In October 2026 on-demand B300 lists at a median of $8.99 per GPU-hour across public price pages (getdeploying), from $7.10 to $9.99 at the specialist clouds below and $15 or more at AWS and Oracle. Reserved B300 on terms of two years or longer costs about $5.40 per GPU-hour at the median of offers through Ancapex, our fee included (19 October 2026): about 40% less. The lower rate comes with a term, a prepayment and a start date weeks or months away.
On-demand is priced for flexibility: start today, stop tomorrow. Reserved capacity is priced for commitment: a block of GPUs for a few months to five years, paid whether you use it or not.
| Provider | B300, per GPU-hour | Source |
|---|---|---|
| Together AI | $9.99 | together.ai/pricing |
| Nebius | $9.50 | nebius.com/prices |
| RunPod | $8.99 (Secure Cloud) | runpod.io/pricing |
| Modal | $7.10 ($0.001972 per second) | modal.com/pricing |
| Oracle, AWS | $15.00, $17.80 | getdeploying.com |
| Median across 16 providers | $8.99 | getdeploying.com |
List prices seen on 20 October 2026; providers change them often. CoreWeave and Lambda showed no public on-demand B300 price that day.
Through Ancapex, our fee included, on terms of two years or longer: median about $5.40 per GPU-hour, middle half $4.90 to $6.10 (15 priced offers from 10 operators, as of 19 October 2026). Large clouds publish their own commitment discounts: CoreWeave offers "up to 60%" off on-demand for committed usage (CoreWeave), and Nebius "up to 35% less than on-demand" for clusters reserved for several months (Nebius). Current numbers: GPU availability.
A reservation is paid for every hour of the term. On-demand is paid only for the hours you run. At $5.40 reserved against $8.99 on-demand, the reservation costs less once your GPUs would be busy more than about 60% of all hours over the term. Training runs, steady inference and research clusters usually are; occasional experiments usually are not.
Two quotes at the same price per GPU-hour can differ in what they include. Check electricity, cooling, on-site operations, egress, storage, the SLA, the prepayment and when the term starts. See the lease checklist.
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.
On-demand: a median of $8.99 per GPU-hour on public price pages, $7.10 to $9.99 at Modal, RunPod, Nebius and Together AI and $15 or more at AWS and Oracle (20 October 2026). Reserved for two years or longer: about $5.40 at the median through Ancapex, our fee included.
Per hour, yes: about 40% less for B300 at current medians. You pay for every hour of the term, so it is cheaper overall when the GPUs are busy more than about 60% of the time.
Offers in our pipeline run from a few months to five years; 1 to 4 years is the most common. Terms of 12 months or shorter cost more per hour.
Often 25% to 30% of the contract value at signing in our offers.
Reserved capacity is priced per deal: term, volume, start date and prepayment all move the rate. Most operators quote it only after an NDA.
