Week 25: Price of NVIDIA's H200 compute at end of week?
Short Answer
The model assigns meaningfully higher odds than the market for the NVIDIA H200 compute price to be Above $2.50 at the end of Week 25, with 97.8% model probability vs 0.0% market probability. This divergence is driven by the U.S. approval of NVIDIA H200 chip sales to Chinese technology companies in June 2026, which is expected to significantly increase demand.
1. Executive Verdict
- U.S. approval of H200 chip sales to China notably increased demand expectations. Specific providers and hyperscalers offer H200 compute at higher market rates. Specialist providers often offer lower H200 on-demand rates than public clouds. NVIDIA's H2 2026 next-gen rollout is expected to impact H200 rental prices. * Upcoming catalysts for late June 2026 include the Vera Rubin platform ramp.
Who Wins and Why
| Outcome | Market | Model | Why |
|---|---|---|---|
| Above $6.50 | 1.0% | 1.5% | Major hyperscalers are noted to offer H200 compute at rates exceeding $10.00/hr. |
| Above $5.50 | 1.0% | 1.5% | Specific providers and major hyperscalers offer H200 compute rates around $6.31/hr and exceeding $10.00/hr. |
| Above $6.00 | 1.0% | 1.5% | Specific providers and major hyperscalers offer H200 compute rates around $6.31/hr and exceeding $10.00/hr. |
| Above $4.00 | 9.0% | 11.6% | U.S. approval of NVIDIA H200 chip sales to Chinese companies significantly increased potential demand. |
| Above $3.50 | 52.0% | 59.3% | U.S. approval of NVIDIA H200 chip sales to Chinese companies significantly increased potential demand. |
Current Context
H200 compute rental prices exhibit significant market fragmentation. As of June 18, 2026, on-demand hourly rates for NVIDIA H200 compute vary widely, ranging from approximately $1.40 to over $10.00 per GPU-hour. This pricing disparity is influenced by the specific provider, the commitment level from the renter, and the geographical region [^][^][^][^][^].
H200 rental rates recently experienced a notable downward trend. In Q2 2026, NVIDIA H200 rental rates dropped by about 40% leading up to June. This decrease is attributed to market supply normalization, primarily due to the ramp-up of the Blackwell series chips [^][^]. Prediction markets reflect ongoing uncertainty regarding where prices will settle by the end of the month. Contracts on Kalshi indicate a 99% probability that prices will remain above $2.19 per hour. Meanwhile, Polymarket estimates a roughly 28.5% probability of prices settling within the $3.00-$4.00 range [^][^].
2. Market Behavior & Price Dynamics
Historical Price (Probability)
Outcome probability
Date
This market's price action has been characterized by an extreme and sudden reversal. The contract began with a very low probability of a YES resolution, opening at 1.0% on June 13 and briefly falling to 0.0% on June 15. This suggests an initial belief that the resolution condition, likely tied to a price of $2.500 based on the market's ticker, was highly unlikely to be met. This sentiment then shifted dramatically on June 18, when the price surged from its low to 98.0%, establishing a powerful upward trend defined entirely by this single movement.
The primary driver for this price spike appears to be the pricing information detailed in the context. The initial low price may have been influenced by reports of a downward trend in H200 rental rates during Q2 2026. However, the repricing on June 18 coincides with reports from that same day showing that while fragmented, the market for H200 compute includes numerous providers with rates well above the apparent $2.500 threshold. The most critical factor in this analysis is the total trading volume of zero. The price changes do not reflect any trading activity or capital-weighted conviction from market participants. This means that concepts like support and resistance levels are not applicable, as no price points have been tested by actual trades. The current 98.0% probability reflects a sentiment that the outcome is almost certain based on available public data, but it is not a sentiment validated by market activity.
3. Significant Price Movements
Notable price changes detected in the chart, along with research into what caused each movement.
📈 June 17, 2026: 46.0pp spike
Price increased from 0.0% to 46.0%
Outcome: Above $3.50
What happened: The primary driver of the 46.0 percentage point spike was the U.S. approval of NVIDIA H200 chip sales to approximately 10 major Chinese technology companies in June 2026 [^]. This development indicated a significant increase in potential demand for H200 compute capacity, directly supporting higher prices for H200 cloud compute and making the "Above $3.50" outcome more probable [^]. While NVIDIA's stock declined on June 17, 2026, due to a $25 billion bond sale, the China sales approval was a distinct positive catalyst for compute demand [^][^][^][^]. Based on the provided information, social media activity was irrelevant.
4. Market Data
Contract Snapshot
The market resolves "Yes" if the value of NVIDIA H200 compute per hour is above $3.50 at 4 PM ET on June 19, 2026; otherwise, it resolves "No." Resolution is based on data reported by Ornn (dashboard.ornnai.com). If specific data is not published, the most recently available data will be used, and revisions made after expiration will not be accounted for.
Available Contracts
Market options and current pricing
| Outcome bucket | Yes (price) | No (price) | Last trade probability |
|---|---|---|---|
| Above $3.00 | $0.98 | $0.03 | 97% |
| Above $3.50 | $0.52 | $0.49 | 52% |
| Above $4.00 | $0.09 | $0.92 | 9% |
| Above $5.50 | $0.01 | $1.00 | 1% |
| Above $6.00 | $0.01 | $1.00 | 1% |
| Above $6.50 | $0.01 | $1.00 | 1% |
| Above $2.50 | $0.99 | $0.02 | 0% |
| Above $4.50 | $0.02 | $0.99 | 0% |
| Above $5.00 | $0.02 | $0.99 | 0% |
Market Discussion
NVIDIA H200 cloud rental prices have experienced significant volatility in Q2 2026, with on-demand rates ranging from $2.50 to $7.00 per GPU-hour following a reported ~40% decline in some segments throughout May [^][^][^][^]. Prediction market sentiment as of mid-June 2026 indicates skepticism that rental prices will remain in the lower $3.00–$4.00/hour band, with participants expecting rates to stabilize outside this range or fluctuate based on hyperscaler availability and Blackwell supply absorption [^][^][^][^]. The H200 market is characterized as transitioning from an acute shortage to a more segmented, quota-driven environment where hardware is differentiated by workload-specific software and provider infrastructure [^][^].
5. What effect could new U.S. export controls on AI hardware, potentially announced in mid-2026, have on the global availability and pricing of H200-level compute?
| H200 Export Tariff & Remittance | 25% remittance to U.S. government, 25% Section 232 tariff on foreign purchases [^][^][^] |
|---|---|
| Non-U.S. H200 Pricing Premium | Often 25% higher than U.S. equivalents [^][^][^] |
| H200 On-Demand Price Range | $2.40 to $4.00 per GPU-hour (mid-June 2026) [^][^] |
New U.S. policies in early 2026 significantly altered H200 export conditions. The U.S. administration implemented a policy permitting H200 exports to China under a revenue-sharing model, which required a 25% remittance to the U.S. government [^][^][^]. Concurrently, a 25% Section 232 tariff was applied to foreign purchases of H200 chips [^][^]. Further clarifications were issued on May 31, 2026, by the U.S. Bureau of Industry and Security (BIS), confirming that existing license requirements for advanced computing items, including H200s, apply to entities headquartered in Country Group D:5 or Macau, even if those entities are physically located outside these regions [^][^][^].
These controls resulted in higher global prices and persistent supply constraints. By mid-June 2026, global pricing for H200 compute exhibited a notable "tariff premium" in non-U.S. cloud regions, leading to prices often 25% higher compared to domestic U.S. equivalents [^][^][^]. This situation was compounded by persistent supply limitations, as the legal demand for H200 capacity from Chinese hyperscalers created competition with Western cloud providers [^][^][^]. Public on-demand pricing for H200 compute during the week of June 15–19, 2026, typically ranged from approximately $2.40 to $4.00 per GPU-hour, with variations depending on the provider and geographic region [^][^].
6. How do Wall Street analyst forecasts for NVIDIA's Q3 and Q4 2026 Data Center revenue reflect the price erosion of H200-era chips versus the adoption of the Blackwell platform?
| Q4 2026 Data Center Revenue Estimate | $58-60 billion (near) [^][^][^][^][^] |
|---|---|
| FY 2027 Annual Data Center Revenue Projection | $311 billion (around) [^][^][^][^][^] |
| Mid-2026 H200 Compute Price Stability | above $2.19/hour [^][^][^] |
Wall Street projects strong NVIDIA Data Center revenue driven by dual demand. NVIDIA's Data Center revenue for Q3 and Q4 2026 is projected by Wall Street analysts to be shaped by the initial adoption of the Blackwell platform alongside persistent demand for H200-era chips [^][^][^][^][^]. Data Center revenue is identified as the primary growth driver for the company, with Q4 2026 estimates hovering between $58 billion and $60 billion. Annual projections for fiscal year 2027 reach approximately $311 billion, despite some lingering uncertainty regarding Blackwell's specific financial contribution to these figures [^][^][^][^][^].
H200 chips maintain price stability amidst Blackwell's market entry. Despite the introduction of the Blackwell platform, H200 compute prices have demonstrated volatility but largely remain stable, primarily due to high demand for the H200 as a dependable 'workhorse' option [^][^][^][^][^][^][^][^]. The current scarcity and deployment complexities associated with Blackwell hardware contribute to the H200 maintaining its strong market position, sometimes even commanding a premium [^][^][^][^][^][^][^][^]. Prediction markets in June 2026 further indicate continued price stability for H200 compute, with significant odds favoring prices above $2.19 per hour [^][^][^].
Analysts' forecasts reflect Blackwell's potential alongside H200's resilient demand. Analysts' forecasts integrate the superior performance of the Blackwell platform, particularly its FP4 throughput, as it gradually becomes available in the market [^][^][^][^][^][^][^][^][^][^][^][^][^]. However, the ongoing challenges within Blackwell's supply chains to meet intense demand from hyperscalers and sovereign AI programs allow the H200 to preserve its market share and price stability [^][^][^][^][^][^][^][^][^][^][^][^][^]. This indicates that revenue projections during this transitional period carefully balance the anticipated growth from Blackwell with the robust, non-eroding demand and pricing for H200 chips [^][^][^][^][^][^][^][^][^][^][^][^][^].
7. How do the H200 on-demand rental rates from large-scale providers like CoreWeave and Lambda Labs compare to those from major public clouds like Azure in June 2026?
| Specialist Provider H200 Rate | Approximately $3.60-$3.80/hr (June 2026) [^] |
|---|---|
| Major Public Cloud H200 Rate | Approximately $4.20-$4.50/hr (June 2026) [^] |
| Major Public Cloud 8x H200 Node Rate | $40-$45/hr (June 2026) [^][^][^] |
Specialist providers offer lower H200 on-demand rates than public clouds. On-demand rental rates for a single NVIDIA H200 GPU are consistently lower at specialist providers compared to major public cloud hyperscalers in June 2026 [^]. Specialist providers such as Lambda Labs and CoreWeave generally offer H200s at typical hourly rates around $3.60-$3.80 [^]. Lambda Labs specifically provides competitive rates starting around $2.99-$3.60/hr for single-GPU instances, while CoreWeave offers single-GPU instances at approximately $6.31/hr in certain configurations [^][^][^].
Public clouds charge higher H200 rates, often in bundled nodes. In contrast, major public cloud hyperscalers, including Azure, typically charge approximately $4.20-$4.50/hr for a single H200 GPU [^]. These hyperscalers frequently bundle H200 capacity into larger nodes, such as 8x GPU nodes, leading to higher total hourly costs ranging from $40-$45 per node [^][^][^]. Hyperscalers also often prioritize long-term capacity reservations, which can make pure on-demand H200 access sporadic and relatively expensive [^][^][^].
H200 rental prices exhibit significant market fragmentation and volatility. Prediction markets for June 2026 reflect substantial fragmentation and volatility in H200 rental prices, tracking various outcomes from below $3 to above $7/hr [^][^][^][^]. This highlights the inherent challenge in determining a single market price for compute [^][^][^][^].
8. Which public data aggregators and market indices provide the most reliable daily or weekly tracking of H200 spot rental prices for June 2026?
| H200 Price Tracking Frequency | Daily or weekly (GPUS.io, ByteCosts, RentGPU, ComputePrices.com, MiningBoard) [^][^][^][^][^] |
|---|---|
| Specialized Indices for H200 Pricing | AxonIndex, GPU Markets, Gerra, AIMultiple's Cloud GPU Index [^][^][^][^][^] |
| Programmatic Access Sources for H200 Data | GPUs.io API, Compute Atlas, Hugging Face datasets (e.g., afhubbard/gpu-prices) [^][^][^] |
Several aggregators and indices track H200 spot rental prices. As of June 2026, public data aggregators such as GPUS.io, ByteCosts, RentGPU, ComputePrices.com, and MiningBoard offer daily or weekly tracking of H200 hourly spot rental prices [^][^][^][^][^]. These platforms obtain near-real-time updates by scraping provider APIs and public pricing surfaces. For a more normalized, index-level analysis of H200 spot and on-demand pricing, specialized indices like AxonIndex, GPU Markets (an open-source index), Gerra, and AIMultiple's Cloud GPU Index are available, applying methodologies to manage price dispersion across various providers [^][^][^][^][^].
Differentiating pricing types and programmatic access is key. For reliable H200 rental price tracking, it is crucial to distinguish between 'spot' (interruptible/peer-to-peer) and 'on-demand/reserved' (hyperscale/dedicated) pricing, as enterprise service levels for the latter often command significant premiums [^][^][^]. To enable programmatic access for constructing daily or weekly time-series models of H200 rental pricing, primary sources include APIs from GPUs.io, open-source projects such as Compute Atlas, and datasets found on platforms like Hugging Face, specifically afhubbard/gpu-prices [^][^][^].
9. How might NVIDIA's announced rollout schedule for its next-generation 'Blackwell' and 'Rubin' architectures in H2 2026 impact the residual demand and rental prices for H200 GPUs?
| H200 rental price drop | ~40% in late Q2 2026 [^][^] |
|---|---|
| Vera Rubin initial shipments | Q3 2026 [^] |
| H200 on-demand rental price range | $2.00/hr to $4.50/hr (late May/early June 2026) [^][^] |
NVIDIA's next-gen rollout significantly impacted H200 GPU rental prices. H200 rental prices experienced a substantial approximately 40% drop in late Q2 2026 [^][^]. This decline was attributed to market supply normalization and a demand shift towards Blackwell-based systems, influenced by NVIDIA's announced H2 2026 schedule for its next-generation Blackwell and Vera Rubin architectures [^][^].
Next-generation architecture launches are balancing the H200 market supply. The Vera Rubin architecture is slated for initial shipments in Q3 2026, with a volume ramp planned for Q4 2026 [^][^][^]. This rollout adds to the ongoing and historically fast deployment of the Blackwell architecture [^][^][^]. This influx of next-generation hardware coincided with the H200 market's transition from acute scarcity to a more balanced phase around late May/early June 2026 [^][^]. During this period, on-demand H200 rental prices from major providers generally ranged from $2.00/hr to $4.50/hr, though hyperscaler rates remained potentially higher [^][^]. Prediction markets for June 2026 H200 rental prices indicated high volatility, with significant activity reflecting expectations that prices might settle near or slightly above the $3.00-$4.00 band, driven by persistent enterprise inference demand [^].
10. What Could Change the Odds
Key Catalysts
Key Dates & Catalysts
- Strike Date: June 19, 2026
- Expiration: June 26, 2026
- Closes: June 19, 2026
11. Decision-Flipping Events
- Trigger: Key bullish catalysts for NVIDIA in late June 2026 include the Vera Rubin platform production ramp, expansion into the PC CPU/GPU market with RTX Spark, and the Annual Meeting of Stockholders on June 24, 2026 [^] [^] [^] [^] .
- Trigger: Conversely, bearish catalysts and risks include the company's high beta of 2.20, uncertainty surrounding U.S.-China export controls, and potential macroeconomic volatility following the FOMC meeting on June 16-17 [^] [^] [^] [^] .
13. Historical Resolutions
No historical resolution data available for this series.
