If you've overlooked data storage, you're not taking AI seriously (www.businessinsider.com)

AI workloads are forcing a rethink of data storage: Solidigm argues that SSDs are no longer a niche performance upgrade but a foundational requirement for modern AI infrastructure. With GPUs hungry for ever-larger, lower-latency data streams, HDDs — still used for nearly 90% of data-center capacity and costing roughly $0.011/GB — struggle to meet the throughput and predictable latency AI demands. Solidigm’s white paper claims SSDs can deliver lower total cost of ownership over a 10-year, exabyte-scale deployment, and the company demonstrated a liquid-cooled, direct-to-chip, hot-pluggable enterprise SSD built with NVIDIA at GTC 2025. Technically, SSDs provide parallel read/write, deterministic latency and much higher density (Solidigm ships 122 TB SSDs versus current HDD tops near 30 TB), plus substantial operational gains: up to 77% power savings and 90% less rack space when replacing HDDs. That enables real-time, data-intense workflows—like Los Alamos’ seismic simulations—that HDD seek-and-spin mechanics can’t sustain. While HDDs and tape will remain cost-effective for cold archives (e.g., CERN), the shift toward always-on, high-throughput AI pipelines makes investing in SSD-focused architectures, cooling/form-factor innovations, and reallocated power budgets essential for maximizing GPU utilization and scaling AI systems.