GPU hosting

AI data center hosting and GPU colocation

Own enterprise GPU hardware. Host it on Rax power at a contracted rate. Earn income from AI compute without building a data centre.

Book a consultationSee the infrastructureCall 718-766-8559

98%uptime guarantee
24/7monitoring
100+ MWavailable
60 daysto operational

Why Rax for AI compute

Power, cooling and deployment speed decide whether a GPU deployment is viable. Those are the parts Rax already runs every day.

Power is the constraint

Power is the largest running cost in AI compute and the hardest thing to secure. Rax secures power through its energy partnerships, so the rate is contracted and fixed for the life of the agreement rather than exposed to the spot market.

No interconnect queue

NatGas MDUs run on natural gas from a pipeline or well-head and are fully off-grid capable. Deployment is not gated behind a utility interconnection study.

Liquid cooling supported

Each module supports air cooling and hydro (liquid) cooling. Dense GPU nodes produce heat that air alone struggles with at scale.

About 60 days to operational

Order to running, covering manufacturing, assembly, logistics and commissioning. Units arrive fully assembled and tested.

You own the hardware

You purchase the AI GPU server outright and it is registered to you. Rax provides hosting, power, cooling, networking and maintenance under a service agreement.

Operated, not just rented

Racking, networking, cooling, remote monitoring and maintenance are handled by Rax. Starlink connectivity and remote monitoring are included in every module.

AI GPU Server configurations

Dense 8-GPU nodes on dual AMD EPYC 7003 (Milan), with the core count and PCIe lanes to feed every card.

Spacer

RTX 4090 Server

192 GB total VRAM

GPUs8x RTX 4090
Total VRAM192 GB
CPUDual EPYC 7003 (Milan)
System RAM512 GB DDR4 ECC
Storage4 TB NVMe
Power4x 2000W Platinum
WarrantyRax, 2 years

Book a consultation

Flagship

RTX Pro 6000 Blackwell

768 GB total VRAM

GPUs8x RTX Pro 6000 Blackwell
Total VRAM768 GB
CPUDual EPYC 7003 (Milan)
System RAM1 TB DDR4 ECC
Storage12 TB NVMe
Power4x 2000W Titanium
WarrantyRax, 2 years

Book a consultation

Income from AI-compute hosting varies with utilisation, workload mix and market demand, and is not guaranteed.

Talk to a Rax AI specialist

Call 718-766-8559Book a consultation

How it works

1

Buy the server

You purchase the GPU server outright. It is your hardware, registered to you.

2

Rax hosts and powers it

We rack it, cool it, connect it and monitor it in a Rax facility on Rax power.

3

Earn AI-compute income

The server rents its GPU capacity to AI workloads. You hold the asset and the revenue it produces.

What these servers power

AI model training

Multi-GPU runs where VRAM per node and interconnect matter more than single-card speed.

Inference at scale

Serving production models, where cost per token and sustained utilisation decide the economics.

Fine-tuning

Adapting open-weight models on private data that should not leave hardware you own.

Rendering and simulation

3D rendering, VFX and CFD work that is GPU-bound and bills by the hour.

GPU colocation for AI workloads

If you own GPU servers and need somewhere serious to run them, Rax provides GPU colocation built for AI workloads: high density rack space, industrial cooling, physical security and 24/7 monitoring, priced by power. GPU hosting starts at $0.125 per kWh for small fleets, with rates down to $0.10 per kWh at larger fleet sizes. Your team keeps full control of the hardware and the software stack, and our technicians are on site around the clock. For megawatt scale power, see our NatGas modular data centers.

GPU hosting, priced by power

$0.125per kWh, small fleets

$0.10per kWh, larger fleet sizes

Book a consultation

AI inference server hosting

Production AI inference is always on work, and it needs hosting that treats uptime as the product. Rax hosts inference servers with 98% uptime, redundant power and cooling, and 24/7 monitoring, so the models your customers rely on stay available. Start with a small deployment and scale rack by rack as demand grows. Book a consultation and we will map power, cooling and connectivity to your workload.

Frequently asked questions

Do I actually own the server hardware?

Yes. You purchase and own the physical AI GPU server outright. Rax provides hosting, power, cooling, networking and maintenance under a service agreement. The hardware is registered to you.

How is power billed?

At a fixed rate for the life of the agreement, so it does not move with the spot market. Current rates and tiers are published on the NatGas MDU page.

What income can I expect?

Income from AI compute varies with utilisation, workload mix and market demand, and is not guaranteed. In a consultation we can talk through what is realistic for the hardware and the workload you are planning.

Where will my server be hosted?

In a Rax facility. Capacity is available across the states and regions on our locations page. Site selection depends on deployment size and available gas supply.

Can this run off-grid?

Yes. NatGas MDUs run on pipeline or well-head gas and need no grid connection, which removes the utility interconnection queue from the timeline. Full specifications are on the NatGas MDU page.

How long does deployment take?

Approximately 60 days from order to operational, covering manufacturing, assembly, logistics and commissioning.

What happens if hardware fails?

Every configuration carries a 2-year Rax warranty. Monitoring is continuous and maintenance is handled on site as part of the hosting agreement.

Power your AI infrastructure with Rax

Tell us the workload and the scale, and we will come back with a configuration, a site and a rate.

Book a consultationCall 718-766-8559