Nvidia H200 GPU rental rates fall 40 percent in three weeks, raising questions over AI chip scarcity - 디지털투데이

Nvidia's H200 GPU rental rates dropped by nearly 40% over three weeks, sparking speculation that years-long AI chip shortages may be easing. The decline is attributed to shifting demand toward the new... Read original →

The 40% plunge in H200 rental rates within three weeks signals not oversupply but a pivotal shift from speculative procurement to rational AI infrastructure deployment. Technologically, Blackwell’s 3nm EUV-based architecture and fifth-gen Tensor Cores are rapidly displacing Hopper, forcing cloud providers to reassess legacy GPU ROI. On compliance, tightening U.S. export controls on advanced AI chips compel non-U.S. clients to build Nvidia-alternative stacks, inflating supply chain redundancy costs. Competitively, AMD’s MI300X ramp-up and Huawei’s Ascend 910B penetration in Greater China are eroding Nvidia’s mid-tier training dominance. Over the next 12–24 months, GPU pricing will bifurcate: Blackwell commands premium rates while Hopper/H200 slide into edge inference roles. The real threat isn’t easing scarcity—it’s whether AI capex at Microsoft or Meta can generate positive IRR. Persistent negative returns would trigger structural spending cuts, undermining the entire AI hardware investment thesis.