GPU rental prices · Taipei Ledger

5 min read Original article ↗
Taipei LedgerData Analysis on Taiwan's Companies and Economy

AI infrastructure · updated daily

What one GPU-hour costs to rent, in US dollars, by GPU model and provider. Prices of 14 November 2026.

GPUHyperscalersNeocloudsGPU specialistsMarketplacesthird-party hostsRange
AWSon-demandAzurepay-as-you-goOracleon-demandCoreWeaveon-demandLambdaon-demandNebiuson-demandCrusoeon-demandRunPodSecure CloudRunPodCommunity CloudVast.aimedian
B300288 GB$17.80–$15.00––$9.50$8.96$8.99$6.94$11.84from $10.63 · 4 offers$6.94 – $17.80
B200180 GB$14.24–$14.00$8.60–$8.50–$7.99$5.98$10.00from $8.75 · 13 offers$5.98 – $14.24
H200141 GB$7.91–$10.00$6.30–$5.40$4.29$5.29$3.59$4.94from $2.37 · 14 offers$2.37 – $10.00
H100 SXM80 GB$6.88$12.29$10.00$6.16–$4.50$3.90$4.39$2.69$2.34from $1.79 · 17 offers$1.79 – $12.29
H100 PCIe / NVL80–94 GB–$6.98–––––$2.89$1.99$2.93from $1.60 · 11 offers$1.60 – $6.98
A100 80GB80 GB$3.43$4.10$4.00$2.70––$2.00$2.09$1.19$0.87from $0.84 · 21 offers$0.84 – $4.10
MI300X192 GB––$6.00–––$3.45$2.39––$2.39 – $6.00
L40S48 GB$1.86–$3.50$2.25–$1.55$1.50$1.09$0.79$0.80from $0.44 · 23 offers$0.44 – $3.50
RTX PRO 600096 GB––$4.50$2.50–$1.80–$2.69$1.69$1.67from $0.80 · 110 offers$0.80 – $4.50
RTX 509032 GB–––––––$1.39$0.69$0.55from $0.40 · 64 offers$0.40 – $1.39
RTX 409024 GB–––––––$0.89$0.34$0.51from $0.36 · 64 offers$0.34 – $0.89

Compute only: storage and data transfer cost extra (for Vast.ai, see below). Hover a price for the instance it comes from. * = source not updated today; last price shown.

Announced price changes (published by the provider, not yet in effect):

  • Crusoe B300: $8.96 → $9.90 from 01 Dec 2026
  • Crusoe H100 SXM: $3.90 → $4.50 from 01 Dec 2026
  • Crusoe H200: $4.29 → $5.40 from 01 Dec 2026

GPU rental prices, every Friday
The week's changes in data-centre GPU rental prices come in the free Friday email, with Taiwan's AI supply chain. Every listed Taiwan company reports its November sales by 10 Dec. The free Friday email on 11 Dec has the summary: which supply-chain groups grew and which shrank.

After subscribing, check your inbox (and spam folder) and confirm your subscription. We do this to protect you from spammers.

Data transfer and disk on Vast.ai

Each Vast.ai host sets its own charge for data sent to and from the machine, billed per GB on top of the hourly price. Most charge little, but some charge US$30 or more per TB in each direction, which can outweigh a lower hourly price when large datasets or model weights are moved. Charges of the same day's offers, in US$ per TB (1 TB = 1,000 GB).

GPUOffersFree both waysDownloadmedian, per TB Uploadmedian, per TBHighestper TBUS$30+ per TB Extra diskmedian, per GB-month
B30040$2.44$3.66$3.660$0.19
B200170$36.62$36.62$36.629$0.94
H200192$2.57$3.85$25.700$0.20
H100 SXM2310$2.60$3.91$39.064$0.20
H100 PCIe / NVL140$2.60$3.91$100.004$0.27
A100 80GB310$2.60$3.91$52.084$0.20
L40S310$2.60$3.91$13.020$0.13
RTX PRO 60001443$2.64$3.91$100.0027$0.20
RTX 5090990$5.21$5.21$40.008$0.20
RTX 4090961$3.91$4.58$53.3316$0.27

Highest and US$30+ count the dearer direction of each offer. The hourly price includes a small default disk; more disk is charged per GB-month. AWS and Azure charge for data sent out to the internet: see AWS data transfer and Azure bandwidth pricing.

Price history

Daily since 02 November 2026.

B300: US$ per GPU-hour
$0$10$2002 Nov14 Nov 2026

AWSOracleNebiusCrusoeRunPod SecureRunPod CommunityVast.ai median

B200: US$ per GPU-hour
$0$8$1602 Nov14 Nov 2026

AWSOracleCoreWeaveNebiusRunPod SecureRunPod CommunityVast.ai median

H200: US$ per GPU-hour
$0$6$1102 Nov14 Nov 2026

AWSOracleCoreWeaveNebiusCrusoeRunPod SecureRunPod CommunityVast.ai median

H100 SXM: US$ per GPU-hour
$0$7$1402 Nov14 Nov 2026

AWSAzureOracleCoreWeaveNebiusCrusoeRunPod SecureRunPod CommunityVast.ai median

H100 PCIe / NVL: US$ per GPU-hour
$0$4$802 Nov14 Nov 2026

AzureRunPod SecureRunPod CommunityVast.ai median

A100 80GB: US$ per GPU-hour
$0$2$502 Nov14 Nov 2026

AWSAzureOracleCoreWeaveCrusoeRunPod SecureRunPod CommunityVast.ai median

MI300X: US$ per GPU-hour
$0$3$702 Nov14 Nov 2026

OracleCrusoeRunPod Secure

L40S: US$ per GPU-hour
$0$2$402 Nov14 Nov 2026

AWSOracleCoreWeaveNebiusCrusoeRunPod SecureRunPod CommunityVast.ai median

RTX PRO 6000: US$ per GPU-hour
$0$2$502 Nov14 Nov 2026

OracleCoreWeaveNebiusRunPod SecureRunPod CommunityVast.ai median

RTX 5090: US$ per GPU-hour
$0$1$202 Nov14 Nov 2026

RunPod SecureRunPod CommunityVast.ai median

RTX 4090: US$ per GPU-hour
$0$0$102 Nov14 Nov 2026

RunPod SecureRunPod CommunityVast.ai median

How these prices are measured

  • Per GPU-hour. Clouds rent whole machines, so a machine's hourly price is divided by the GPUs in it (for example AWS p5.48xlarge, 8 × H100, ÷ 8). Compute only: storage, data transfer and support are extra and differ by provider.
  • Hyperscalers. AWS (US East, N. Virginia) and Azure (East US) on-demand Linux prices; Oracle Cloud's price list, which prices GPUs per GPU-hour.
  • Neoclouds (clouds built for GPUs). CoreWeave's North America on-demand price for an 8-GPU machine, ÷ 8; Lambda's on-demand instances (the lowest per-GPU price of its 1, 2, 4 and 8-GPU sizes, usually the 8-GPU one); Nebius and Crusoe on-demand per GPU-hour; RunPod Secure Cloud (its own data centres). Read daily from each provider's public pricing page.
  • Not shown. Committed-use, reserved and cluster prices are usually lower; spot and preemptible prices are lower but can be interrupted. "Contact sales" models have no public price.
  • Marketplaces. RunPod Community Cloud (third-party hosts, list price). Vast.ai lists machines offered by many independent hosts worldwide; the median and the lowest per-GPU price of the rentable on-demand offers that day, with the number of offers. Few offers means the figure can swing.
  • Checks. A list price that moves more than 50% in a day is held at the previous value until checked.
GPUAWS instanceAzure VM
B300p6-b300.48xlarge ÷ 8–
B200p6-b200.48xlarge ÷ 8–
H200p5en.48xlarge ÷ 8–
H100 SXMp5.48xlarge ÷ 8Standard_ND96isr_H100_v5 ÷ 8
H100 PCIe / NVL–Standard_NC40ads_H100_v5 ÷ 1
A100 80GBp4de.24xlarge ÷ 8Standard_ND96amsr_A100_v4 ÷ 8
MI300X––
L40Sg6e.xlarge ÷ 1–
RTX PRO 6000––
RTX 5090––
RTX 4090––

The Week in Taiwan Companies · free, every Friday · only in the email: next week's calendar, foreign-investor streaks, a Japan watch, the chart of the week and the earnings-growth list
Using it at work? Join the team-plan waiting list

After subscribing, check your inbox (and spam folder) and confirm your subscription. We do this to protect you from spammers.

Generated 2026-11-14 00:29.