Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore this 44-minute conference presentation examining the critical intersection of storage technology and AI infrastructure demands projected for 2026. Learn how massive AI deployments requiring one gigawatt of power will support 550,000 NVIDIA Grace Blackwell GB300 GPUs and 25 exabytes of storage, necessitating extremely efficient, high-capacity solid-state drives like Solidigm's 122 terabyte drives to operate within power constraints. Discover the revolutionary impact of NVIDIA's Vera Rubin platform with Bluefield 4 DPUs, which introduces a new "inference context memory storage platform" (ICMSP) layer that fundamentally transforms AI storage architecture by creating a critical intermediate tier between direct-attached storage and object storage for rapid KV cache data access. Understand how this architectural evolution redistributes the 25 exabytes across high-capacity network-attached storage, 6.4 exabytes of context memory storage, and 6.1 exabytes of direct-attach storage, driving a projected 5x or greater compound annual growth rate in high-capacity storage demand. Examine Vast Data's disaggregated shared-everything (DASE) architecture that separates compute from storage while utilizing Solidigm SSDs for dense capacity, achieving 3-4x more effective capacity through global data reduction techniques including compression, deduplication, and similarity-based reduction. Learn how Vast's C-node storage logic can be deployed directly on Bluefield 4 DPUs to create highly optimized ICMSP solutions that accelerate time to first token, boost GPU efficiency by offloading context computation, and dramatically reduce power consumption by eliminating intermediate compute layers, enabling AI inference workloads to operate at unprecedented speed and scale with shared, globally accessible context.
