GPU rental prices rising despite oversupply forecasts

GPU rental prices have risen sharply in 2026 across major Nvidia chip types, defying predictions of an oversupply-driven collapse. Using a16z data, Milk Road AI reported the H200 is up about 160% since the start of 2026, the B200 up approximately 120%, and the H100 up about 35%; only the A100 has remained roughly flat. FastGPU data tracking 81 provider instances over 60 days shows the Nvidia B300 up 52.8% from $8.50 to $12.99 per GPU hour, the H200 up 33.3%, the B200 up 27.8%, the H100 up 30%, and the RTX PRO 6000 up 307%. IREN co-CEO Dan Roberts said on CNBC that IREN's 2026 GPU capacity is largely sold out, with new three-year contracts priced at more than double last year's rates, and that the bottleneck is getting chips online fast enough, not finding buyers. Customers are prepaying 45 to 55% of GPU costs upfront. Nebius's CRO indicated the company could theoretically sell out its entire 2027 capacity immediately at current prices. Separately, Ramp's enterprise spending data shows that among the top 1% of AI-spending firms, more than a third of AI spend goes to raw GPU compute rather than LLM subscriptions, coding agents, or API tokens. On the supply side, Shadeform launched Flex VMs and Shadeform Spot, which convert bare-metal GPU servers into rentable cloud instances within hours using only an IP address and an SSH key, aiming to bring idle GPU capacity to market without API integration. AI infrastructure stocks including IREN, CRWV, and NBIS fell over 8% on oversupply and financing-cost fears during this period, even as rental rates rose.