GPU VPS for generative AI: hourly and monthly pricing, 5 providers compared
DigitalOcean, Scaleway, Google Cloud, RunPod and Akamai Cloud price lists for RTX 4000 Ada and L4 GPUs: hourly pricing, VRAM and estimated monthly cost.
UptimeMag editorial team · 16 October 2026 · 3 min read
Anyone generating images or video with diffusion models doesn't need an H100. A mid-range GPU with 20–24 GB of VRAM — RTX 4000 Ada or L4 — is sufficient for Stable Diffusion, upscaling models and short clips, at a significantly lower hourly rate. The problem is that price lists are scattered across dozens of providers and change frequently. We've compared five providers by verifying figures on official pages or industry trackers updated daily, as of 16 October 2026.
The five price lists compared
| Provider | GPU | VRAM | Price/hour (on-demand) | Estimated monthly (730h) | Billing |
|---|---|---|---|---|---|
| DigitalOcean | RTX 4000 Ada | 20 GB | $0.76 | $554.80 (standard cap 672h ≈ $510.72) | per second, 5 minute minimum |
| Scaleway | L4 | 24 GB | from €0.79 | ≈ €576.70 | as used, storage and IP separate |
| Google Cloud (g2-standard-4) | L4 | 24 GB | $0.70 | ≈ $511.00 | per second, 1 minute minimum |
| RunPod | RTX 4000 Ada (Community) / L4 (Secure) | 20 / 24 GB | from $0.18 (RTX 4000 Ada) / from $0.39 (L4) | from $131.40 to $284.70 | per second |
| Akamai Cloud (Linode) | RTX 4000 Ada | 20 GB | $0.52 | ≈ $379.60 | hourly |
The DigitalOcean price is from the official price list: RTX 4000 Ada: $0.52 /GPU/hr On-demand is the figure for Akamai, confirmed by two other independent trackers reporting the same value for the same period. On the DigitalOcean configuration, the GPU comes with 32 GiB of system memory, 8 vCPU, 500 GiB NVMe storage and 10,000 GiB of traffic included in the hourly rate, billed per second with a 5 minute minimum; the official note also clarifies that beyond the standard 672 monthly hours, you continue to pay based on actual seconds used.
For Scaleway, the published price starts at €0.79 per GPU/hour on the PAR-2 zone, with vCPU and system RAM included in the price but block storage and flexible IP charged separately.
On Google Cloud, the price for the g2-standard-4 configuration with one L4 is g2-standard-4, 1x L4 24 GB: $0.706832276/hr on demand in us-central1 (Iowa), a figure that includes 4 vCPUs and 16 GiB RAM, whilst storage and networking remain separate.
On RunPod two different tariffs coexist depending on service level: RunPod lists L4 pods from $0.44/hr (Community Cloud) and $0.39/hr (Secure Cloud), whilst for the RTX 4000 Ada several independent trackers (GetDeploying, ComputePrices, gpus.io) converge on a range of between $0.18 and $0.28 per hour depending on region and current availability.
What's missing from the price lists
None of the five providers publishes instance startup time (provisioning time) on their pricing page: this is data measured only by actually launching an instance, not a figure from a price list. Anyone planning a burst workload (for example overnight batch rendering) should therefore run a direct test before choosing, rather than relying on a stated average.
How to read the table
Monthly cost is simple multiplication (hourly price × 730 hours), useful for comparison but not a discount price list: all five providers offer reserved or spot plans at lower rates for those committing over months or accepting interruptions, which we haven't included here because access conditions differ from pure on-demand pricing.
Written with the help of artificial intelligence and checked by the editors (EU AI Act, art. 50).