New NVIDIA B300 · 288 GB HBM3e servers — from $4,019/mo per GPU

See the B300
Pay in crypto Deploy a GPUDeploy

Market check · 23 September 2026

CryptGPU vs the
GPU cloud market.

We collected the public on-demand prices of 40 GPU clouds and hosts. Here is every number, how we used it, and where hourly clouds remain the better choice.

  • 40providers checked
  • 9 of 10GPUs below the lowest price
  • −30%vs the median
  • 23 Sep 2026last check

Our price on the market range.

Each bar spans the lowest to the highest on-demand list price we found for the GPU, per GPU per month.

9 of 10GPUs are priced below the lowest on-demand list price we found — not just below the median.

  • B300288 GB · 8 quotes $4,019$5,745−30%
  • B200180 GB · 10 quotes $3,439$4,920−30%
  • MI355X*288 GB · 2 quotes $1,409$2,022−30%
  • H200141 GB · 12 quotes $2,279$3,259−30%
  • H10080 GB · 15 quotes $1,779$2,548−30%
  • RTX PRO 600096 GB · 9 quotes $959$1,380−31%
  • L40S48 GB · 10 quotes $789$1,139−31%
  • RTX 509032 GB · 15 quotes $329$481−32%
  • RTX 409024 GB · 16 quotes $269$387−30%
  • RTX 5080*16 GB · 2 quotes $199$299−33%

Monthly price per GPU. Each bar starts at $0 and is scaled to the highest list price found for that GPU.
* Few public list prices exist for the MI355X and the RTX 5080: their reference is the median of the public prices we found (starting prices for the MI355X).

Every GPU, every reference price.

Monthly price per GPU (hourly × 730). “vs lowest” compares our price with the cheapest on-demand list price we found.

Market prices per GPU per month compared with CryptGPU
GPUQuotesLowestMedianHighestCryptGPUvs medianvs lowest
NVIDIA B300288 GB · Blackwell Ultra 8 $4,818$6.60/h $5,745$7.87/h $6,278$8.60/h $4,019$5.51/h −30% −17%
NVIDIA B200180 GB · Blackwell 10 $3,964$5.43/h $4,920$6.74/h $6,278$8.60/h $3,439$4.71/h −30% −13%
AMD MI355X*288 GB · CDNA 4 2 $1,891$2.59/h $2,022$2.77/h $2,154$2.95/h $1,409$1.93/h −30% −25%
NVIDIA H200141 GB · Hopper 12 $2,548$3.49/h $3,259$4.46/h $4,606$6.31/h $2,279$3.12/h −30% −11%
NVIDIA H10080 GB · Hopper 15 $1,964$2.69/h $2,548$3.49/h $4,497$6.16/h $1,779$2.44/h −30% −9%
NVIDIA RTX PRO 600096 GB · Blackwell 9 $975$1.34/h $1,380$1.89/h $2,190$3.00/h $959$1.31/h −31% −2%
NVIDIA L40S48 GB · Ada Lovelace 10 $472$0.65/h $1,139$1.56/h $1,643$2.25/h $789$1.08/h −31% +67%
NVIDIA RTX 509032 GB · Blackwell 15 $384$0.53/h $481$0.66/h $999$1.37/h $329$0.45/h −32% −14%
NVIDIA RTX 409024 GB · Ada Lovelace 16 $283$0.39/h $387$0.53/h $887$1.22/h $269$0.37/h −30% −5%
NVIDIA RTX 5080*16 GB · Blackwell 2 $285$0.39/h $299$0.41/h $313$0.43/h $199$0.27/h −33% −30%

* Indicative: no public on-demand price exists yet for the MI355X (public starting prices used), and only two public prices were found for the RTX 5080. GeForce references come mostly from monthly dedicated-server prices, divided by 730 for the hourly figure.

AWS, Google Cloud, Azure and Oracle.

Public on-demand prices per GPU-hour, for reference. They are two to three times the market median, so they are excluded from it.

Hyperscaler on-demand prices per GPU-hour compared with the CryptGPU hourly equivalent
GPUAWSGoogle CloudAzureOracle CloudCryptGPU
NVIDIA B300 $17.80p6-b300.48xlarge · 3.2× — — $15.00per GPU · 2.7× $5.51$4,019/mo
NVIDIA B200 $14.24p6-b200.48xlarge · 3.0× — — $14.00per GPU · 3.0× $4.71$3,439/mo
AMD MI355X — — — $8.60per GPU · 4.5× $1.93$1,409/mo
NVIDIA H200 $7.91p5en.48xlarge · 2.5× — $13.78ND96isr H200 v5 · 4.4× $10.00per GPU · 3.2× $3.12$2,279/mo
NVIDIA H100 $6.88p5.48xlarge · 2.8× $11.07a3-highgpu-8g · 4.5× $12.29ND96isr H100 v5 · 5.0× $10.00per GPU · 4.1× $2.44$1,779/mo
NVIDIA RTX PRO 6000 $3.36g7e.2xlarge · 2.6× $4.50g4-standard-48 · 3.4× — $4.50per GPU · 3.4× $1.31$959/mo
NVIDIA L40S $1.86g6e.xlarge · 1.7× — — $3.50per GPU · 3.2× $1.08$789/mo

US dollars per GPU-hour, on-demand, 23 September 2026. “×” = multiple of our hourly equivalent (monthly price ÷ 730). — = no public on-demand price found. AWS and Azure from their public price lists (via instances.vantage.sh), Oracle from its public price list, Google Cloud via getdeploying.com. Committed-use discounts can lower hyperscaler prices.

What we counted, and what we left out.

Counted in the median

  • Public, on-demand list prices, published on the provider’s own pricing page or API.
  • One price per provider and GPU: the per-GPU price of the 8-GPU node when published, otherwise the single-GPU price.
  • For GeForce cards, the cheapest monthly price of each dedicated host, and hourly cloud prices × 730.

Left out

  • AWS, Google Cloud, Azure and Oracle Cloud (shown above for reference).
  • Spot, preemptible and interruptible prices.
  • Peer-to-peer marketplaces: Vast.ai, RunPod Community Cloud, Salad, TensorDock.
  • Prices that need a commitment, quote-only offers, and listings that were ambiguous or abnormal.

The 40 providers in the sample

  • Lambda
  • CoreWeave
  • RunPod
  • Nebius
  • Crusoe
  • Hyperstack
  • Verda
  • Together AI
  • DigitalOcean
  • OVHcloud
  • Scaleway
  • Vultr
  • Hetzner
  • LeaderGPU
  • HOSTKEY
  • Civo
  • Jarvislabs
  • Lyceum
  • Massed Compute
  • Paperspace
  • Latitude.sh
  • TensorWave
  • GPU Mart
  • CloudClusters
  • Primcast
  • ProlimeHost
  • SeiMaxim
  • Cloudzy
  • Trooper.AI
  • BreezeHost
  • ServerMO
  • Hostease
  • Ranch Computing
  • RackNerd
  • Hostcode
  • Bthorio
  • GPUYard
  • CloudRift
  • Nova Cloud
  • Oblivus

Each provider contributes to the median of at least one GPU. Monthly price = median hourly price per GPU × 730 hours; our price = 70% of it, rounded down to end in 9.

When hourly is the better deal.

A lower price is not the whole story. Here is where each model wins.

Choose an hourly cloud if

  • Your GPUs run less than about 17 hours a day on average.
  • You need capacity for a few hours or days, or burst to many GPUs at once.
  • You train across several servers joined by InfiniBand or a similar fabric.
  • You want managed services: notebooks, managed Kubernetes, serverless inference.

Choose CryptGPU if

  • Your GPUs run most of the day: serving, training runs, rendering queues.
  • You want a fixed monthly budget with no metering.
  • You want dedicated GPUs and root access on your own server.
  • You prefer to pay in BTC, ETH, USDT, USDC or LTC.
Find your break-even

The GPU market, September 2026.

  • Most powerful rentable GPU

    NVIDIA’s Rubin generation is not publicly rentable yet. The B300 (Blackwell Ultra) is the most powerful GPU available on demand, from 8 providers in our sample.

  • Prices are rising

    Nebius raises its on-demand prices for H100, H200, B200 and B300 by 17 to 21% on 1 October 2026.

  • AMD MI355X

    No provider publishes an on-demand price for the MI355X yet; public offers are starting prices, which is why its market reference is indicative.

  • Hyperscalers

    AWS, Google Cloud and Azure list GPUs at two to three times the market median: for example $6.88 per H100-hour on AWS p5 and $17.80 per B300-hour on AWS p6-b300.

About the comparison.

Why leave out AWS, Google Cloud, Azure and Oracle?

Their on-demand GPU prices are two to three times the market median. Including them would inflate our discount, so they are shown separately for reference and kept out of the median.

Why not compare with marketplaces such as Vast.ai?

Marketplaces list GPUs from third-party hosts, often interruptible or unverified, and some offers are below our prices. It is not the same product as a dedicated server rented by the month, so we keep them out of the median, together with spot offers.

Are you really cheaper than every provider?

For 9 of our 10 GPUs, our monthly price is below the lowest on-demand list price we found. The exception is the L40S: one dedicated host lists it at about $472 per GPU per month in an 8-GPU server, below our $789.

How often is this comparison updated?

This page shows the check of 23 September 2026. Every update of the comparison is dated here, and our prices are derived from it.

Why a median rather than an average?

A few very expensive list prices would pull an average up and make our discount look larger. The median is the price in the middle of the sample: half the providers are cheaper, half are more expensive.

The math checks out. Pick your GPU.

Every price on this site comes from the same comparison, dated 23 September 2026.

Configure a server Price list