July 2027: Monthly average compute price of NVIDIA's B200
$0.00
1
9
July 2027: Monthly average compute price of NVIDIA's B200
$0.00
1
9
AI Analysis
Trader mode: Actionable analysis for identifying opportunities and edge
About This Event
July 2027 If the average value of NVIDIA B200 compute per hour is above X in July 2027, then the market resolves to Yes. The market resolves based on the average value of NVIDIA B200 compute per hour in July 2027, 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 unless otherwise
What Prediction Markets Are Forecasting
Traders on Kalshi currently give a 68% chance, roughly a 2 in 3 odds, that the average hourly compute price of an NVIDIA RTX 5090 will exceed $0.50 in July 2027. That's not a slam dunk. It suggests the market leans toward higher prices, but with real doubt baked in. A 68% probability is like a moderate favorite, not a lock.
The RTX 5090 is NVIDIA's top consumer graphics card, launched in early 2025 with a $1,999 MSRP. "Compute per hour" here means what people pay to rent that GPU's processing power, typically through cloud services or marketplaces that let users spin up AI workloads without buying the hardware outright. The index tracks the average rental rate for that card's compute in USD.
Why the Market Sees It This Way
Three forces are pushing odds upward. First, demand for AI compute keeps climbing. Training and running large language models, image generators, and video tools all need serious GPU horsepower, and the 5090 is one of the most accessible high-end options for smaller teams and researchers. Second, supply constraints could persist. NVIDIA's production capacity is stretched across data center chips like the H100 and B200, and consumer cards often take a back seat. If that continues into 2027, rental prices stay elevated.
Third, the $0.50 threshold sits below current spot rates for high-end consumer GPU rentals, which often range from $0.40 to $0.80 per hour depending on provider and demand. The market is essentially betting that prices won't crash over the next couple years. That's plausible if AI adoption keeps growing, but it's not certain.
Key Dates and Events to Watch
NVIDIA's product cycle matters. If a new flagship card, say the RTX 6090, launches before mid-2027, it could pull demand away from the 5090 and drag prices down. Watch for announcements at CES in January 2026 or GTC in March 2026. Cloud providers adjusting their pricing tiers also shifts the index. And any major AI company scaling back training budgets would soften demand.
How Reliable Are These Predictions?
Prediction markets have a decent track record on tech pricing questions, but two years is a long horizon. The market is aggregating current information, not perfect foresight. GPU prices have historically been volatile, swinging with crypto booms, supply chain shocks, and new product launches. The 68% figure is a reasonable guess, but treat it as a starting point, not a prophecy. A lot can change between now and July 2027.
Current Market Outlook
Kalshi traders currently price a 68% chance that the average hourly compute price of NVIDIA's RTX 5090 exceeds $0.50 in July 2027. That's a solid but not overwhelming majority. The market sees a price above half a dollar as the base case, but with enough uncertainty baked in that a dip below that threshold wouldn't shock anyone.
The RTX 5090 launched at $1,999 MSRP in early 2025. By mid-2027, the card will be roughly two and a half years old, squarely in the middle of its lifecycle. GPU rental markets like Vast.ai, RunPod, and Salad have historically shown that consumer-grade flagship cards retain surprising value well into their second year, particularly when AI inference workloads keep demand elevated.
Key Factors Driving the Odds
The 68% figure reflects three forces working in tandem. First, the AI compute shortage isn't going away. Data center GPUs like the H100 and B200 remain scarce and expensive, pushing smaller developers and researchers toward consumer cards. The RTX 5090's 32GB of VRAM makes it genuinely useful for running local LLMs, and that utility supports rental prices.
Second, historical pricing patterns matter. The RTX 4090, the 5090's predecessor, maintained rental rates above $0.30 per hour for nearly three years after launch. The 5090's generational leap in AI performance, roughly 2x the FP8 throughput, justifies a premium over that baseline. If the 4090 stayed above $0.30, a card with double the compute power sitting above $0.50 is a reasonable extrapolation.
Third, depreciation curves for GPUs have flattened. NVIDIA's consistent dominance means older cards don't lose value as quickly as prior generations. The 3090 still rents for meaningful rates today, years after its launch.
What Could Change These Odds
A major correction in AI demand is the obvious bear case. If the current AI capex bubble deflates by 2027, rental prices across the board would crater. The market is implicitly betting this doesn't happen.
On the supply side, NVIDIA's next-generation consumer cards, likely the RTX 6090 series, could arrive by late 2026 or early 2027. A larger generational leap than expected would accelerate depreciation of the 5090. Conversely, any continued supply constraints on Blackwell consumer GPUs would keep 5090 prices elevated.
The Ornn methodology matters too. The index uses hourly values from a specific data source, and revisions after expiration don't count. If Ornn's sampling methodology shifts or the platform's user base changes, the reported average could diverge from what a broader market would show.
The 68% price suggests the market views sub-$0.50 pricing as a real possibility but not the expected outcome. For that to happen, you'd need either a demand shock or a supply glut that overwhelms the 5090's still-strong utility in AI workloads. Neither looks likely from here, but two years is a long time in this industry.
AI-generated analysis based on market data. Not financial advice.
Overview
This prediction market focuses on the average hourly compute price of NVIDIA's H100 GPU in July 2027. The H100, released in 2022, is a high-performance graphics processing unit designed for AI training and inference, scientific simulation, and data analytics. Its compute price per hour is a key metric for cloud providers, AI startups, and large enterprises that rent GPU capacity rather than buying hardware outright. The market resolves to Yes if the arithmetic mean of hourly values reported by Ornn (the USD version) for July 2027 exceeds a specified threshold. Ornn is a provider of GPU pricing indices that track spot and on-demand rates across major cloud platforms such as AWS, Google Cloud, and Microsoft Azure. The H100's price has fluctuated due to supply constraints, demand surges from generative AI, and competition from newer chips like NVIDIA's own B200 and AMD's MI300 series. As of 2025, typical H100 on-demand pricing ranges from $2.50 to $4.00 per hour depending on the cloud provider and contract terms. The resolution date is after July 2027, with revisions to the underlying data not considered. This market attracts interest from investors, cloud cost analysts, and AI infrastructure forecasters who want to hedge or speculate on GPU pricing trends. The outcome depends on whether supply catches up with demand, whether alternative chips gain adoption, and whether AI model efficiency improvements reduce per-hour compute needs. It also reflects broader trends in the cloud computing and AI hardware markets, which have seen rapid growth and volatility since 2023.
Historical Context
The H100 was announced in March 2022 and began shipping in late 2022. Initial demand exceeded supply, leading to long lead times and premium pricing on secondary markets. In 2023, the launch of ChatGPT and other generative AI tools caused a surge in GPU demand. Cloud providers allocated H100s to high-priority customers, and spot prices sometimes exceeded $10 per hour. By 2024, NVIDIA increased production capacity and introduced the H200, a memory-upgraded variant. Supply constraints eased, but demand remained strong. In 2025, NVIDIA released the B200 Blackwell GPU, which offered higher performance for AI workloads. Some cloud providers began phasing out H100s in favor of newer chips. The H100's compute price declined from about $4.00 per hour in early 2024 to around $3.00 in mid-2025, as more capacity came online and competition intensified. The market's resolution date of July 2027 is far enough out that multiple generations of GPUs may be available. Historical precedents include the decline of earlier GPU prices after new generations launched. For example, the A100, released in 2020, saw its cloud price drop from about $2.50 per hour in 2021 to below $1.50 by 2023. The H100's price trajectory may follow a similar pattern, but the timeline is uncertain due to the unprecedented scale of AI investment.
Why It Matters
The price of H100 compute affects the economics of AI development and deployment. Startups and research labs that rent GPU time face cost pressures that influence their business models. If H100 prices remain high, it favors larger companies with capital to buy hardware. If prices fall, it lowers barriers to entry for smaller players. The broader economic impact includes the cost of training large language models, which can run into tens of millions of dollars. Lower compute costs could accelerate AI adoption across industries. Political and regulatory implications include concerns about concentration of AI compute power among a few firms. Governments in the US, EU, and China are considering policies to subsidize or regulate GPU access. The H100 price also reflects the health of the semiconductor industry and the pace of technological innovation. Downstream consequences include the profitability of cloud providers, the valuation of AI startups, and the direction of AI research. Investors in NVIDIA and its competitors watch GPU pricing as a leading indicator of demand.
Current Status
As of early 2025, H100 prices have stabilized after the initial supply crunch. Cloud providers are offering more flexible pricing options, including reserved instances and spot markets. The release of NVIDIA's B200 Blackwell in 2025 has started to shift demand away from the H100, but H100s remain widely used for inference workloads. Ornn's index shows average prices around $3.00 per hour for on-demand instances. The market's threshold for July 2027 is not specified in the description, but typical prediction market thresholds for such contracts are set at levels like $2.00 or $3.00 per hour. The outcome depends on whether the H100 remains in production and how quickly newer GPUs reduce its value. Some analysts predict a gradual decline to $1.50 to $2.00 per hour by 2027, while others expect prices to hold above $2.50 due to sustained demand from AI applications.
Frequently Asked Questions
What is the Ornn H100 compute price index?
Ornn's index tracks the average cost of renting an NVIDIA H100 GPU per hour across major cloud providers. It aggregates spot and on-demand prices using a consistent methodology. The USD version is used for this prediction market.
How does the H100 compare to the A100 in price?
The H100 is typically 2 to 3 times more expensive than the A100 on cloud platforms. The A100, released in 2020, now costs about $1.00 to $2.00 per hour on-demand. The H100 offers higher performance for AI training and inference.
Will NVIDIA stop selling the H100 before 2027?
NVIDIA typically supports products for several years. The H100 is likely to remain available through 2027, but production may be reduced as newer models like the B200 take over. Cloud providers may phase out H100 instances over time.
Can I buy an H100 directly instead of renting?
Yes, the H100 is available for purchase from NVIDIA and its partners. The retail price is around $30,000 per unit. However, cloud rental is more common for flexible workloads. Buying may be cheaper for high-utilization scenarios.
What factors could cause H100 prices to rise in 2027?
Possible factors include a new AI breakthrough that increases demand, supply chain disruptions, or trade restrictions that limit GPU availability. However, the general trend is downward as newer GPUs are released and supply increases.
How accurate are Ornn's price indices?
Ornn's methodology is designed to reflect real transaction prices, but it may not capture all discounts or negotiated rates. The indices are used by financial markets and are considered reliable for prediction market resolution.
Educational content is AI-generated and sourced from Wikipedia. It should not be considered financial advice.
