infrastructure company, announced a collaboration designed to simplify how AI teams manage data across the AI lifecycle. As AI teams generate enormous volumes of data across ingestion, training, checkpointing, inference, and downstream workflows, the solution gives customers a validated way to keep performance-sensitive workloads on Weka NeuralMesh while retaining large datasets, checkpoints, and outputs in Backblaze B2, with the integration, sizing, tuning, and testing already done.

“AI teams need their GPUs fed and an infrastructure with the performance and capacity to support the full AI data workflow. Weka has mastered the performance tier. We’ve spent nearly two decades doing the same for capacity storage. Together, AI teams get a qualified, complete solution to ensure fast and efficient production,” said Gleb Budman, CEO, Backblaze.

Weka’s NeuralMesh is built for performance-intensive AI and accelerated computing environments, delivering predictable performance as workloads, datasets and GPU clusters scale. Backblaze B2 provides cloud object storage capacity for large datasets and retained AI assets. The work validates the two platforms together, so customers can deploy a proven integration instead of building and testing one themselves.

“AI workloads are stretching storage in two directions at once. GPUs need microsecond access to data to stay fed, while datasets and checkpoints are growing to exabyte scale,” said Nilesh Patel, CSO, Weka. “Our collaboration with Backblaze gives customers a validated path to both – without the cost of building and testing that integration themselves. Speed where it matters, scale wherever you need it.”

Customers can retain raw, unstructured data (training sets, media libraries, source files) in Backblaze B2. When those datasets become part of a performance-sensitive workload, they can be made available to NeuralMesh and served to accelerated compute. Checkpoints, outputs and other assets that no longer require high-performance access can be retained in B2 for reuse in future workloads or in situations where recovery to an earlier stage of testing is necessary. NeuralMesh’s Snap-to-Object capability, tested with Backblaze, lets teams revert to a checkpoint from a training run or recover saved inference data without improvising a fix mid-run, pulling from the same B2 capacity tier everything else already lands in.

Availability
Certification of B2 Cloud Storage for NeuralMesh is underway.