Cloud GPU Pricing, Performance & Provider Comparison
Cloud GPU list prices for the same model can differ several times over between providers. The tables compare lowest rates, provider names, market ranges and listing medians by GPU, memory and interconnect.
Cloud GPU price per throughput
Compare measured throughput per dollar in the benchmark calculator:
Cloud GPU Pricing & Performance
1,245 offerings across 28 providers
IONOS
Vast AI
Best PriceVast AI
Vast AI
Verda
Verda
Verda
Verda
Vast AI
Runpod
See cloud GPU benchmark methodology for details.
On-demand billing charges for usage by the hour or a smaller time unit, depending on the provider, without a long-term reservation.
On-demand instances allow usage without a long reservation. Availability, maintenance and service commitments depend on the provider and instance terms.
On-demand cloud GPU prices
Ranking: Subscribers are linked and highlighted at the top. Other rows are sorted by lowest on-demand price. Range and median summarize captured listings for the same GPU, memory and interconnect grouping; region, host resources and GPU count can still differ. Prices use the November 5, 2026 catalog capture. IONOS rows retain the separately maintained sponsor reference prices. Unknown prices are excluded.
On-demand rates apply without a reservation commitment. Published listings can include unavailable or unknown stock states, so a listed price does not guarantee that an instance can be started.
Spot cloud GPU prices
Ranking: Subscribers appear first; other rows are sorted by lowest spot price. Median summarizes captured spot listings for the same GPU grouping. N/A means no spot rate is documented for the sponsor reference row.
Spot capacity can be reclaimed by the provider. In 154 matched provider/GPU/memory/interconnect comparisons, the middle 50% of observed discounts is 26–55% versus on-demand. Eight matches have higher spot prices. Each comparison uses the lowest captured rate per billing tier; remaining instance differences can affect the result. Jobs using spot need to tolerate interruptions and restart costs.
Reserved cloud GPU prices
Ranking: Subscribers appear first; other rows are sorted by lowest reserved price. The catalog combines different commitment lengths. Median summarizes captured reserved listings for the same GPU grouping. N/A means no reserved rate is documented for the sponsor reference row.
In 134 matched provider/GPU/memory/interconnect comparisons, the middle 50% of reserved discounts is 15–44% versus on-demand. Three matches have higher reserved prices. Each pair uses the lowest captured rate in its tier. Commitment length, payment schedule and capacity guarantees must be checked in the individual offer.
Cloud provider performance comparison
The same GPU model can perform slightly differently across providers because of host CPU choice, network fabric, driver configuration, and virtualization overhead. To quantify this, we ran identical text and image generation workloads on AMD MI300X 192GB at DigitalOcean and Runpod:
Key Observations:
- For text generation, Digital Ocean demonstrated a slightly higher throughput, processing approximately 0.4% more tokens per second.
- Conversely, for image generation, Runpod showed a marginal advantage, processing about 0.4% more images per second.
These runs differ by less than 0.5% on both measured workloads. The comparison does not establish a performance ranking for other models or instance configurations.
Buy on-prem or rent in the cloud
The cost of owning a GPU depends on purchase price, utilization, power, cooling, maintenance and resale value. Compare those costs over the expected service life with cloud rental at the intended usage.
A workload can also use owned capacity for steady demand and cloud instances for temporary peaks. The split depends on operational requirements and the cost of moving data between environments.
Consumer GPUs such as RTX 4090 and RTX 5090 appear in rental catalogs. Compare their memory, interconnect, support and applicable software-license terms with the requirements of the deployment.
Add as preferred sourceCloud GPU benchmark methodology
Throughput benchmarks use 4-bit FP quantization across all tests. The pipeline runs:
- Text finetuning: Llama 3.2 on the first 5,000 conversations from FineTome, 5 epochs, 1M total tokens, Unsloth framework. Throughput = (tokens × epochs) / total time.
- Text inference: 1M tokens generated with llama-cpp-python.
- Image finetuning: YOLOv9 on 100 images from SkyFusion, 4 epochs, Unsloth.
- Image inference: Finetuned YOLOv9 on ~500 images at 640×640.
Throughput per dollar expresses workload output per unit of cost. The measured workload, runtime, hourly rate and GPU configuration determine the result. Other models and settings can produce different throughput.
FAQs
A GPU model can include different memory and interconnect variants. A100 includes 40 GB and 80 GB versions; V100 includes 16 GB and 32 GB versions. The tables separate memory and interconnect where identified, but GPU count, region and bundled CPU/RAM can still vary within a row.
The calculator compares workload output with cost for the measured configuration. A higher throughput-per-dollar value means more output per dollar for that workload. The workload and embedded price assumptions must be checked before applying the ranking to another job.
The pricing tables in this update use the catalog captured on November 5, 2026.
Further reading
- Multi-GPU Benchmark: B200 vs H200 vs H100 vs MI300X
- Cloud GPU Providers & Their GPUs
- GPU Concurrency Benchmark
- Top 25+ AI Chip Makers: NVIDIA & Its Competitors
- Cloud GPU Rental Price Index
- DGX Spark vs Mac Studio & Halo: Benchmarks & Alternatives
Cite this benchmark
Pick the format that matches where you're publishing. Pasting the link version into your CMS preserves the backlink.
@misc{dilmegani2026,
author = {Dilmegani, Cem and Sarı, Ekrem},
title = {{Cloud GPU Pricing, Performance & Provider Comparison}},
year = {2026},
month = nov,
howpublished = {\url{https://aimultiple.com/cloud-gpu-pricing}},
note = {AIMultiple. Retrieved November 6, 2026}
}Results and timestamps of 106 data points. Download the summary data shown in this article's charts and tables as a ZIP file containing 5 CSV files.
Want the granular data behind it? Join Premium
Changelog
15 updatesAdded spot and 1-year reserved cloud GPU pricing tables alongside the on-demand table.
Removed the "Lowest prices for most popular GPUs" section.
Cem's work at AIMultiple has been cited by leading global publications including Business Insider, Forbes, Morning Brew, and Washington Post, global firms like Deloitte and HPE, NGOs like World Economic Forum, and supranational organizations like European Commission. [1], [2], [3], [4], [5]
Throughout his career, Cem served as a tech consultant, tech buyer and tech entrepreneur. He advised enterprises on their technology decisions at McKinsey & Company and Altman Solon for more than a decade. He also published a McKinsey report on digitalization.
He led technology strategy and procurement of a telco while reporting to the CEO. He has also led commercial growth of deep tech company Hypatos that reached a 7 digit annual recurring revenue and a 9 digit valuation from 0 within 2 years. Cem's work in Hypatos was covered by leading technology publications like TechCrunch and Business Insider.
Cem regularly speaks at international technology conferences. He graduated from Bogazici University as a computer engineer and holds an MBA from Columbia Business School.
Comments 2
Share Your Thoughts
Your email address will not be published. All fields are required. Comments are left in their original language.
Cem - great article, I'd love to pick your brain on private networking or direct connects to these GPU instances.
Hi Ashley, thank you! Sure, happy to chat.
Hi there, fantastic article and very well-researched. Would you mind checking out Dataoorts at https://dataoorts.com
Sure, we'll review to see if we can include Dataoorts in the next edit.