Backblaze and WEKA have partnered to simplify data management across the AI lifecycle by combining WEKA NeuralMesh with Backblaze B2 cloud object storage.
AI storage layers
The arrangement is intended to give AI infrastructure teams a validated setup for handling data at different stages of model development and deployment, rather than keeping all workloads on the same type of storage.
Under the partnership, WEKA's NeuralMesh will serve workloads that need fast data access, while Backblaze B2 will provide the capacity layer for larger datasets, checkpoints, and retained AI assets. The model is aimed at organisations managing data across ingestion, training, checkpointing, inference, and later workflows.
The tie-up reflects a broader challenge in AI infrastructure as companies try to balance GPU demands with the cost of storing and retrieving large volumes of data. Training datasets, media libraries, and model outputs can grow rapidly, while not every stage of the process requires the same level of storage performance.
Data access
Customers will be able to retain raw and unstructured data in B2, then make it available to NeuralMesh when it becomes part of a workload that needs faster access. Once those workloads are complete, checkpoints, outputs, and other assets can be returned to B2 for later use or recovery.
The integration has already been sized, tuned, and tested, which the companies argue could reduce the engineering work required to deploy the two platforms together. That may appeal to AI teams seeking to avoid building their own connections between high-speed storage systems and lower-cost capacity tiers.
A NeuralMesh feature known as Snap-to-Object has also been tested with Backblaze B2, according to the companies. They said it allows teams to revert to a checkpoint from a training run or recover saved inference data from the same object storage layer used for other retained assets.
Certification of B2 Cloud Storage for NeuralMesh is under way. No further technical details on the certification process were provided.
Storage performance
Backblaze has focused its pitch on storage capacity for AI workloads, while WEKA has concentrated on storage and data access for compute environments that rely heavily on GPUs. The partnership brings those two layers together in a single design aimed at customers running large-scale AI systems.
The announcement comes as infrastructure suppliers compete to offer more integrated AI stacks, particularly around storage, networking, and compute orchestration. For many users, storage architecture has become a practical concern as organisations move from experimental projects to sustained model training and inference at larger scale.
Company comments
Gleb Budman, Chief Executive Officer, Backblaze, said the partnership addresses both performance and storage volume requirements.
"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 Budman.
WEKA described the issue as one of diverging storage needs within AI systems, where some tasks require very low-latency access and others involve long-term data retention at scale.
"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, Chief Strategy Officer, 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."