NVIDIA B300 Waitlist
Reasoning inference · frontier training
Expected early 2027Quoted on reservation. No list price.
Join the waitlist Chat on WhatsAppReselling GPUs to your own customers? Partner pricing on request.
Join the waitlist. We confirm the system, location, price and date in writing before you commit to anything.
Capacity is not live yet. No price until we quote. Nothing is binding until you sign.
Businesses we’ve hosted and supported since 2009
Reasoning inference · frontier training
Expected early 2027Quoted on reservation. No list price.
Join the waitlist Chat on WhatsAppReselling GPUs to your own customers? Partner pricing on request.
NVIDIA lists 288GB of HBM3e per GPU at 8TB/s, 15 PFLOPS dense NVFP4 (20 sparse), NVLink 5 at 1.8TB/s per GPU and up to 1,400W.
Capacity is not live yet. Your written offer states the system, GPU count, location, reported memory, price and date.
| GPU | Memory | Availability | Choose it when |
|---|---|---|---|
| NVIDIA H200Ready now | 141GB | Available now | Your model fits in 141GB and you need capacity now. |
| NVIDIA B200On request | 180GB | Quoted within one business day | You need Blackwell, FP4 or 180GB per GPU this year. |
| NVIDIA B300Waitlist | 288GB | Expected early 2027 | You need 288GB per GPU or Blackwell Ultra, and can plan for early 2027. |
All eight-GPU HGX systems. NVIDIA rates Blackwell Ultra at 1.5x the dense FP4 of Blackwell, with the same 1.8TB/s NVLink. All GPUs on the GPU cloud page.
The model or workload, the GPU count and when you need it.
The system, location, price and date, once our partners confirm capacity.
Nothing is binding until you sign. Change or decline the offer at any time.
Large models and long context on Hopper. Test a node first.
Blackwell in HGX systems. Written quote within one business day.
Hopper for training and inference that fits in 80GB.
Visual AI, rendering and single-card inference up to 96GB.
Your waitlist place stays while you run on another card.
We quote the B300 on reservation and publish no list price. Capacity is expected in early 2027. Join the waitlist with your GPU count and timing; we send the system, location, price and date in writing, and nothing is binding until you sign.
Both are Blackwell GPUs in eight-GPU HGX systems. NVIDIA lists 288GB of HBM3e per B300 GPU against 192GB on B200 (about 180GB visible to software), 1.5x the dense FP4 compute and up to 2x faster attention. NVLink stays at 1.8TB/s per GPU.
NVIDIA lists 288GB of HBM3e per Blackwell Ultra GPU at 8TB/s. Its HGX B300 spec table lists 2.1TB for eight GPUs, so the figure your software reports sits a little below 288GB. We state the exact system and reported memory in writing.
HGX B300 is an eight-GPU system with x86 hosts that fits a standard rack row, and the form we expect to offer first. GB300 NVL72 is a liquid-cooled rack of 72 GPUs and 36 Grace CPUs for the largest clusters. Tell us the scale and we advise.
Start now if your workload fits today: H200 is ready now and B200 is on request, both with a written quote within one business day. Wait only if you need 288GB per GPU or Blackwell Ultra FP4. You can join the waitlist and run on H200 meanwhile.
Nothing. It tells us your model, GPU count and timing so we can line up capacity with our partners and come back with a written offer. You accept, change or decline it. No deposit, no term, nothing binding until you sign.