NVIDIA GPU Rental Prices Surge Past $3 for H100, $5.80 for B200

GPU rental rates for NVIDIA A100, H100, H200, and B200 chips are climbing steadily, contradicting the idea that AI infrastructure is already overbuilt. Buyers and operators should note that compute costs are rising rather than falling as new capacity comes online. The upward movement in rental rates indicates that demand for high-end AI accelerators continues to exceed available supply.

Rising cloud costs for AI accelerators contradict overbuild claims

Data center operators are facing higher hourly costs for these specific GPU generations, which impacts their overall cloud pricing strategies. The H100, a workhorse for large language model training, has seen its rental price jump from just under $2 per hour in January 2026 to nearly $3 per hour. The newer B200, designed for even heavier workloads, has moved from below $5 per hour to a range of $5.50 to $5.80 per hour.

Rising rental prices serve as a direct rebuttal to claims that the industry is overbuilding AI compute resources. Ricky Ho of Silicon Data noted that GPU rental prices are not collapsing as new capacity comes online. He stated, "The chart is one of the clearest rebuttals to the claim that AI compute is already being overbuilt. GPU rental prices are not collapsing as new capacity comes online. They are rising across almost every generation. B200 rates have moved toward roughly $5.50 to $5.80 per…" This data from Silicon Data, Evergreen Capital, and P Equity Research highlights the tightening supply dynamics.

Microsoft is adjusting its financial approach to these assets by extending the estimated useful life of its data centers from 15 to 25 years starting in fiscal year 2027. Amy Hood explained, "effective at the start of FY 27, we are extending the estimated useful life of our data centers and office buildings from 15 to 25 years." Microsoft's decision to extend the estimated useful life of its data centers from 15 to 25 years may reduce annual depreciation expenses, potentially offsetting some of the rising rental or acquisition costs.

We looked at NVIDIA A100, H100, H200, B200 closely — see our earlier coverage on xAI Reports Only 11% GPU Usage. That report highlighted low utilization rates on massive H100 and H200 deployments, which contrasts with the current pricing strength. Market dynamics reflect a balance between high utilization demands for specific workloads and periods of lower usage across broader clusters.

The confirmed facts show a clear upward trajectory in GPU rental costs for NVIDIA's major AI accelerators. H100 rates approach $3 per hour while B200 rates hover near $5.80 per hour. These figures, combined with Microsoft's extended depreciation timeline, define the current economic landscape for AI hardware operators.