SK Hynix and SanDisk's High Bandwidth Flash Standard Arrives as Zimbabwe Bans Raw Lithium Exports

The Day's Thesis

Today crystallized a fundamental tension in the AI infrastructure buildout: while memory architects engineer solutions for tomorrow's inference bottlenecks, resource nationalism is reshaping the critical mineral supply chains that underpin the entire technology stack. The convergence signals a new phase where technical innovation must navigate increasingly fragmented commodity markets.

AI & Research Frontier

MIT researchers delivered a breakthrough in LLM training efficiency that could reshape compute economics — their new method doubles training speed by utilizing idle computing resources while maintaining model accuracy. This development arrives as NVIDIA's Jensen Huang publicly pushed back against market concerns that AI threatens software companies, arguing instead that AI will enhance traditional software businesses.

The efficiency gains from MIT's approach could significantly reduce demand for GPU computing capacity, potentially easing the supply constraints Huang warned would persist for "a couple of quarters." Meanwhile, Nokia and AWS are piloting AI automation for real-time 5G network slicing, with AI agents managing traffic and service quality autonomously — a development that could accelerate the edge computing buildout essential for distributed AI workloads. The cellular AI research emerging from MIT promises to revolutionize drug discovery by providing holistic cellular information, creating new compute-intensive workloads that will demand the very infrastructure improvements these other breakthroughs enable.

Technology & Infrastructure

SK Hynix and SanDisk's joint announcement of the High Bandwidth Flash (HBF) standard represents a critical infrastructure evolution, specifically targeting AI inference servers where storage performance has become the new bottleneck. This memory technology addresses the growing mismatch between compute capacity and data access speeds in AI workloads. The timing is particularly significant as NVIDIA warns of constrained gaming GPU supply for the next two quarters, potentially creating shortages that extend beyond consumer markets into data center expansion plans.