# Weka's new storage platform reduces AI model load Caching 100% of pre-calculated tokens to optimize GPU usage. - **Source:** [VentureBeat](https://venturebeat.com) - **Published:** 2026-07-21 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/d1b6da06-6ce4-4609-99bf-86a6fdf9ac5e ## GPU Memory Challenge AI models constantly recalculate the same information in long context windows and multi-turn conversations, rapidly depleting expensive GPU memory. Weka addresses this with flash storage technology that extends GPU memory capacity at lower cost. ## NeuralMesh 6 Platform Weka's new NeuralMesh 6 software and Wekapod 3 hardware introduce Augmented Memory Grid technology. This system leverages NAND flash storage to function like GPU memory at a fraction of the cost, specifically designed for large-scale AI deployments. ## Multi-Tenancy Support NeuralMesh 6 offers composable and virtual multi-tenancy supporting up to 50,000 tenants. With hardware-level isolation and provisioning under 30 minutes, it enables enterprises to flexibly manage AI infrastructure at scale. ## Unified Storage System Weka's platform unifies file and object storage in a single system. Unlike traditional solutions, it allows reading the same physical data through both paths without duplication, significantly improving storage efficiency. ## Context Window Optimization Augmented Memory Grid caches 100% of pre-calculated tokens, eliminating recomputation overhead. This dramatically optimizes GPU usage, especially for multi-turn interactions like chat and coding assistants. ## Competitive Positioning Unlike traditional storage vendors like Dell and NetApp, Weka offers a native AI solution from the ground up. Analysts highlight its unique value proposition for organizations with rapidly growing AI workloads requiring efficient GPU utilization. ## Sources covering this story - https://venturebeat.com/data/stop-adding-more-gpus-wekas-new-storage-platform-reduces-load-by-caching-100-of-ai-models-pre-calculated-tokens — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._