New architecture transforms OCI Bare Metal NVMe into the fastest and most efficient AI storage tier, redefining GPU performance and data access
Hammerspace announced a major breakthrough in AI infrastructure performance with its Tier 0 architecture, demonstrated live at Oracle AI World 2025. The innovation transforms Oracle Cloud Infrastructure (OCI) Bare Metal NVMe into an ultra-fast shared storage tier, enabling continuous GPU utilization and eliminating the data bottlenecks that traditionally slow AI workflows.
By re-engineering local NVMe storage on OCI GPU shapes into a persistent, high-speed shared storage layer, Hammerspace effectively removes data silos, allowing organizations to operate across unified, high-performance environments. This architecture empowers AI training, checkpointing, inference, and agentic AI workloads to run faster, more efficiently, and with lower latency.
“With Hammerspace on OCI, enterprises achieve the highest performance at scale, with unified data orchestration and global access optimized for AI workloads,”
said Tony Asaro, Chief Strategy and Business Development Officer, Hammerspace.
Recent benchmark testing on OCI has shown remarkable results, including:
- Up to 7x improvement in latency vs. traditional cloud file storage.
- Up to 6x improvement in storage performance.
- Checkpointing at extreme speeds, drastically reducing GPU idle time.
- Throughput fast enough to keep GPUs fed 24/7, ensuring no performance gaps.
- Policy-driven tiering for seamless migration of cold data to lower-cost storage.
“Our Tier 0 architecture ensures GPUs are never starved for data — it’s a game-changer for AI training, inference, and agentic workloads,”
added Raj Sharma, Cloud Field CTO, Hammerspace.
According to Asaro, customers are already seeing significant performance and scalability benefits. “The combination of Hammerspace and OCI provides a complete AI-ready solution — from small projects to massive, globally distributed environments,” he said.
Hammerspace will showcase this innovation live:
- Session: Increase Performance and Reduce Idle Time of Your GPU Workloads in OCI
- Presenter: Raj Sharma, Cloud Field CTO, Hammerspace
- Location: NVIDIA Booth #1013
- Date/Time: Tuesday, October 14, 5:30–6:00 p.m.
To explore the joint OCI–Hammerspace innovation, attendees can visit Hammerspace at Booth #8008 or access the OCI benchmark blog for detailed results.