VDURA, Wasabi link AI storage with cloud archiving lifecycle

VDURA and Wasabi have formed a technology alliance focused on AI data storage, linking VDURA's storage systems with Wasabi's S3-compatible cloud storage.

The arrangement targets organisations running AI factories, neoclouds and enterprise high-performance computing environments that need to manage both active and inactive data.

Under the alliance, VDURA will manage data used for dataset staging, model loading, training, checkpointing and inference, while Wasabi will provide cloud object storage for long-term retention and archiving. The design is intended to keep active datasets close to graphics processing units while moving older material into cloud storage when it is no longer regularly used.

This model addresses a persistent issue in AI infrastructure: many teams keep large volumes of inactive data on expensive performance systems because moving it can add complexity, slow access or create uncertain cloud costs.

As a result, storage built for high-speed processing can end up holding checkpoints, dataset versions and model artefacts that no longer need to sit near compute resources. That reduces available capacity for current workloads and complicates protection, governance and later reuse.

VDURA's systems provide a parallel file system with RDMA data paths, full POSIX workflows, support for NVMe flash and hard disk drives, and a native S3 interface within a single global namespace. Wasabi will store datasets, checkpoints, model versions and derived artefacts for retention, disaster recovery and distribution to other sites or computing environments.

Cost focus

A central element of the alliance is the cloud storage pricing model. Wasabi said its standard terms do not include per-gigabyte egress fees or API request charges, which the companies argue makes storage and retrieval costs easier to predict for data users may want to reuse later.

The approach reflects a broader infrastructure pattern used by major cloud operators, where fast storage tiers are paired with deeper, lower-cost capacity tiers. VDURA and Wasabi are applying that design to customers building their own AI environments, using open interfaces rather than a single integrated stack.

Ken Claffey, Chief Executive Officer of VDURA, outlined the rationale for the deal.

"The infrastructure that feeds GPUs is engineered for velocity, and data belongs there while it is doing active work. It should not live there permanently," said Claffey.

"Our customers generate enormous volumes of checkpoints, dataset versions and model artifacts that hold real long-term value. Pairing VDURA's GPU-adjacent performance with Wasabi's open, predictable cloud layer gives them a place for that data to be retained, protected and reused, and makes data movement a native part of the AI data lifecycle," he said.

Data lifecycle

The companies are targeting a part of the market where storage architecture is becoming more important as AI workloads scale. Training and inference systems generate growing volumes of supporting data, and operators increasingly need to decide what must remain on premium infrastructure and what can move elsewhere without becoming inaccessible.

That question is becoming more pressing as enterprises seek to reuse earlier training assets, compare model versions, retain audit trails and meet internal governance requirements. The alliance positions cloud object storage as the destination for those older assets once they leave the active training cycle.

Laurie Mitchell, Senior Vice President, Global Marketing, Wasabi Technologies, said the value of AI data extends beyond a single training run.

"AI data does not lose its value when a training run ends. It becomes the raw material for the next model, the audit trail for governance and the baseline for comparison," said Mitchell.

"This alliance makes it straightforward to connect GPU-adjacent infrastructure with independent, S3-compatible cloud object storage, so organizations keep control of their data and their costs without locking either one into a hyperscaler," she said.

The deal brings together two specialist providers at a time when AI infrastructure buyers are weighing whether to assemble systems from separate vendors or rely on larger cloud platforms for storage and compute. By linking a high-speed on-premises or near-GPU storage layer with an external object store, the companies are making the case for a mixed model in which active and inactive data are handled differently.

For customers building in-house AI estates, that could mean reserving premium storage for live workloads while shifting retained data to a cheaper cloud tier that remains available for retraining, compliance and cross-site access.