Insight: The Critical Role of High-Bandwidth Memory in Next-Gen AI Datacentre Networking

AI‑driven memory demand outstrips supply as HBM diverts capacity, tightening DRAM and NAND and raising lead times, costs, and risk across data centers.

The memory market has reached a structural inflection point where AI demand outpaces supply. Hyperscalers have locked in long-term DRAM commitments, HBM consumes roughly 3x to 4x as many wafers per bit as DDR5, and NAND suppliers are shipping virtually everything they produce — yet still cannot meet demand. The rapid expansion of AI datacentre and infrastructure is exerting significant pressure on the memory ecosystem. Project timelines can be delayed when memory becomes the primary component for server builds, expansions, or upgrades. Capacity planning becomes more uncertain, with longer lead times increasing the risk of under-provisioning. Memory unavailability can also reduce architectural flexibility for scaling RAM-intensive workloads, including databases, analytics, and AI inference. This insight report assesses the supply-side diversion favoring HBM and how it constricts the networking ecosystem, alongside advice for mitigating the resulting lead-time and cost volatility.