TheAIUS DATA CENTERS
B300 nodes in a Tier-3 facility in the United States; H200 nodes in US East. Root access, no virtualization, no oversubscription.
WHOLE 8× B300 NODES · 1 TO 24 MONTHS · 30% LOCKS SLOT & RATE
Reserve whole 8× B300 nodes.
From $5.30/GPU·hr on a 1-year term — 23% under Verda's published annual rate.
Delivery window confirmed in your quote — we match your start date against current allocations.
No payment until you accept your quote. 30% prepayment then locks slot and rate.
Build your reservation.
Pick nodes and term. See your rate and take-or-pay term total.
NVIDIA B300
8× B300 SXM6 · 288GB HBM3e · NVLink 5 · 400G InfiniBand · whole node
= 8× B300 SXM6 · 288GB
TERM TOTAL $371,424.00 take-or-pay · 30% to lock $111,427.20
30% PREPAYMENT LOCKS SLOT & RATE · TAKE-OR-PAY · BILLING AFTER ACCEPTANCE
NVIDIA H200 — available now
8× H200 141GB SXM5 · 2× Intel Xeon 8558P (48c) · 3TB DDR5 · ~32TB local storage · whole node
available now · US East
= 8× H200 141GB SXM5
TERM TOTAL $224,256.00 take-or-pay · 30% to lock $67,276.80
MINIMUM 3 MONTHS · 30% AT SIGNING · TAKE-OR-PAY · BILLING AFTER ACCEPTANCE
ON-DEMAND (PER-MINUTE) — PLANNED · DATE TBA.
Reserved B300 pricing vs published reserved rates.
| Provider · B300 $/GPU·HR | On-demand | 1 month | 3 months | 6 months | 1 year | 24 months |
|---|---|---|---|---|---|---|
| TheAI Cloud reserved · 1-24 month terms | planned | |||||
| Verda· published reserved tiers | $7.50 | $7.35 | $7.28 | $7.20 | $6.90 | $5.625 · full prepayment |
| DigitalOcean· GPU Droplets, HGX B300 | — | — | — | — | $7.94 | — |
PUBLISHED PRICE LISTS, VERIFIED SEP 3, 2026 · B300 PER GPU·HOUR · PROVIDER NAMES LINK TO THE SOURCE
HYPERSTACK AND NEBIUS PUBLISH ON-DEMAND ONLY · COREWEAVE AND LAMBDA QUOTE RESERVED VIA SALES · AWS CAPACITY BLOCKS LIST AT $14.04/GPU·HR
VERDA: PUBLISHED 1-YEAR/2-YEAR RESERVED TIERS (FULL PREPAYMENT); ON-DEMAND/SPOT NOT COMPARED.
One-year reserved at $5.30 is ~23% under Verda's published annual rate and ~33% under DigitalOcean's. Six-month at $5.70 is ~21% under Verda's. Two-year reserved at $4.90 is ~13% under Verda's two-year rate — which requires full upfront payment; ours is 30% at signing. On-demand is planned, not live.
NVIDIA H200 — available now · vs published rates
| Provider · H200 $/GPU·HR | On-demand | 3 months | 6 months | 1 year |
|---|---|---|---|---|
| TheAI Cloud available now · US East · 3 months minimum | — | |||
| Verda· on-demand minus published commitment discounts | $4.00 | $3.88 | $3.84 | $3.68 |
| DigitalOcean· GPU Droplets, HGX H200 | $4.47 | — | — | $3.40 |
PUBLISHED PRICE LISTS, VERIFIED SEP 7, 2026 · H200 PER GPU·HOUR · ON-DEMAND NOT OFFERED BY US
VERDA: 1-YEAR COMMITMENT = 8% OFF ON-DEMAND; ITS 2-YEAR TIER ($3.00, PAID FULLY UPFRONT) IS NOT COMPARED — WE OFFER H200 ON 3-12 MONTH TERMS.
H200 three-month at $3.70 is ~5% under Verda's three-month commitment; one-year at $3.20 is ~13% under Verda's published annual rate and ~6% under DigitalOcean's 12-month reserved. Minimum term 3 months; 30% at signing.
Written answer in 24h · no payment until you accept
How much Blackwell actually exists — and how little is rentable short-term → read the count
Built for training runs and production inference.
NVLINK 5 + 400G INFINIBAND
InfiniBand NDR fabric between nodes, 400G per GPU — multi-node training works day one.
BILLING AFTER ACCEPTANCE
Acceptance criteria are written into your agreement before delivery. If the cluster doesn't pass them, the prepayment is refunded in full. NCCL benchmarks run before the first invoice; 99.5% SLA with service credits.
SCREENED CUSTOMERS
Every organization is verified before workloads run — U.S. export rules enforced.
Node spec.
Two SKUs — a whole 8× B300 SXM6 node (reserved) and a whole 8× H200 SXM5 node (available now). Exact configuration, nothing shared.
| Node | Supermicro SYS-822GS-NB3RT 8U · 8× NVIDIA B300 SXM 288GB (HGX B300 NVL8) · 2.3TB HBM3e per node |
|---|---|
| CPU | 2× Intel Xeon 6767P — 64C / 128T each (128 cores, 256 threads per node) · Intel TDX-capable |
| Memory | 3TB DDR5-6400 ECC RDIMM (32× 96GB) |
| Storage | 4× 3.84TB NVMe PCIe 5.0 E1.S (15.36TB) + 2× 1.92TB M.2 NVMe boot (3.84TB) · ~19TB local per node |
| Networking | 8× ConnectX-8, 400G per GPU (3.2 Tb/s per node) · InfiniBand NDR · NVLink 5 + NVSwitch in-node · public IPv4 per node · no egress fees |
| Security | TPM 2.0 · Intel TDX for confidential computing, BIOS-enabled on request |
| Location | United States, Tier III facility |
| Cluster | 8 nodes / 64 GPUs · whole-node allocations |
| Every contract | NCCL acceptance benchmarks before billing · 99.5% SLA with credits · monthly invoicing per agreement |
NVIDIA H200 node — available now
| GPUs | 8× NVIDIA H200 141GB SXM5 (HGX H200) · 1.1TB HBM3e per node |
|---|---|
| CPU | 2× Intel Xeon 8558P — 48 cores each (96 cores per node) |
| Memory | 3TB DDR5 |
| Storage | ~32TB local storage per node |
| Location | US East |
| Allocation | whole node · minimum term 3 months · 30% at signing |
Final site and configuration are confirmed in your capacity agreement.
Before you reserve.
How fast do I get access after signing?
The delivery window is written into your capacity agreement — typically two to four months from signing, and often sooner when an allocation is already in flight. Tell us your target date in the request; earlier allocations open up regularly, and we'll say plainly what's possible. Billing starts only after acceptance benchmarks pass.
Is the reservation paid or binding?
Requesting a configuration is free and creates no obligation. Your quote — site, delivery window, spec and rate — comes within 24 hours. A 30% prepayment then locks your slot and rate; from that point the reservation is take-or-pay.
How do I lock my slot and rate?
Request a configuration → written quote within 24 hours → you accept → 30% prepayment invoice (wire or USDC) → slot and rate locked, take-or-pay. Remaining 70% per the agreement, monthly invoicing is standard.
Can I cancel a locked reservation?
A locked reservation is a rental contract, not a hold: the 30% prepayment is credited to your term and is not refunded if you cancel. Until you accept a quote, nothing is owed.
What if the delivery window slips?
If the delivery window in your agreement is missed, you choose: a rate reduction for the delay, or a full refund of the prepayment if the delay is material. Billing never starts before acceptance, and your rate does not move.
What if the cluster doesn't pass acceptance?
Acceptance criteria are written into your agreement before delivery. If the cluster doesn't pass them, the prepayment is refunded in full. The criteria are spec-based until first delivery: nccl-tests all_reduce intra-node over NVLink and inter-node over InfiniBand NDR against bandwidth thresholds, DCGM diagnostics at level 3, and a burn-in run.
What happens at zero balance?
Reserved terms don't run on a balance: you're invoiced per your capacity agreement, and a missed invoice is handled under its payment terms. On-demand, once it opens, will run from a prepaid balance — at zero balance on-demand instances are terminated immediately and drives are wiped, no grace period; we email you at ~24h and ~4h of remaining runway.
Can I train across multiple nodes?
Yes. B300 nodes interconnect over a 400G InfiniBand NDR fabric (400G per GPU) — reserve two or more and they're delivered as one cluster.
How does billing work?
Reserved terms: a 30% prepayment invoice (wire or USDC) after you accept the quote, then monthly invoices per the capacity agreement. Billing starts only after acceptance benchmarks pass. No egress fees. Card top-ups and minute-based billing arrive with on-demand — planned, date TBA.
What SLA do you offer?
99.5% monthly uptime per node, with service credits for shortfalls — see Terms for the credit schedule.
Who can use the service?
Our nodes are hosted in the United States; U.S. export controls apply and we follow them: every organization passes screening during quote acceptance, before workloads run.
From the blog.
How we price whole-node B300 — and what we learned scanning the market.
How Much Blackwell Actually Exists?
Blackwell shipments set records every quarter — and the median public B300 on-demand price rose 57% since November. We reconstruct how much has actually shipped (~7M packages), how little of it is rentable (two sub-12-month listings), and why prices rise while supply booms.
The Blackwell Fact Table
Every figure from “How Much Blackwell Actually Exists?” with its source: shipments, the B300 slice, capex and anchor deals, manufacturing ceilings, Rubin, the H100 precedent — plus methodology notes and unit conventions.
Kimi K3 (2.8T) and six other open models on a single 8×B300 node — full vLLM numbers
We ran seven current open models — from Gemma 4 31B to Kimi K3's 2.78T parameters — through out-of-box vLLM on one 8× B300 node. TTFT, throughput, what didn't launch, and what surprised us.
Team.
Oleksiy Sitnichenko
CEO
Over 20 years of global executive leadership: business management, entrepreneurship, and strategic development across diverse international markets.
Viacheslav Zhvart
CTO, GPU infrastructure supply
Leads a unit that puts datacenter GPU capacity into production. Talks to DCs, engineers, and buyers in one thread — from rack constraints to contract language.
Mykola Baichak
Head of ML Infrastructure
Senior engineer with 10+ years building high-performance systems across autonomous driving, quantitative trading, and large-scale ML inference.
Pavel Solovskyi
ML Infrastructure & Performance Engineering
Infrastructure engineer specializing in confidential AI compute, GPU performance engineering, and large-scale inference operations.