June 2027: Monthly average compute price of NVIDIA's RTX 5090
$0.00
1
7
June 2027: Monthly average compute price of NVIDIA's RTX 5090
$0.00
1
7
AI Analysis
Trader mode: Actionable analysis for identifying opportunities and edge
About This Event
June 2027 If the average value of NVIDIA RTX 5090 compute per hour is above X in June 2027, then the market resolves to Yes. The market resolves based on the average value of NVIDIA RTX 5090 compute per hour in June 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 o
What Prediction Markets Are Forecasting
Traders on Kalshi currently give a 72% chance, roughly a 7 in 10 shot, that the average hourly compute price of an NVIDIA RTX 5090 will exceed $2.00 in June 2027. That might sound like a narrow technical detail, but it's really a bet on how fast AI infrastructure costs will fall over the next three years.
The RTX 5090 is NVIDIA's flagship consumer GPU, launched in early 2025. It's not just for gamers. People rent these cards by the hour through cloud providers to run AI models, render graphics, or train small neural networks. The "compute price" here means what it costs to rent one hour of that card's processing power, averaged across all providers tracking it.
A 72% probability means the market is fairly confident, but not certain. There's still a 28% chance prices stay below $2.00. That's like the odds of rolling a 5 or 6 on a single die. Real uncertainty remains.
Why the Market Sees It This Way
Three forces are pushing prices up toward that $2.00 threshold.
First, the RTX 5090 is a premium card with massive demand. It has 32GB of VRAM and serious compute power, making it attractive for running AI models locally or in small cloud setups. When demand outstrips supply, rental prices rise.
Second, electricity costs are climbing in many regions. GPU rental prices track power prices closely, since electricity is a large share of operating costs. If energy stays expensive, so does compute.
Third, there's the AI arms race. Companies keep buying GPUs faster than manufacturers can make them. Even with NVIDIA ramping up production, the 5090 remains a high-end product with limited supply compared to the broader market.
That said, the market isn't pricing this above $2.00 with total confidence. Newer GPUs could arrive by 2027, pulling demand away. Cloud providers could overbuild capacity. Or energy prices could drop. Each of those would push prices below $2.00.
Key Dates and Events to Watch
NVIDIA's annual GPU architecture updates matter most. If a new flagship card launches in late 2026 or early 2027, it could steal demand from the 5090 and lower rental prices. Watch for announcements at CES in January or GTC in March.
Electricity price trends in major data center regions, especially Texas and Virginia, will also shift the outlook. Any significant energy policy changes or grid disruptions could move the market.
Finally, watch NVIDIA's production numbers. If they announce a major supply increase for the 5090 line, cloud providers might lower prices to stay competitive.
How Reliable Are These Predictions?
Prediction markets have a decent track record with tech pricing questions, but this one is unusually specific. The market depends on a single data source, Ornn, which tracks GPU rental prices across various providers. That's a niche dataset, and the market's accuracy depends on how well Ornn captures real-world transactions.
Markets tend to be better at forecasting broad trends than precise price levels. A 72% probability on a specific dollar amount two years out carries real uncertainty. The market is saying "more likely than not," not "almost certainly." For anyone watching AI infrastructure costs, this is a useful signal, but not a crystal ball.
Current Market Outlook
The market is pricing a 90% probability that NVIDIA's A100 compute will average above $0.50 per hour in June 2027. That means traders see this as nearly certain. For context, the A100 launched in 2020 and has been the workhorse for AI training through the 2023-2024 boom. The $0.50 threshold is relatively low compared to peak pricing of $1.50-$2.00 per hour in 2023 when GPU shortages were acute.
Key Factors Driving the Odds
The A100 is now a last-generation chip. NVIDIA's H100 replaced it as the premium option in 2023, and the B100 is expected in 2025. Older hardware typically sees price declines as supply expands and demand shifts to newer products. But the A100 retains a specific advantage: it's widely available on cloud platforms like AWS, GCP, and Azure, and many AI workloads don't require the latest hardware. Inference tasks, fine-tuning, and smaller model training run fine on A100s.
Two forces are keeping prices above $0.50. First, total AI compute demand is still growing rapidly. The market for GPU cloud services hit $20 billion in 2024 and could double by 2027. Second, hyperscalers have signed long-term contracts locking in A100 capacity at fixed prices. This creates a floor. Even if spot prices fall, the average stays up because most compute is sold under those contracts.
What Could Change These Odds
The main risk to the 90% probability is a supply glut. NVIDIA shipped massive volumes of A100s in 2022-2023, and those GPUs are now entering secondary markets. If too many cloud providers overbuilt A100 capacity, spot prices could crash below $0.30. That happened with older NVIDIA chips like the V100, which fell from $1.00 to $0.20 within two years of being superseded.
The other catalyst is the B100 ramp. If NVIDIA ships the B100 on schedule and it's dramatically more efficient, demand for A100s could collapse. But the market is betting that legacy workloads and inference demand will keep the A100 relevant at $0.50 through mid-2027. That's a reasonable bet given how slowly enterprise AI adoption actually moves.
AI-generated analysis based on market data. Not financial advice.
Overview
This prediction market concerns the average hourly compute price of NVIDIA's A100 GPU in June 2027. The A100, released in 2020, is a data center GPU based on the Ampere architecture. It became a foundational hardware component for AI training and inference, particularly for large language models and scientific computing. The market resolves to Yes if the average price per hour, as reported by Ornn's USD index, exceeds a specified threshold (X). Ornn is a provider of cloud compute pricing data, aggregating spot and on-demand rates from major cloud providers. The price of A100 compute has fluctuated due to supply constraints, demand from AI startups and hyperscalers, and competition from newer GPUs like the H100 and Blackwell series. By June 2027, the A100 will be a 7-year-old product, but its installed base remains large, and it is still widely used for inference workloads. The market's outcome will reflect the balance between ongoing demand, the depreciation of older hardware, and the availability of cheaper alternatives. Interest in this market stems from its use as a proxy for AI infrastructure costs and the broader health of the cloud computing industry. Investors and analysts use GPU pricing trends to gauge capacity utilization, chip lifecycle economics, and the pace of technological substitution.
Historical Context
NVIDIA's A100 GPU was announced in May 2020 and began shipping later that year. It was the first GPU based on the Ampere architecture, featuring third-generation Tensor Cores and support for structural sparsity. Initial list prices were around $10,000 to $15,000 per unit, but cloud providers offered hourly rates ranging from $1 to $3 depending on instance type and commitment. During 2021-2022, demand surged due to the AI boom, causing spot prices to spike above $5 per hour on some providers. Supply chain constraints and cryptocurrency mining demand also pushed prices higher. In 2023, the introduction of the H100 (Hopper) began to displace the A100 for training workloads, but the A100 remained popular for inference due to its lower cost and sufficient performance for many models. By 2024, cloud providers had increased A100 capacity, and spot prices stabilized around $1-$2 per hour. The 2024 release of the Blackwell B200 GPU in late 2024 further accelerated the shift away from A100 for new deployments. Historically, older GPU generations see price declines of 30-50% per year as newer products enter the market and supply chains mature. However, the A100's longevity has been unusual due to sustained AI demand and the slower-than-expected adoption of newer GPUs in some regions.
Why It Matters
The price of A100 compute is a direct indicator of AI infrastructure costs. Many AI startups and research labs rely on A100 instances for model training and inference. If prices remain high, it suggests that supply is still constrained or that demand for older hardware persists, which could delay the adoption of newer GPUs and increase costs for AI development. Conversely, falling prices indicate oversupply or technological obsolescence, which could lower barriers to entry for AI companies but also signal reduced profitability for NVIDIA and cloud providers. This metric also affects the economics of AI training. For example, training a large language model like GPT-4 required thousands of A100 GPUs running for weeks. A 10% change in hourly price translates to hundreds of thousands of dollars in cost differences. Investors in AI companies, cloud providers, and semiconductor stocks watch these prices to assess market conditions. Furthermore, the A100's pricing trajectory informs debates about whether AI hardware will become a commodity or remain a premium product. The outcome of this market will provide a data point for analysts forecasting NVIDIA's future revenue mix and the broader shift toward specialized AI chips.
Current Status
As of late 2025, the A100 remains in production but is being phased out in favor of the H100 and Blackwell series. Cloud providers still offer A100 instances, but new capacity additions are primarily for newer GPUs. Spot prices for A100 have stabilized between $1.00 and $2.00 per GPU per hour on most providers. The Ornn index for June 2025 is not yet available, but monthly averages in early 2025 have been around $1.50-$1.80. The key uncertainty for June 2027 is whether demand for A100 will collapse as more efficient GPUs become available, or if it will remain steady due to legacy workloads and price-sensitive customers. NVIDIA has not announced a discontinuation date for the A100, but typical product lifecycles in data centers are 3-5 years, suggesting the A100 will be end-of-life by 2027.
Frequently Asked Questions
What is the Ornn index and how is it calculated?
Ornn is a data provider that tracks compute prices across cloud providers. Its USD index for NVIDIA A100 compute per hour is the arithmetic mean of hourly values reported from major clouds. The index includes both on-demand and spot prices, weighted by availability.
Will the A100 still be relevant in 2027?
Yes, but its role will shift. By 2027, the A100 will be a legacy product used primarily for inference on older models and cost-sensitive workloads. New training will likely use H100 or Blackwell GPUs. However, the large installed base will keep some demand alive.
How does spot pricing affect the average?
Spot instances can be 60-80% cheaper than on-demand. Because Ornn averages across all reported prices, a high proportion of spot usage can lower the monthly average significantly. The market outcome depends on the mix of spot and on-demand usage in June 2027.
What factors could cause A100 prices to rise in 2027?
A rise could occur if there is a supply shock, such as a manufacturing issue with newer GPUs, or if demand unexpectedly increases due to a new AI application that requires massive inference capacity. Geopolitical disruptions affecting NVIDIA's supply chain could also push prices up.
How reliable is Ornn's data?
Ornn is a reputable provider used by researchers and analysts. Its methodology is transparent, and it pulls data directly from cloud provider APIs. However, users should be aware that the index may not capture every instance type or region, and revisions can occur.
Educational content is AI-generated and sourced from Wikipedia. It should not be considered financial advice.
