Artemis X

The cloud

Frontier compute, dedicated by design

Every campus is built for one platform: NVIDIA Vera Rubin, the current frontier. No legacy tiers, no mixed stack — rack-scale compute, operated for tenants who need whole pods, not GPU-hours.

The platform

Vera Rubin NVL72, rack-scale

72 GPUs in a single NVLink domain per rack at 200 kW, liquid-cooled — the most productive inference silicon in the market, deployed as whole racks, never as slices.

The fabric

Bare metal, reference fabric

Rack-scale NVLink domains scaled out on the NVIDIA reference fabric, delivered bare-metal — no virtualization tax between the tenant and the silicon.

The tenancy

Dedicated, not spot

Capacity is sold in dedicated 5 MW blocks under long-term terms — single-tenant, isolated environments with managed Kubernetes and Slurm, run by an experienced cloud operator from the first rack.

The operations

Run like infrastructure

Workload telemetry, 24/7 operations and a SOC 2 program — with power, cooling and the cloud layer under one roof and one accountable team.

Row of NVIDIA Vera Rubin NVL72 racks — NVIDIA Vera Rubin NVL72 — rack row · Image: NVIDIA

Cooling

Closed loop. Zero water.

Modern AI racks reject most of their heat into liquid. Campus One is built on the NVIDIA DSX POD reference design: the racks, coolant distribution units, plumbing and cabling of each pod are integrated and validated in the factory, then the pod rolls into the hall and connects to a sealed warm-water loop and dry coolers — radiators and fans, no evaporation, no compressors on the compute path. The building's own air system rides on new sealed trim chillers, with the existing central plant valved in behind it as emergency backup.

Loop 1 · Compute

DSX POD units → warm-water loop → dry coolers

Factory-sealed pod loop→ Pod coolant distribution units→ Warm-water headers→ Dry coolers

Each DSX POD unit arrives with its internal loop already flushed, pressure-tested and validated. Rack heat crosses a metal wall in the pod's coolant distribution units and leaves through warm-water headers running on the campus's existing industrial pipe racks. Warm-water design means plain radiators reject it year-round.

Loop 2 · Building

Hall air → trim chillers

Existing air handlers→ Sealed chilled water→ Air-cooled trim chillers

The hall's existing pharma-grade air and humidity system is kept unchanged — only its source of cold water is new: sealed, air-cooled machines with a built-in spare.

The architecture

Built on NVIDIA DSX POD

NVIDIA's reference design for factory-built AI factory units on the Vera Rubin platform. Compute, cooling, power distribution and networking inside each pod are assembled and tested before shipment, to one validated specification, with 800 VDC at the pod input. What arrives at the dock is a finished unit, not a kit.

Why it matters

Less site work, less schedule risk

The integration that usually happens late in a build — rack plumbing, cabling, commissioning — moves upstream into a factory, in parallel with site preparation. That shortens the path from a ready hall to productive capacity and takes the largest execution risk off the critical path. Every campus in the program repeats the same unit.

Day one

Sealed campus from the July 2027 switch-on — dry coolers and trim chillers, no evaporative equipment

0 gal/day

Planned campus water consumption from July 2027 — one-time loop fills only

N+2

Two spare dry coolers on the compute loop; the existing 3,000-ton plant stays valved in as emergency backup for the building loop

800 VDC

Pod power input per the NVIDIA DSX design — more power through less copper, generation and conversion engineered to it

Cutaway of an NVIDIA DSX POD unit showing Vera Rubin racks, coolant distribution and power bays inside the enclosure — NVIDIA DSX POD · factory-integrated unit, installed inside the Campus One hall · Image: NVIDIA

Dry cooler array seen from above — Dry cooler array · closed-loop heat rejection, zero water