WEKA and Oracle have jointly validated 10x AI inference performance gains for long-context workloads running on Oracle Cloud Infrastructure (OCI), demonstrating that purpose-built storage and cloud infrastructure working together can dramatically exceed the performance of conventional approaches. The benchmark results, focused on WEKA’s Augmented Memory Grid technology running on OCI, were validated by both companies as representative of production-scale deployments.
The 10x improvement applies specifically to long-context inference workloads — AI tasks that require maintaining very large context windows across multiple conversational turns or analytical queries. This category of workload includes agentic AI applications, multi-turn code assistance, complex document analysis, and extended reasoning tasks. As AI models grow in capability and enterprises deploy them for increasingly complex tasks, long-context inference is becoming a dominant workload class that conventional storage architectures struggle to serve efficiently.
WEKA’s Augmented Memory Grid on NeuralMesh, first announced for commercial availability at SC25 in November 2025 with OCI as the exclusive launch partner, works by treating distributed storage as an extension of GPU memory. This architectural innovation allows AI inference engines to access much larger effective context windows than would be possible with GPU memory alone, without the performance penalties that arise from conventional paging or caching approaches.
The validation results are particularly significant for enterprise customers running agentic AI applications — systems where AI agents maintain persistent memory across long interaction sessions and need to retrieve and reason over large bodies of historical context. For these applications, the latency and throughput characteristics of the underlying storage infrastructure directly impact the quality and speed of AI responses.
OCI’s partnership with WEKA reflects Oracle’s broader strategy of building cloud infrastructure purpose-optimized for AI workloads. Oracle has made significant investments in GPU cluster infrastructure, high-performance networking, and storage systems specifically designed for the demands of training and inference at scale. The WEKA validation adds another dimension to OCI’s AI infrastructure differentiation, particularly for customers with demanding long-context inference requirements.
The 10x performance improvement benchmark demonstrates that AI infrastructure optimization isn’t just about GPU hardware — the entire stack from storage to networking to cloud orchestration contributes to total system performance. As enterprises invest in AI at scale, the choice of cloud infrastructure and storage architecture will increasingly determine whether their AI deployments can deliver the responsiveness and reliability that business applications demand.
