Own your GPUs. We rack them, run them, and hand you root.
Stop renting the same silicon every month. Buy a purpose-built multi-GPU server, and Nexus sources it, burns it in, and hosts it in a certified facility on one flat fee. Live in weeks.
- No facility search, no 100 kW minimums, no NVIDIA allocation queue
- Titled to you — a balance-sheet asset, not a cloud bill
- Power, cooling, network, remote hands, and monitoring included
The lineup
Four servers. One architecture.
Datacenter-licensed GPUs on a dual-EPYC, ECC, NVMe backbone. Built on ASUS, Gigabyte, or Supermicro chassis — we quote whichever has allocation when you order.
Nexus N4
Entry-scale inference and fine-tuning · 2U
- GPUs
- 4× RTX PRO 6000 Blackwell
- VRAM
- 384 GB GDDR7 ECC
- CPU
- Dual EPYC 9005
- RAM
- 512 GB DDR5
- Storage
- 8 TB NVMe
- Network
- 2× 25GbE
- Draw
- ~3.5 kW
Nexus N8
The workhorse · 4U
- GPUs
- 8× RTX PRO 6000 Blackwell
- VRAM
- 768 GB GDDR7 ECC
- CPU
- Dual EPYC 9555
- RAM
- 768 GB DDR5
- Storage
- 16 TB NVMe
- Network
- 2× 100GbE
- Draw
- ~6.5 kW
Nexus N8X
Multi-node training over RoCE · 4U
- GPUs
- 8× RTX PRO 6000 Blackwell
- VRAM
- 768 GB GDDR7 ECC
- CPU
- Dual EPYC 9655
- RAM
- 1.5 TB DDR5
- Storage
- 30 TB NVMe
- Network
- 400GbE CX-8
- Draw
- ~7 kW
Nexus H8
Frontier training on HGX · 8U
- GPUs
- 8× H100 SXM5
- VRAM
- 640 GB HBM3
- CPU
- Dual Xeon / EPYC
- RAM
- 2 TB DDR5
- Storage
- 30 TB NVMe
- Network
- 8× 400Gb IB
- Draw
- ~10.2 kW
Already have a Dell, HPE, Lenovo, or Supermicro quote?
Send it over. We match it, source it, and host it on the same terms.
Prices are indicative and itemized per build at time of order.
How it works
You own it. We operate it. You run on it.
Spec and quote
Tell us the workload. We return an itemized build, chassis options, hosting terms, and a date.
Build and burn in
72 hours of GPU, memory, and NVMe stress on our bench. You get the acceptance report.
Rack and hand off
Cabled and networked in a Nexus facility. SSH keys, IPMI access, and Ubuntu + CUDA + Docker ready.
We run the iron
Power, cooling, swaps, and monitoring on us. You run your stack and get one flat invoice.
Own it, or rent it first.
Own & host
You buy the server; Nexus sources, builds, racks, and operates it. Title is yours from day one.
- Generally Section 179 eligible (consult your CPA)
- Full root and IPMI access
- Ship it out or sell it whenever you choose
- Lease-to-own available through a finance partner
Rent instead
Not ready for capex? Reserve a whole Nexus server by the month — same hardware, same facilities, published pricing.
- RTX PRO 6000 from $1.50/GPU-hr (12-month)
- Month-to-month, 6-month, or 12-month terms
- Bare metal, single tenant, root access
- Switch to ownership later
One flat fee
Everything below is in the hosting fee.
Billed per kW per month against rated draw. No per-hour markups.
Power & cooling
A+B redundant feeds, air or liquid cooling matched to density.
Network & IPs
Committed uplink, dedicated IP block, VPN-only management network.
Remote hands
Racking, cabling, hot-swaps, firmware and driver updates, reimaging.
Burn-in & acceptance
72-hour stress test before shipping, with a per-GPU report.
Monitoring
DCGM health, thermals, ECC tracking, alerts before failures hit you.
Monthly statement
Power consumed, uptime, incidents, support hours. Per server.
Common questions
Why RTX PRO 6000 instead of RTX 5090?
Who owns the server?
What does hosting cost?
How fast can you deploy?
Can I run my own orchestration?
Do you handle multi-node training?
Get a quote
Tell us what you're running.
We come back with an itemized build, hosting terms, and a deployment date.
Final configuration, pricing, and delivery depend on component availability and hosting capacity at time of order.
Buying to rent out, not to run?
Enroll the same servers in the GPU Owner Program: Nexus rents them to AI teams on your behalf.
GPU Owner ProgramNeed racks, not boxes?
For multi-rack and multi-megawatt deployments, Nexus sources dedicated capacity across its network.
AI & GPU Colocation