NVIDIA GPU rental prices surge as AI demand hits full capacity
The cost of renting high-end AI chips is climbing sharply, with some models doubling in price as demand for AI inference grows.
The short version
- Hopper-architecture GPU lease prices have reached up to twice the original price.
- Blackwell-architecture B300 GPU rental prices rose by more than 60% since launch.
- NVIDIA's computing resources for cloud providers are currently operating at full capacity.
- Consumer GPU prices are rising, with RTX series models seeing recent cost adjustments.
- NVIDIA is requesting that cloud partners allocate at least 51% of services to non-China customers.
Why it matters
For businesses, this means the cost of running AI models is becoming significantly more expensive as hardware becomes scarce. It also signals a shift where even older, proven hardware is seeing massive price hikes due to a lack of available supply.
Surging costs for AI hardware
NVIDIA recently disclosed during an earnings call that the rental prices for its graphics processing units (GPUs) are climbing across multiple generations. This price hike is being driven by a sustained increase in demand for AI inference, which is the process of using a trained model to generate outputs. As more companies deploy AI tools, the need for the underlying hardware has reached a critical point.
The data shows that the price increases are not limited to the newest chips on the market. For example, the cost for renewing multi-year leases on Hopper-architecture GPUs has reached levels that are up to twice the initial price. This suggests that even established hardware is seeing a premium because the demand is outpacing the available supply of chips.
Newer hardware is also seeing significant price jumps. The lease price for Blackwell-architecture B300 GPUs has risen by more than 60% since they first launched. These figures indicate a rapidly tightening market where the cost of compute power is becoming a primary hurdle for AI development and deployment.
The drivers of high demand
A major factor contributing to these high rental prices is a decrease in the cost of AI tokens. Colette Kress, NVIDIA’s interim CEO and Chief Financial Officer, noted that AI token prices have dropped by nearly 60% since May. Because it is now cheaper for developers to use AI, the overall volume of computing demand has surged significantly.
This increased usage creates a feedback loop where lower costs for the end user lead to higher demand for the hardware that powers those services. As more people and companies use AI, the competition for the limited number of available GPUs intensifies. This competition allows providers to raise rental rates to meet the high volume of requests.
NVIDIA confirmed that the computing resources it provides to both hyperscale and emerging cloud providers are currently operating at full capacity. When a market reaches full capacity, any additional demand must be met by either finding new supply or increasing the price of existing resources. Currently, the market is leaning heavily toward the latter.
Impact on consumer markets
The high demand for AI hardware is not just affecting large-scale cloud providers and enterprise customers. The consumer GPU market is also experiencing price adjustments. Factories for major brands like Gigabyte and ASUS have implemented new cost adjustment policies for November that affect most models.
Specific examples of these price increases include the RTX series, which saw a price hike of 1,000 yuan for some models and 150 yuan for others. These changes reflect the broader trend of rising production and demand costs for the chips that power both gaming and local AI tasks.
For the average consumer, these price hikes mean that the entry point for high-performance computing is becoming more expensive. As the demand for AI-capable hardware grows, the hardware used by gamers and hobbyists is being pulled into the same supply constraints as the massive data centers used by major tech corporations.
Restrictions on Chinese access
The tightening of the GPU market is also creating hurdles for Chinese AI companies. Reports from the Financial Times indicate that NVIDIA has asked its global data center and cloud service partners to give at least 51% of their services to customers who are not in China. This request applies to future contracts and does not affect any existing lease agreements that are already in place.
This move by NVIDIA may further limit the ability of Chinese firms to access advanced AI computing power through overseas channels. The report noted a significant price discrepancy in the Chinese market, where a B300 rack system reportedly sold for RMB 18.5 million last month. This is considerably higher than the international sale price of approximately $600,000.
These restrictions and price premiums suggest a bifurcated market where access to high-end compute is becoming increasingly difficult for certain regions. As NVIDIA manages its global distribution and complies with various requirements, the cost and availability of its most powerful chips will continue to be a major point of contention for international AI development.
What to watch
- Whether the 60% drop in token prices will eventually stabilize the demand for GPUs.
- How many cloud providers will comply with NVIDIA's 51% non-China allocation request.
- Future price adjustments for the RTX series consumer cards.
Try it
Check your current cloud computing costs or upcoming hardware purchases to see if you can find more affordable alternative models as high-end GPU prices continue to rise.
Sources and links
- ReportingKuCoin: the article we sourced
First reported by KuCoin. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.