B200 / B300 · NVIDIA Blackwell
Liquid-cooled Blackwell,
reserved for your run.
B200 and B300 in NVL-class, direct-to-chip liquid-cooled racks for frontier training and large-scale inference. Reserved and built to order, backed by an open purchase order with Supermicro — brought online as commitments land.
- GPU
- B200 / B300
- Memory class
- ~192 GB HBM3e
- Cooling
- Direct-to-chip liquid
- Path
- Reserved, built to order
01Hardware
The densest GPUs NVIDIA ships, on paper.
Direct-to-chip liquid cooling for the parts air can no longer handle — committed capacity, built to order and brought online as reservations land.
| Component | Specification | Notes |
|---|---|---|
| GPU | NVIDIA B200 / B300 | top of the Blackwell line |
| GPU memory | ~192 GB HBM3e-class | per GPU |
| Racks | NVL-class, high-density | direct-to-chip liquid cooling |
| Procurement | Open Supermicro purchase order | systems built to order |
| Allocation | Reserved & committed | term pricing, scoped per cluster |
02The reservation path
Start on managed capacity. Reserve Blackwell to order.
Start on managed capacity today
On-demand GPUs run open-weights models now through an OpenAI-compatible endpoint, billed by the second — no reservation required while your Blackwell block is scoped.
Reserve a Blackwell block
Commit to B200 / B300 capacity and we scope the cluster, the direct-to-chip cooling build, and the timeline against our open purchase order with Supermicro.
Built to order, brought online
Reservations convert into racked, running Blackwell as commitments land — no speculative inventory, no stranded capacity, and no claims we can't stand behind.
Or colocate your own systems
Bring your own B200 / B300 systems into liquid-ready racks as our halls come online. You run the fleet; we handle power, cooling, and networking.
03Who it's for
Training-heavy teams and committed inference.
Frontier labs. Pre-training and large-scale fine-tuning on dedicated, liquid-cooled Blackwell reserved to you — not a marketplace slot that vanishes mid-run.
Large enterprises. In-house foundation models on committed capacity, with a named point of contact and reserved pricing.
Liquid colocation. Bring your own B200 / B300 systems into our liquid-ready racks as halls come online — power and cooling handled.
Where this is going
Spec the cluster. Buy it through us.
Few teams hold Supermicro or vendor accounts. We do — an open purchase order that backs every reservation, so a commitment converts into racked, running Blackwell on a predictable path.
We're building toward a configurator: spec your B200 / B300 cluster — GPU count, interconnect, cooling — and buy it through us at procurement pricing, delivered to our halls or yours. Tell us what you need and it shapes what we build.
04B200 / B300 inquiry
Tell us about your training run.
Inquiry-first — no card required. We'll scope your B200 / B300 cluster or liquid-ready colocation, the cooling build, and timing, then follow up, usually within a business day.
05FAQ