In October 2026 the same NVIDIA B300 lists at $17.80 per GPU-hour on demand at AWS and $15.00 at Oracle, against $9.50 to $9.99 at Nebius, Crusoe and Together AI; a three-year AWS EC2 Instance Savings Plan brings AWS down to $6.69 to $7.69. GB300 NVL72 has a public price only at Oracle, $18.00 per GPU-hour. Reserved B300 through Ancapex for two years or longer costs about $5.40 per GPU-hour at the median, our fee included.
Prices below are as each provider's own pricing page, price file or price API showed them on 20 October 2026. Where a provider prices a server or a VM, we divide by its GPU count; a GPU-hour at different providers comes with different CPUs, memory and network, so these are list prices, not exact equivalents.
| Provider | Offer | Per GPU-hour |
|---|---|---|
| AWS price list | p6-b300.48xlarge on demand, $142.416 an hour, US East | $17.80 |
| AWS Capacity Blocks | Capacity Block, $129.168 an hour (AWS prints $16.146 per GPU) | $16.15 |
| AWS Savings Plans | EC2 Instance Savings Plan, 3 years, all upfront to no upfront | $6.69 to $7.69 |
| Oracle | BM.GPU.B300.8, pay as you go | $15.00 |
| Together AI | GPU clusters, on demand (preemptible $4.99) | $9.99 |
| Crusoe | HGX B300 on demand | $9.90 |
| Nebius | HGX B300 on demand | $9.50 |
| CoreWeave | HGX B300 spot, $35.84 an hour in North America; on demand: contact sales | $4.48 (spot) |
Azure and Google Cloud have no HGX B300 machine with a public price; Lambda does not list B300.
| Provider | Offer | Per GPU-hour |
|---|---|---|
| AWS price list | p6-b200.48xlarge on demand, $113.9328 an hour | $14.24 |
| AWS Capacity Blocks | Capacity Block, $113.666 an hour | $14.21 |
| AWS Savings Plans | EC2 Instance Savings Plan, 3 years | $5.35 to $6.15 |
| Oracle | BM.GPU.B200.8, pay as you go | $14.00 |
| Google Cloud | a4-highgpu-8g: Flex-start $64.44 an hour; 3-year committed use $56.7072 | $8.06; $7.09 |
| CoreWeave | HGX B200 on demand, $68.80 an hour | $8.60 |
| Nebius | HGX B200 on demand | $8.50 |
| Crusoe | B200 on demand | $8.50 |
| Together AI | GPU clusters, on demand; 91 to 180 days $6.79 | $8.19 |
| Lambda | B200 on demand, 8 GPUs; 1-Click Clusters of 16 GPUs for 2 weeks to 1 year $9.86 | $6.69 |
| Provider | Offer | Per GPU-hour |
|---|---|---|
| Oracle | BM.GPU.GB300.4, pay as you go | $18.00 |
| Oracle | BM.GPU.GB200.4, pay as you go | $16.00 |
| AWS Capacity Blocks | u-p6e-gb200x72 Capacity Block, $761.904 an hour for 72 GPUs, Dallas Local Zone | $10.58 |
| CoreWeave | GB200 NVL72, $42.00 an hour for an instance of 4 GPUs | $10.50 |
| Azure retail prices | ND GB200 v6, 4 GPUs, East US: on demand / 1-year / 3-year reservation | $27.04 / $17.31 / $11.90 |
GB300 NVL72 has no public price at AWS (P6e-GB300 UltraServers through the sales team, AWS, 2 December 2025), Azure (no ND GB300 v6 price in its price API), Google Cloud (A4X Max by reservation only), CoreWeave, Nebius, Together AI or Crusoe. Rack-scale capacity is mostly sold on contract.
Earlier prices come from archived copies of each provider's page in the Internet Archive.
Through Ancapex, B300 on terms of two years or longer costs about $5.40 per GPU-hour at the median of 15 priced operator offers, with the middle half between $4.90 and $6.10, our fee included, and can start from Q4 2026. Operators quote reserved capacity for one to five years, often with 25% to 30% of the contract paid up front. See the B300 Price Index, reserved vs on-demand B300 and what buying a B300 server costs.
Requirements are handled confidentially and shared with operators only on a need-to-know basis, never with your name attached before you approve it. We reply within one business day from dima@ancapex.io.
p6-b300.48xlarge costs $142.416 an hour on demand in US East, $17.80 per GPU-hour; a Capacity Block is $16.146 per GPU-hour, and a three-year EC2 Instance Savings Plan $6.69 to $7.69 (AWS, 20 October 2026).
Azure publishes no price for ND GB300 v6 in its price API. Its ND GB200 v6 lists at $27.04 per GPU-hour on demand in East US, $17.31 on a one-year and $11.90 on a three-year reservation.
Google Cloud has no HGX B300 machine type with a public price. Its A4 machines carry B200, and A4X Max (GB300) is available by reservation only.
CoreWeave lists HGX B300 on demand as contact sales and its spot price at $35.84 an hour in North America, $4.48 per GPU-hour; it states discounts of up to 60% for committed usage.
Among published on-demand prices on 20 October 2026: Nebius at $9.50 per GPU-hour, then Crusoe at $9.90 and Together AI at $9.99; AWS is $17.80 and Oracle $15.00.
Only Oracle publishes it: $18.00 per GPU-hour, pay as you go. Other providers sell GB300 NVL72 through their sales teams or by reservation.
