NVIDIA B300 price: rental per GPU-hour and server cost
Series 1 · GPU prices · NVIDIA · snapshot 7 Nov 2026
Blackwell Ultra B300 rents for $9.50 per GPU-hour at the median published on-demand rate, +38% against the B200, and an eight-GPU HGX B300 server is listed for $599,995 at the median. The spread between sellers is wider than the gap between the two generations.
Figure 1B300: recorded snapshots, USD per GPU-hour
| Snapshot | Median | Floor | Ceiling | Reserved 12 mo |
|---|---|---|---|---|
| 2026-11-07 | $9.50 | $5.89 | $17.80 | $7.22 |
Price trend: not enough data yet — 1 dated snapshot, and a trend needs at least 2. The chart replaces this table from the next snapshot of the B300 series.
Reading the B300 market
11 on-demand list prices define the range. DeepInfra's dedicated B300 at $5.89 is the floor, followed by a cluster of GPU clouds between seven and ten dollars: Hyperstack $7.40, Civo $7.49 (an eight-GPU instance split per GPU), Crusoe $8.96, Runpod Secure Cloud $8.99, Nebius $9.50 and Together $9.99. Verda sits a step higher at $11.69.
The top of the range is set by large clouds selling whole eight-GPU machines: Oracle's BM.GPU.B300.8 and Fireworks both work out to $15.00 per GPU, and AWS's p6-b300.48xlarge in US East to $17.80 — 1.9× the median. Interruptible capacity runs far below every on-demand list: CoreWeave's spot node works out to $4.48 per GPU, Together's preemptible tier $4.99, DigitalOcean spot $6.17. Those rates are not counted in the on-demand range because the capacity can be reclaimed.
Commitment is where B300 pricing differs most from older parts. Civo publishes a full ladder for an eight-GPU node — $6.49 for 12 months, $5.99 for 24 and $5.49 for 36 per GPU-hour — and DigitalOcean lists $7.94 for a 12-month term. The index carries $7.22 as the 12-month figure, 24% under the on-demand median: a deeper published discount than the H100 or H200 get, because B300 capacity is still being filled.
Table 1B300 prices by provider, USD per GPU-hour
| Provider | Offer | USD / GPU-hr | Detail | As of |
|---|---|---|---|---|
| CoreWeave | Spot | $4.48 | NVIDIA HGX B300 8-GPU node, North America, spot $35.84/node-hr ÷ 8 GPUs. On-demand = Contact sales | 2026-11-07 |
| Together AI | Spot | $4.99 | NVIDIA HGX B300 'Preemptible Compute' column (pay as you go), per GPU-hour; reserved = Contact us | 2026-11-07 |
| Civo | Reserved, 36 mo | $5.49 | 8 x NVIDIA B300, 36-month commitment $43.92/hr ÷ 8 | 2026-11-07 |
| Verda (ex-DataCrunch) | Spot | $5.85 | 1x B300 SXM6 268GB spot instance | 2026-11-07 |
| DeepInfra | On-demand | $5.89 | Dedicated B300 270GB GPU for custom LLM deployments, billed per minute | 2026-11-07 |
| Civo | Reserved, 24 mo | $5.99 | 8 x NVIDIA B300, 24-month commitment $47.92/hr ÷ 8 | 2026-11-07 |
| DigitalOcean | Spot | $6.17 | NVIDIA HGX B300 spot GPU Droplet, interruptible, 'Current Spot Price' | 2026-11-07 |
| Civo | Reserved, 12 mo | $6.49 | 8 x NVIDIA B300, 12-month commitment $51.92/hr ÷ 8 (6-month commitment shown N/A) | 2026-11-07 |
| Runpod | Marketplace | $6.94 | 1x B300 pod, Community Cloud (third-party hosts) | 2026-11-07 |
| Modal | Serverless | $7.10 | Serverless GPU, billed per second: $0.001972/sec × 3600 | 2026-11-07 |
| Hyperstack | On-demand | $7.40 | NVIDIA B300 288GB, on-demand per GPU-hour (28 pCPU, 240GB RAM per GPU). No B300 reservation price listed | 2026-11-07 |
| Civo | On-demand | $7.49 | Extra Large 8 x NVIDIA B300 instance: $59.92/hr ÷ 8 GPUs. Row columns in order: On-demand, 6mo, 12mo, 24mo, 36mo commitment | 2026-11-07 |
| DigitalOcean | Reserved, 12 mo | $7.94 | HGX B300, 12-month reserved | 2026-10-20 |
| Crusoe | On-demand | $8.96 | NVIDIA B300 288GB HGX, current on-demand per GPU-hr. Page flattens table; value matched to row by column order (cross-checked twice) | 2026-11-07 |
| Runpod | On-demand | $8.99 | B300 288GB pod, listed rate | 2026-10-20 |
| Nebius | On-demand | $9.50 | HGX B300, per GPU-hour | 2026-10-20 |
| Together AI | On-demand | $9.99 | NVIDIA HGX B300, GPU Clusters on-demand, per GPU-hour | 2026-11-07 |
| Verda (ex-DataCrunch) | On-demand | $11.69 | 1x B300 SXM6 268GB VRAM instance (Verda lists 268GB usable), 30 CPU, 255GB RAM | 2026-11-07 |
| Oracle Cloud (OCI) | On-demand | $15.00 | BM.GPU.B300.8 (8x B300 263GB), pay-as-you-go per GPU-hour, Oracle GPU price list (part B112237) | 2026-11-02 |
| Fireworks AI | On-demand | $15.00 | B300 288 GB on-demand deployment GPU rate ($0.250/min) | 2026-11-07 |
| AWS | On-demand | $17.80 | p6-b300.48xlarge (8x B300), US East (N. Virginia), Linux on-demand $142.416/hr ÷ 8. AWS EC2 on-demand price list | 2026-11-07 |
Rows are the prices recorded in the tracker note for this model; 21 of the 21 come from the provider's own price page, the rest through comparison trackers. Third-party index rows are shown for context and not counted as providers. Every page behind this table: Sources, B300.
The B300 quotes and history are in the index download: JSON · CSV — free to reuse under CC BY 4.0; credit “Compute Census (computecensus.com)” with a link. Licence terms
Specifications
Table 2DGX B300 system (8 GPUs); per-GPU memory = system total / 8
| Memory per GPU (index) | 288 GB |
| GPUs per system | 8x NVIDIA Blackwell Ultra SXM |
| GPU memory, system | 2.1 TB total (~288 GB per GPU) |
| FP4 Tensor Core, system | 144 PFLOPS sparse | 108 PFLOPS dense |
| FP8 Tensor Core, system, with sparsity | 72 PFLOPS |
| NVLink, system | 14.4 TB/s aggregate |
| Networking | 8x ConnectX-8 VPI, up to 800 Gb/s InfiniBand/Ethernet |
| CPU | Intel Xeon 6776P |
| System power | ~14 kW |
| Form factor | 10U |
Quoted from NVIDIA — DGX B300, checked 7 Nov 2026. Vendor tensor figures are "with sparsity" unless stated; dense throughput is half.
What moves the B300 price
- Memory: 288 GB of HBM3e per GPU in the index (NVIDIA lists 2.1 TB for an eight-GPU DGX B300), against 180 GB for the B200 — the reason long-context inference moves to this part first.
- Low-precision compute: NVIDIA rates the DGX B300 at 108 PFLOPS of dense FP4 for eight GPUs; workloads that can use FP4 get the most out of the premium, FP8 training much less.
- Power density: about 14 kW per eight-GPU system on NVIDIA's DGX page, in a 10U chassis. HGX servers ship in air-cooled 8U and liquid-cooled versions, and the cooling choice decides which halls can host them.
- Few large-cloud listings: Azure and Google publish no B300 price yet, and CoreWeave, Nebius and Crusoe sell clusters by quote above a certain size, so volume buyers rarely pay the list rates shown here.
Rent or buy
Rented around the clock, one eight-GPU B300 node costs $55,480 a month at the median, $32,062 on Civo's 36-month term and $103,952 at the AWS rate.
Bought, the same node is a capital item: 7 dated listings for complete HGX B300 servers run from $518,139 to $827,308, median $599,995, and NVIDIA's own DGX B300 is listed from $734,654 (plus euro-priced listings) (Table 3). The median server equals about 10.8 months of median-rate rental, or 18.7 months at Civo's three-year rate — before roughly 14 kW of power and cooling per node, networking and the hosting site. Listings are starting configurations; memory, storage, NICs and support terms move the final quote by a wide margin.
Table 3B300 purchase prices from dated public listings, USD
| Configuration | Listings | Low | Median | High | Seller type | Checked |
|---|---|---|---|---|---|---|
| Eight-GPU server (HGX) e.g. Supermicro SYS-822GS-NB3RT HGX B300 8-GPU 8U air-cooled · Supermicro SYS-822GS-NB3RT-01-G2 Gold Series HGX B300 8-GPU 8U | 7 | $518,139 | $599,995 | $827,308 | reseller, system builder | 2026-11-07 |
| NVIDIA DGX system e.g. NVIDIA DGX B300 · NVIDIA DGX B300 2.1TB | 2 | $734,654 | fewer than 3 listings | $1,300,478 | system builder, reseller | 2026-11-07 |
List prices published by resellers and system builders on the dates shown, before volume discounts; tax and shipping as listed by each seller. We record every listing with its page and a quote of the price, but do not name or link sellers: availability and final terms differ for each order, so the request form below returns current offers for your configuration.
Where 14 kW-per-node Blackwell systems can be hosted: AI data center census.
Other models with their own page: GB200, B200, MI355X, H200, H100, MI300X, A100, L40S, RTX 4090, GB300. All models: GPU rental price index. How the figures are built: methodology.