NVIDIA B300 GPU prices

Compare source-linked B300 offers for high-memory Blackwell Ultra deployments. Treat delivery timing, topology, and contract scope as part of the price.

Dataset checked 2026-09-08. Prices and availability may change.

Verified records4
Lowest normalized$6.97per GPU-hour
Median normalized$7.79per GPU-hour
Highest normalized$15.00per GPU-hour

Current public offers

Source-linked records with the published billing context preserved.

Provider and nodeRegionTermNormalized priceSource
Gcore8x B300 SXMNot publishedhourly$6.97/GPU-hr48.00Provider source ↗Checked 2026-09-04
DataPacket8x B300Not published3–6-mo prepay typ.$7.57/GPU-hr38085Provider source ↗Checked 2026-09-08
Latitude.sh8x B300 (g4.b300.large)Not publishedno commitment$8.00/GPU-hr$64.00/hr ($46,720/mo)Provider source ↗Checked 2026-08-20
Oracle Cloud8x B300 (BM.GPU.B300.8)Select OCI regionsno commitment$15.00/GPU-hr$15.00/GPU/hr (pay as you go)Provider source ↗Checked 2026-08-20

Plan around delivered capacity

B300 is suited to workloads that can use its larger memory footprint and newer platform capabilities. Compare completed-workload economics using the intended model, precision, batch size, serving stack, and cluster topology.

Validate early-market offers

Confirm the exact system, quantity, region, delivery date, minimum term, support boundary, software image, and evidence path. Record which fields are provider-confirmed and which remain public-source observations.

Request matching capacity

Send one specification. We will use it to identify matching public and unlisted supply.

Non-binding. By submitting, you agree to the Terms of Use and Privacy Policy.

Questions buyers ask

What does a B300 price include?

It depends on the source. Review whether the rate covers a full node or individual GPU, the commitment, network, storage, support, and setup fees.

How should a buyer compare B300 and older GPUs?

Use model-specific throughput and memory requirements to calculate completed-workload cost rather than comparing hourly rates alone.