As AI adoption expands beyond specialized HPC teams into mainstream enterprises, the definition of “best-in-class” is shifting from infrastructure to how data is accessed, governed, and used.
Leading platforms must:
- Work across heterogeneous infrastructure
- Deliver extreme performance without specialized complexity
- Automate data movement, placement, and lifecycle services
- Avoid creating new silos or operational debt
- Integrate directly with AI pipelines, agents, and inference systems
Hammerspace meets these requirements by design. Unified access alone isn’t enough; the platform also embeds automation, policy-driven orchestration, and performance-aware intelligence directly into the data plane, not bolted on as separate systems. This enables enterprises to gain AI-ready infrastructure without sacrificing manageability or governance.
In addition to Tier 0 enhancement, with v5.2, Hammerspace also delivered broader security controls, including Kerberos authentication and labeled NFS support that enables consistent enforcement of security labels across NFS. Furthermore, the release expanded cloud integration including deep support for Oracle Cloud Infrastructure while maintaining seamless hybrid operations. This is in addition to existing tight integration with AWS, Azure, and GCP environments.
Looking Forward: The Platform for AI Anywhere
Inference and agentic AI fundamentally changes what a data platform must do. As AI moves beyond discrete training and runs into continuous, always-on workflows, data platforms must evolve to support persistent, low-latency access to distributed data while dynamically adapting to changing models, pipelines, and compute locations across GPU-accelerated environments.
This is realized in the Hammerspace AI Data Platform (AIDP), which extends the data layer into AI workflows themselves. AIDP enables data to be discovered, curated, vectorized, and published directly to AI pipelines and agents through MCP endpoints—without moving or duplicating the data.
This integration with NVIDIA-accelerated compute infrastructure and enterprise software extends the Hammerspace data plane with AI-aware orchestration and rich metadata intelligence, enabling enterprises to discover, prepare, and continuously feed the right data to GPU-driven AI pipelines (training, inference, and agentic workflows) without reorganizing data or creating new silos.
The “why now” is clear: as AI becomes operational across the enterprise, success increasingly depends on making data access continuous, intelligent, and automated. AI is not a series of isolated jobs. It is continuous interaction with data.
Crucially, the Hammerspace AI data platform solution leverages Hammerspace performance innovations like Tier 0 activation and affinitization. Now, decisions around which data is needed and where it should flow are automated, while Tier 0 affinitization ensures that once data reaches the execution environment, it is placed as close as possible to the GPUs and compute nodes consuming it. This unifies intelligence and performance within a single platform, automatically aligning data discovery, orchestration, and placement without user intervention. This brings together data discovery, preparation, and execution into a single system.
As AI becomes a core enterprise workload rather than an experimental initiative, this data-centric approach of combining open standards, intelligent orchestration, and topology-aware performance will be essential for sustainable scale, operational simplicity, and long-term flexibility.
Hammerspace was designed to support this evolution. Its open, standards-based architecture allows enterprises to continuously adapt by activating data wherever it lives and feeding compute resources wherever they are, and does so without introducing new vendor-locked silos or fragmenting metadata across systems.
Proven Impact at Enterprise and Hyperscale
Hammerspace’s data platform innovation is validated by production deployments at extreme scale.
Meta uses Hammerspace to power Llama 2, 3, and 4 workloads across thousands of storage servers and GPU servers in its AI Research Super Cluster. As demand grew beyond on-prem facilities, Hammerspace enabled seamless extension of these workloads to cloud-based GPUs in AWS and OCI, without data migrations or workflow changes for data scientists.
To applications and researchers the environment continued to look like a standard NFS file system with Hammerspace orchestrating data in the background via its S3 connector technology. This powerful feature allows NFS-based workloads to access data in cloud storage through S3 with no gateways, data copies or caches required. Users don’t need to know or care about this, since their workloads run the same whether on-premises, in the cloud, or a hybrid of both.
Additional customers demonstrate breadth. These examples illustrate a consistent pattern: activating existing data without creating new silos:
- A global game development and interactive entertainment company streamlined collaboration with a single unified namespace that spans seven data centers across two continents
- Netflix Animation accelerated global rendering workflows across Dell Isilon environments using pNFS v4.2
- Mitsubishi Electric Research Laboratories doubled performance to NVIDIA DGXTM B200 systems while unifying legacy storage platforms
- Vanderbilt University slashed storage costs for its ACCRE compute cluster—activating Tier 0 capacity and unifying incompatible storage silos to reduce overall storage spend by 48%
- Texas Tech HPCC enabled a high-performance global file system that spans HPC workloads running pNFS v4.2 between multiple clusters and ultimately three wind-powered data centers.
- A global digital payments and financial technology company improved productivity for 3,000 AI researchers and data scientists with a 4 PB global file system that spans on-prem and Google Cloud, gained hybrid-cloud agility, and reduced storage costs by $3-5M by moving to commodity object storage for capacity storage
- BUCK, a leading creative agency, enabled global collaboration at scale by creating a single global namespace that spans four sites across three continents with the performance to handle all of their creative media workflows.
- VHB, an engineering and design firm, eliminated manual data copying and improved productivity for their engineering and IT teams with a hybrid-cloud global file system spanning primary and DR data centers and a Microsoft Azure cloud region
Benchmarks reinforce these outcomes. At Supercomputing 2025, Hammerspace announced top-20 results in the IO500 10-Node Production list using standard NFS—an industry first—delivering performance historically associated only with complex proprietary HPC file systems. Earlier MLPerf benchmarks showed similar results, with Hammerspace achieving HPC-class performance using roughly half the servers and network ports of traditional scale-out NAS architectures.
Ecosystem Role: Platform Innovator That Enables and Consolidates
Hammerspace is unique in creating a data platform that redefines the traditional relationship between data and infrastructure, and in the process solves a critical problem. AI exposes the limitations of infrastructure-first approaches. Classic enterprise IT is organized hierarchically, with new data coming in at the top and aging down to lower tiers until they are relegated to an archive. AI flips this model on its head, since inference and agentic AI need immediate access to any data on any storage at any time, and at HPC-class performance levels and scale not usually seen in the enterprise. In doing so, Hammerspace also disrupts legacy siloed models, enables existing investments to deliver new value, and consolidates fragmented data environments into a coherent, intelligent whole.
The future of AI will not be built on new storage systems, but on how effectively data can be accessed, governed, and activated. By aligning open standards, extreme performance, and automation into a single data platform, Hammerspace is reshaping what enterprise data infrastructure looks like in the AI era.
Learn More
- Blog Post: Guest Post: Building Meta’s GenAI Infrastructure
- Data Unchained Podcast: Hammerspace Breaks IO500 Barriers: How They Built the Fastest NFS-Based Benchmark Ever w/ Jon Flynn
- Blog Post: Get AI Ready by Putting Data First: Why You Don’t Need Another Storage Silo for AI
- White Paper: Hammerspace Data Platform
