**Strong AI inference demand combined with persistent supply constraints in high-bandwidth memory (HBM) and advanced packaging (CoWoS) has driven recent firmness in H100 rental prices.** As of September 2026, median on-demand rates sit near $3.35 per GPU-hour across tracked providers, up about 7% over the past 90 days and 12% year-over-year in some indices, after bottoming near $1.70 for one-year contracts in late 2025. Spot and marketplace rates range from roughly $0.35–$2.00, while neoclouds like Lium, RunPod, and Vast.ai offer on-demand from $1.25–$2.50; hyperscalers (AWS, Azure, Google Cloud) command $6–$12+ premiums. Traders note divergence between falling spot volatility and rising committed-contract pricing, signaling production workloads rather than experimentation. Providers such as Nebius have announced 17–21% on-demand hikes effective October 2026, reflecting sold-out capacity and extended customer visibility into 2028. Blackwell ramp-up could eventually displace some H100 demand, but near-term bottlenecks in power, cooling, and memory allocation continue to support elevated rates. Key watchpoints include hyperscaler capacity releases, additional neocloud expansions, and any acceleration in HBM production.
基於Polymarket數據的AI實驗性摘要。這不是交易建議,也不影響該市場的結算方式。 · 更新於$10,275 交易量
↑ $3.50
78%
↓ $2.25
45%
↓ $2.00
45%
↓ 1.50美元
33%
$10,275 交易量
↑ $3.50
78%
↓ $2.25
45%
↓ $2.00
45%
↓ 1.50美元
33%
The resolution source for this market is Ornnai.com (ornnai.com), specifically, the H100 Index chart data available at https://dashboard.ornnai.com. The daily values shown on the chart will be used for resolution. Daily data will be considered finalized once the following day’s data point is published.
This market will resolve as soon as the listed price is hit, or once the value for the specified end date is finalized. If not all relevant data has been finalized by the end of the 14th calendar date afterward (ET), this market will resolve based on the data available at that time.
Revisions made to previously published data points, before all relevant data points have been finalized, will be considered; however, they will not disqualify a previously published data point from resolving this market. Revisions made after all relevant data points have been finalized will not be considered.
市場開放時間: Jul 10, 2026, 10:21 AM ET
The resolution source for this market is Ornnai.com (ornnai.com), specifically, the H100 Index chart data available at https://dashboard.ornnai.com. The daily values shown on the chart will be used for resolution. Daily data will be considered finalized once the following day’s data point is published.
This market will resolve as soon as the listed price is hit, or once the value for the specified end date is finalized. If not all relevant data has been finalized by the end of the 14th calendar date afterward (ET), this market will resolve based on the data available at that time.
Revisions made to previously published data points, before all relevant data points have been finalized, will be considered; however, they will not disqualify a previously published data point from resolving this market. Revisions made after all relevant data points have been finalized will not be considered.
**Strong AI inference demand combined with persistent supply constraints in high-bandwidth memory (HBM) and advanced packaging (CoWoS) has driven recent firmness in H100 rental prices.** As of September 2026, median on-demand rates sit near $3.35 per GPU-hour across tracked providers, up about 7% over the past 90 days and 12% year-over-year in some indices, after bottoming near $1.70 for one-year contracts in late 2025. Spot and marketplace rates range from roughly $0.35–$2.00, while neoclouds like Lium, RunPod, and Vast.ai offer on-demand from $1.25–$2.50; hyperscalers (AWS, Azure, Google Cloud) command $6–$12+ premiums. Traders note divergence between falling spot volatility and rising committed-contract pricing, signaling production workloads rather than experimentation. Providers such as Nebius have announced 17–21% on-demand hikes effective October 2026, reflecting sold-out capacity and extended customer visibility into 2028. Blackwell ramp-up could eventually displace some H100 demand, but near-term bottlenecks in power, cooling, and memory allocation continue to support elevated rates. Key watchpoints include hyperscaler capacity releases, additional neocloud expansions, and any acceleration in HBM production.
基於Polymarket數據的AI實驗性摘要。這不是交易建議,也不影響該市場的結算方式。 · 更新於

警惕外部連結哦。
警惕外部連結哦。
Frequently Asked Questions