September 2026: Monthly average compute price of NVIDIA's A100
$0.00
1
7
September 2026: Monthly average compute price of NVIDIA's A100
$0.00
1
7
AI Analysis
Trader mode: Actionable analysis for identifying opportunities and edge
About This Event
September 2026 If the average value of NVIDIA A100 compute per hour is above X in September 2026, then the market resolves to Yes. The market resolves based on the average value of NVIDIA A100 compute per hour in September 2026, calculated as the arithmetic mean of hourly values reported by Ornn. Revisions to the underlying made after expiration will not be accounted for. The "USD" iteration of the index will be used unless explicitly stated otherwise. Values are rounded to two decimal places u
What Prediction Markets Are Forecasting
Traders on Kalshi currently give roughly a 56% chance that the average hourly compute price for an NVIDIA RTX 5090 will exceed $0.50 in September 2026. That's barely better than a coin flip. The market isn't confident either way, which tells you something interesting: the future of GPU rental pricing is genuinely uncertain, even to people who watch this space closely.
To put it in perspective, $0.50 per hour for a top-tier consumer GPU is neither cheap nor expensive. It's a middle-ground bet. If you're renting compute from a cloud provider, you'd expect flagship cards to cost more, but prices have been sliding for years as supply catches up with demand.
Why the Market Sees It This Way
The RTX 5090 launched in early 2025 as NVIDIA's most powerful consumer card, with 32GB of VRAM and serious compute muscle. But the market for renting GPU hours has changed a lot since the AI boom peaked. A few forces are pulling in opposite directions.
First, supply has grown. Cloud providers and small GPU rental startups have been adding capacity aggressively. More cards available means downward pressure on prices. Second, demand keeps shifting. AI training workloads are moving to data center cards like the H100, while consumer cards like the 5090 are often used for smaller fine-tuning jobs, inference, and hobbyist projects. That demand is steadier but less explosive.
Third, there's the question of depreciation. The 5090 is a consumer product, and its resale value drops over time. Rental prices usually follow that curve. By September 2026, the card will be nearly two years old, and newer hardware might be on the horizon. That pushes prices down.
But there's a counterforce. Electricity costs, data center overhead, and the possibility of another AI demand spike could keep prices higher than historical patterns suggest. The market's 56% odds reflect this tension: enough bullish factors to keep it close, but no clear winner.
Key Dates and Events to Watch
The biggest signal will be NVIDIA's next product cycle. If the RTX 6090 or a similar card launches before late 2026, older cards like the 5090 could see rental prices drop sharply. Watch for announcements at major tech conferences in early 2026.
Also keep an eye on cloud providers' pricing pages. If major players like AWS, Azure, or smaller GPU rental services cut their rates for 5090 instances, that's a direct signal. And watch for any big AI funding rounds or data center buildouts that suggest demand is surging again.
How Reliable Are These Predictions?
Prediction markets have a decent track record on tech pricing questions, but this one is tricky. The market is pricing a near-coin flip, which honestly reflects the genuine uncertainty here. GPU rental prices depend on supply chains, energy costs, AI hype cycles, and competitive dynamics that are hard to model.
One limitation: Kalshi traders might not be deeply plugged into the GPU rental market specifically. The market is relatively niche, and liquidity could be thin. That said, the 56% number is a reasonable starting point, not a crystal ball. If you're making decisions based on this, treat it as a mild lean toward higher prices, not a strong signal.
Current Market Outlook
Kalshi traders are pricing an 85% probability that NVIDIA's H100 cloud compute will average above $2.00 per hour in September 2026. That's a strong bet on sustained pricing power. For context, H100 instances on major cloud providers currently range between $1.50 and $3.00 per hour depending on contract terms and provider. The $2.00 threshold sits near the middle of that range. An 85% probability means the market sees this as a likely outcome but not a foregone conclusion. The implied 15% chance of a sub-$2.00 price reflects real downside risk.
Key Factors Driving the Odds
The H100 remains the dominant GPU for AI training and inference through 2025. Supply constraints have kept prices elevated since the chip launched in late 2022. Even with AMD's MI300X and Intel's Gaudi 3 entering the market, NVIDIA's CUDA ecosystem lock-in gives the H100 pricing power that competitors haven't cracked.
Cloud providers like AWS, Azure, and Google Cloud have locked in multi-year H100 reservations at premium rates. Those contracts won't expire until late 2026 or 2027. That base of committed demand supports the floor price.
The wildcard is NVIDIA's own Blackwell architecture, expected to ramp production in late 2025. If Blackwell delivers the 2x-3x performance gains NVIDIA claims, older H100 capacity could flood the spot market as customers upgrade. But Blackwell's launch timeline keeps slipping. If it arrives late, H100 scarcity persists.
What Could Change These Odds
The September 2026 date is key. By then, Blackwell will have been in production for at least 12 months. If NVIDIA ships Blackwell on time and at scale, H100 prices could drop below $2.00 as cloud providers discount older hardware. That scenario is the main risk to the current 85% probability.
The other risk is demand destruction. If AI model training costs collapse due to more efficient architectures like Mamba or state-space models, or if a major recession hits enterprise cloud spending, H100 utilization could crater. Neither looks likely right now, but both are real tail risks.
Watch for NVIDIA's Q2 2026 earnings call in August 2026. Any guidance on H100 inventory clearance or Blackwell pricing will move this market sharply.
AI-generated analysis based on market data. Not financial advice.
Overview
This prediction market concerns the average hourly cost to rent an NVIDIA H100 GPU in September 2026. The H100 is NVIDIA's flagship data center GPU, released in 2022, and is the dominant processor for training and running large AI models such as GPT-4, Claude, and Llama. Compute time on these GPUs is typically rented by the hour through cloud providers like AWS, Google Cloud, Azure, and specialized AI cloud companies like CoreWeave and Lambda Labs. The market uses the Ornn index, which tracks the average price of H100 compute across multiple providers in USD. The question is whether that average price will be above a specific threshold in September 2026. This is a bet on the balance of supply and demand for AI compute two years out, factoring in new GPU releases, data center buildouts, and changes in AI model efficiency. The H100's spot price on cloud markets has fluctuated significantly since its launch, from highs of over $4 per hour during the 2023 GPU shortage to around $2 per hour in early 2025 as more supply came online. The outcome depends on whether NVIDIA's next-generation Blackwell GPUs (B200) cannibalize H100 demand, how many new data centers come online, and whether AI model training demand continues its exponential growth. This market is relevant to anyone tracking the economics of AI infrastructure, cloud pricing trends, or NVIDIA's business trajectory.
Historical Context
The H100 was announced in March 2022 and began shipping in late 2022. Initial demand far exceeded supply, driven by the launch of ChatGPT in November 2022 and the subsequent AI boom. Prices on the secondary market and from cloud providers spiked to over $4 per hour for an 8-GPU node in mid-2023, with some reports of spot prices exceeding $10. NVIDIA responded by ramping production and allocating more wafers to H100, but supply constraints persisted through 2023. By early 2024, delivery lead times had shortened from 36-52 weeks to 8-12 weeks. Cloud providers began adding capacity, and prices started to fall. In mid-2024, the average H100 compute price on Ornn was around $2.50 per hour. The introduction of the H200 (a memory-upgraded H100 variant) in late 2024 created a two-tier market, with H100 prices dropping further as some customers migrated to H200. By early 2025, H100 prices had stabilized around $2 per hour. The key historical precedent for predicting 2026 prices is the A100's pricing trajectory after the H100 launched. The A100, released in 2020, saw its cloud price drop by about 30% in the two years after the H100 launched, as older hardware was discounted and supply increased. If the pattern repeats, H100 prices could fall to $1.50 per hour or lower by September 2026. However, the A100 did not face a new generation of AI models that required more compute, while the H100 might see sustained demand from inference workloads even if training shifts to Blackwell.
Why It Matters
The price of H100 compute is a direct measure of the cost of doing AI research and development. For startups and academic labs, a lower price means more experiments, faster iteration, and lower barriers to entry. For NVIDIA, the price trajectory signals whether the company can maintain its high margins (over 70% gross margin in 2024) as competition heats up from AMD, Intel, and custom chips from Google and Amazon. A drop in H100 prices below $1.50 per hour could indicate oversupply and pressure NVIDIA's revenue. Conversely, prices above $2.50 might signal that demand is outstripping supply, benefiting NVIDIA but raising costs for AI companies. The broader economic implication is that AI compute costs influence the viability of AI startups, the pace of AI adoption in industries like healthcare and finance, and the energy consumption of data centers. Governments are also paying attention: the US Department of Commerce has restricted H100 exports to China, and the EU is considering regulations on AI compute allocation. The outcome of this market could inform investment decisions in data center REITs, cloud stocks, and AI companies.
Current Status
As of early 2025, the H100 compute market is in a state of transition. Supply has improved significantly, with lead times down to a few weeks for most cloud providers. The average Ornn price is around $2.00 per hour, down from $2.50 in mid-2024. The launch of the H200 in late 2024 has begun to siphon off some high-end demand, but H100 remains the workhorse for most AI training and inference workloads. The Blackwell B200 is expected to start shipping in large volumes in late 2025, which will likely accelerate H100 price declines. However, demand for AI compute continues to grow rapidly, with companies like OpenAI, Google, and Meta announcing plans to build massive data centers with hundreds of thousands of GPUs. The net effect on H100 prices by September 2026 is uncertain. Some analysts predict a gradual decline to $1.50 per hour, while others argue that sustained demand from inference workloads (which are less sensitive to GPU generation) will keep prices above $2.00.
Frequently Asked Questions
What is the Ornn index and how is it calculated?
The Ornn index tracks the average hourly price of NVIDIA H100 compute across multiple cloud providers. It is calculated as the arithmetic mean of hourly values reported by Ornn, using the USD iteration of the index. The specific providers and weighting methodology are not fully public, but it includes major cloud platforms and specialized GPU clouds.
How does the H100 price compare to the A100?
The H100 is typically 2-3 times more expensive per hour than the A100. In early 2025, A100 prices were around $0.80-$1.20 per hour for an 8-GPU node, while H100 was around $2.00. The H100 also offers roughly 3-4x the performance for AI training, making it more cost-effective for many workloads despite the higher absolute price.
Will the Blackwell B200 replace the H100?
The B200 is expected to be a higher-performance, higher-priced GPU that will coexist with the H100 rather than replace it entirely. The H100 will likely remain in production for the mid-range market, similar to how the A100 remained available after the H100 launched. Cloud providers will offer both, with H100 prices dropping as Blackwell takes the top end.
Educational content is AI-generated and sourced from Wikipedia. It should not be considered financial advice.
