Memory chip shortage makes bare metal cloud cheaper than on prem, says Nutanix' Rajiv Ramaswami - Data Center Dynamics

The ongoing memory chip shortage is driving enterprises to reconsider their infrastructure strategies, with cloud-based bare metal servers now offering better value than on-premises deployments. Nutan... Read original →

The memory chip shortage is fundamentally reshaping AI infrastructure deployment logic. Technically, tight DRAM and HBM supply not only inflates server costs but forces enterprises to re-evaluate the TCO boundary between bare metal cloud and on-prem—hyperscalers leverage bulk procurement and inventory buffers to secure structural advantages in delivery lead times and per-watt cost. On compliance, despite tightening data localization rules in the West, sovereign AI stacks promoted by Huawei cannot yet offset operational risks from supply fragility. Strategically, Samsung and SK Hynix will likely prioritize North American hyperscaler orders, further constraining Chinese firms' access to high-end memory; Alibaba Cloud, even with doubled server costs, struggles to pass on pricing pressure—highlighting vulnerability under geopolitical decoupling. Over the next 12–24 months, hybrid deployment will accelerate: non-sensitive workloads shift to bare metal cloud, while core AI training pushes domestic substitution, though HBM3E+ capacity bottlenecks—and limited advanced packaging output from Taiwan, China and mainland China—will delay this transition.