Best Neocloud / B300 and GB300 providers
B300 and GB300 NVL72 providers
Every provider with a public Blackwell Ultra offer, linked to its own page and dated. Plus how HGX B300, DGX B300 and GB300 NVL72 differ.
As of 8 November 2026, CoreWeave and Nebius publicly offer HGX B300, Lambda and Crusoe list both GB300 NVL72 and HGX B300, and TD SYNNEX resells dedicated B300 clusters on Nebius. Each claim below links the provider's own page.
B300 and GB300 cloud providers
| Provider | Platforms | Status | Source |
|---|---|---|---|
| CoreWeaveAI cloud, reserved clusters | HGX B300 | Generally available GA announced 16 Mar 2026. Ask which fabric the allocation uses and whether every node delivers together. |
coreweave.com 22 Oct 2026 |
| NebiusAI cloud, self-service and reserved | HGX B300 | In production Self-service instances plus a sales route for large reservations. Ask for a named region and one connected allocation. |
nebius.com 22 Oct 2026 |
| LambdaSuperclusters, 1-Click Clusters, instances | GB300 NVL72HGX B300 | Listed on site Single-tenant GB300 NVL72 superclusters on InfiniBand. Ask about minimum cluster size and lead time. |
lambda.ai 8 Nov 2026 |
| CrusoeAI cloud and modular data centers | GB300 NVL72HGX B300 | Through sales Both platforms sold through the sales team. Ask which campus the capacity sits in and the start date. |
crusoe.ai 8 Nov 2026 |
| TD SYNNEXReseller channel to Nebius | HGX B300 | Channel route Dedicated B300 clusters on Nebius AI Cloud, announced 16 Apr 2026. Same capacity as Nebius, so not a separate fallback. |
tdsynnex.com 22 Oct 2026 |
| Together AIGPU clusters and inference | TBC | Researching No public B300 or GB300 source confirmed yet. | — |
| Voltage ParkBare-metal GPU cloud | TBC | Researching No public B300 or GB300 source confirmed yet. | — |
| FluidstackDedicated large clusters | TBC | Researching No public B300 or GB300 source confirmed yet. | — |
A public offer means the provider says it sells the platform. It doesn't confirm capacity in your region, at your size or by your start date.
HGX B300, DGX B300 and GB300 NVL72
All three use NVIDIA's Blackwell Ultra GPU, but they're different machines. Make sure every quote names the same one.
| System | What it is | GPUs | Good for |
|---|---|---|---|
| HGX B300 | 8-GPU server board that providers build into their own servers, with x86 CPUs | 8 per node | Most training and inference clusters |
| DGX B300 | NVIDIA's own complete 8-GPU B300 server | 8 per node | Buyers who want NVIDIA's reference build |
| GB300 NVL72 | A liquid-cooled rack: 72 Blackwell Ultra GPUs and 36 Grace CPUs on one NVLink domain | 72 per rack | Frontier training, very large model inference |
Each B300 GPU has 288 GB of HBM3e memory. On GB300 NVL72, all 72 GPUs can talk over NVLink as one large GPU, which helps the biggest models. HGX B300 clusters link nodes over InfiniBand or Ethernet instead.
How big a B300 cluster can you get?
Providers rarely publish free capacity. In our experience, single nodes and small clusters are easy to find; 128+ connected nodes (1,024+ GPUs) in one facility needs a sales conversation and usually a start date weeks or months out. GB300 NVL72 is sold by the rack, so the minimum is 72 GPUs.
Questions to ask a B300 provider
Which exact system is it?
Is it one connected cluster?
What fabric and storage?
Is it running today?
Who's accountable when it breaks?
What does the whole term cost?
More on B300 sourcing
RackShore, who run this site, keep a longer guide to B300 suppliers and channels, including OEM routes such as Supermicro.
Questions
Who has B300 GPUs for rent?
Is B300 the same as GB300?
How much memory does a B300 have?
How much does B300 rental cost?
RackShore runs Best Neocloud and may be paid a referral fee by a provider. That doesn't change which providers are listed or how they're described. Spot a mistake? Tell us.