Compute Screener
Updated 10:29 PM

GPU rental prices and AI token prices

# GPU/Bloomberg ticker USD / GPU-hour 1D 1W 1M Price Range Chart (1M)
1$9.18+3.9%+3.9%+18.3%$6.60–$11.09
2$7.83+0.6%+0.6%+23.9%$5.43–$9.23
3$5.10-0.3%+2.3%+12.3%$3.45–$6.31
4$3.69-0.6%-0.4%+3.9%$2.58–$6.16
5$2.020.0%+4.2%+7.7%$1.61–$2.50
6$1.73+0.4%+0.5%-10.4%$1.12–$3.20

B300

$9.18▲33.4% (All)
USD / GPU-hour
1D
1W
1M
3M
6M
YTD
1Y
+3.9%
+3.9%
+18.3%
+18.9%
+28.9%
—
—
#Exchange24h volumePrice
1Architect AX—Institutional accounts onlyListedView →
#ProviderAs ofUSD / GPU-hour
1Massed ComputeOct 24, 10:29 PM$6.60Rent →
2HyperstackOct 24, 10:29 PM$7.40Rent →
3RunPodOct 24, 10:29 PM$7.97Rent →
4Latitude.shOct 24, 10:29 PM$8.00Rent →
5LiumOct 24, 10:29 PM$8.78Rent →
6NebiusOct 24, 10:29 PM$9.50Rent →
7Together AIOct 24, 10:29 PM$9.99Rent →
8Verda (DataCrunch)Oct 24, 10:29 PM$10.65Rent →
9Vast.aiOct 24, 10:29 PM$11.09Rent →

Break-even throughput by model

#ModelUSD / 1M tokensTokens/s per GPU
1Claude Fable 5.1Anthropic$20.00127
2GPT-6 AstraOpenAI$20.00127
3Claude Opus 5Anthropic$10.00255
4Claude Opus 5.5Anthropic$8.00319
5GPT-5.6 SolOpenAI$8.00319
6Kimi K3Moonshot AI$6.00425
7Claude Sonnet 5.5Anthropic$4.00637
8Claude Sonnet 5Anthropic$4.00637
9GPT-6.1 SolOpenAI$4.00637
10Gemini 4 ArgonGoogle$4.00637
11Kimi K3.1 ProMoonshot AI$4.00637
12Grok 4.7xAI$3.00850
13GLM 5.3Z.ai$2.151,186
14GLM 5.2Z.ai$2.151,186
15Gemini 3.8 FlashGoogle$1.501,700
16Gemini 3 Flash PreviewGoogle$1.132,266
17DeepSeek V4 Pro 0813DeepSeek$0.992,576
18Qwen3.8 27BAlibaba$0.962,666
19Gemini 3.1 Flash LiteGoogle$0.564,533
20MiMo-V2.6-ProXiaomi$0.544,689
21GPT-5.6 LunaOpenAI$0.455,666
22DeepSeek V4.1 Flash 1017DeepSeek$0.269,714
23GLM 5.3 FlashZ.ai$0.2410,736
24Qwen3.8 FlashAlibaba$0.2311,086
25Claude Haiku 5.5Anthropic$0.2012,749
26GPT-6 LunaOpenAI$0.2012,749
27GPT-6 Luna ProOpenAI$0.2012,749
28MiMo-V2.6-FlashXiaomi$0.1814,570
30Qwen3.7 FlashAlibaba$0.0646,360

About NVIDIA B300

The NVIDIA B300 rental price today is $9.18 per GPU-hour, up 3.9% since yesterday, up 3.9% over the past week and up 18.3% over the past month. At that rate one B300 costs about $6,700 for a month of 730 hours. Since Apr 2, 2026 it has ranged from $6.88 to $9.18. The index is the price of one benchmark contract: B300 SXM on HGX, on-demand, 8-GPU node, US hardware, CPU-neutral. It is standardized across 9 cloud providers and marketplaces, from Massed Compute at $6.60 to Vast.ai at $11.09, and is updated every 2 to 3 hours.

The B300 is NVIDIA's Blackwell Ultra GPU, announced on March 18, 2025 and built for reasoning models, which spend more compute generating each answer. Against the B200 it has 1.5 times the memory and 1.5 times the FP4 compute, and twice the speed on attention, the step that dominates long contexts. It ramped quickly: by late 2025 the GB300 version brought in about two-thirds of NVIDIA's Blackwell revenue.

Specifications

SpecB300 in HGX
ArchitectureBlackwell Ultra, two dies, TSMC 4NP, 208 billion transistors
Memory288 GB HBM3e, about 263–270 GB usable in HGX servers; 8 TB/s
FP4 tensorabout 13.5 PFLOPS dense per GPU, 1.5 times the B200
FP8 tensorabout 4.5 PFLOPS dense per GPU, the same as the B200
FP64about 1.25 TFLOPS per GPU, against about 37 on the B200
NVLink1.8 TB/s, NVLink 5
Powerup to 1,400 W per GPU for Blackwell Ultra, NVIDIA's maximum, in GB300 systems; about 14 kW for an 8-GPU DGX B300
Other versionsGB300 NVL72: 72 GPUs and 36 Grace CPUs in a liquid-cooled rack

Per-GPU figures are NVIDIA's 8-GPU HGX totals divided by eight, from its HGX specifications and Blackwell Ultra blog post, accessed October 24, 2026.

Where the B300 is rented

Microsoft built the first large GB300 cluster, more than 4,600 GPUs for OpenAI, announced on October 9, 2025. AWS P6-B300 machines became available on November 18, 2025, at first as reserved capacity. Oracle, Nebius, Crusoe, Together AI and CoreWeave also rent B300s; GetDeploying listed 36 providers on October 23, 2026.

What this price measures

The index is the posted on-demand price of one benchmark contract, a B300 in an 8-GPU HGX node of US hardware, rented by the hour from GPU clouds and marketplaces; prices for other node sizes are converted to that contract, and prices from other regions are counted without a regional adjustment for now. It is not a hyperscaler, reserved or spot price, and it does not cover GB300 rack-scale systems. Few providers post B300 prices yet, so each one carries more weight than in the H100 index.

On October 23, 2026, B300s were listed at $9.50 per GPU-hour at Nebius, $9.90 at Crusoe, $9.99 at Together AI, $15.00 at Oracle and $17.80 at AWS, with spot capacity at CoreWeave for $4.48 and reserved terms from $4.25.

What moves B300 rental prices

Price history

The B300 has been rented publicly for less than a year. Its median posted on-demand price was $9.35 per GPU-hour on October 23, 2026, about 12% higher than 90 days earlier on GetDeploying's page; its weekly provider median rose about 15% over the same period. In September 2026 another tracker put the median at $7.87 on demand and $5.71 reserved; trackers sample different providers, so their levels differ. This index's history before October 23, 2026 is GetDeploying's weekly B300 series, rescaled to join the daily index.

Futures and benchmarks

No B300 futures contract has been filed; CME's filed contracts and ICE's contracts set for listing cover the H100 and B200. See the B200 page.

FAQ

How does the B300 compare with the B200 on price? It cost about 7% to 25% more per hour in October 2026, depending on the source: $15.00 against $14.00 at Oracle, $17.80 against $14.24 at AWS and $9.35 against $7.99 on GetDeploying's medians. For that it has 1.5 times the memory and FP4 compute.

Why are there fewer B300 prices than H100 prices? It is newer, and providers sell much of their capacity on contracts rather than by the hour. The index needs at least five providers to publish.

How often is this index updated? Every 2 to 3 hours, with one value per day in US Central time that is final when the day closes.

Sources

Text reviewed Oct 24, 2026; the figures in it are as of that date. The figures at the top of this page update every 2 to 3 hours. Not investment advice.