Free cookie consent management tool by TermsFeed Generator
AI infrastructure · T3 Elemental Datacenter
Odessa, Texas · 400 acres

Project Landman.

A gigawatt-scale AI campus in Odessa, Texas, built one Elemental Data Center at a time. Owned land, owned power, and self-contained compute buildings that come online one by one, so capacity arrives in months, not years.

1.6GWplanned capacity
400MWinitial deployment
12EDCs standing in Phase 1
What are Elemental Data Centers?

The building blocks of Landman.

Elemental Data Centers (EDCs) are the building blocks of Project Landman, STN’s 400-acre AI campus in Odessa, Texas. Each EDC is a self-contained data center with its own power, cooling, networking and security. Electrical and mechanical systems sit upstairs, leaving the data hall below dedicated to dense racks of next-generation NVIDIA GPUs built for AI training and inference.

Because every EDC stands on its own, we bring capacity online one building at a time instead of waiting years for one massive facility. Twelve EDCs are already standing in Phase 1, Phase 2 is underway, and the campus is planned to scale toward 1.6 GW. STN owns the land, the buildings and the compute inside them: one partner, accountable from the dirt to the GPU.

12EDCs standing in Phase 1
Phase 2underway
1.6GWplanned campus scale
400acres, Odessa, Texas
01
Why here

Build the campus where the power is.

AI capacity is gated by electricity, not GPUs. Project Landman sits on land STN owns in the Permian Basin, with generation we control and one utility feed into the site. No waiting in an interconnection queue, no landlord between the compute and the grid.

  • Owned landRoom to scale from 400MW to 1.6GW on the same site
  • Owned powerGeneration and delivery controlled by STN
  • Single utility feedOne controlled entry point, cleanly managed
STN Mod42 data center module at dusk
Mod42, dusk. The module every hall at Landman is built from.
02
The POD

24 data halls. One block.

Landman is built one POD at a time. Each POD is a block of 24 identical data halls on a shared power and cooling backbone, and each POD can be tuned to the density the tenant needs. Pick one.

Per hall1.9MW
PODs at 400MW9
PODs at 1.6GW36

Illustrative math from the published figures: 400MW initial, 1.6GW full build, 24 halls per POD.

Project Landman POD of 24 data halls

17 of 24 halls lit at this density

One POD. Twenty-four halls, one backbone. 45MW.
03
The data hall

One hall. Ready to run.

A hall arrives as finished modules, not as a construction site. Racks, liquid cooling and power distribution are built and tested in the factory, then set on the slab and connected. Every hall is engineered for NVIDIA B300, GB300 and Vera Rubin, and every hall is single-tenant.

NVIDIA B300GB300Vera RubinSingle-tenant by designLiquid cooling, direct to rack
A single assembled hall: five modules, one floor, one system.
04
The kit

5 modules.1 system.

5modules
Five parts, one build

Built apart. Shipped as one.

Connected by a single thread and built as one. Keep scrolling to walk the kit module by module, or tap any module to jump to it.

Scroll
01 / 05 · Vestibule

The front door.

Controlled entry, airlock and staging. People come through here; dust and weather do not.

02 / 05 · GPU data hall

Where the compute lives.

B300, GB300 and Vera Rubin ready. Single-tenant by design.

  • Production-density racksTwo parallel rack rows per module, cooled at the rack door
  • B300 / GB300 / Vera RubinSized for current and next-generation NVIDIA hardware
  • Single-tenant by designYour workload, your hardware, your container
03 / 05 · Cooling distribution

Liquid cooling, factory-built.

Heat managed before it ever reaches a GPU.

  • Chilled water mainsTwo parallel insulated pipes running the full length
  • Direct-to-rack distributionLiquid at every rack position, branched at the factory
  • Redundant pump skidThree pumps with VFDs: variable flow, full failover
04 / 05 · GPU data hall

The second hall.

Mirrors the first. Two rack rows, cooled at the door, one tenant.

  • Production-density racksTwo parallel rack rows per module, cooled at the rack door
  • Single-tenant by designYour workload, your hardware, your container
05 / 05 · Electrical infrastructure

Power, kept upstairs.

Service the breakers without ever entering the data hall.

  • Stacked architectureDistribution panels and switchgear in their own dedicated module
  • Single utility feedOne controlled entry point, cleanly managed
  • Compute floor stays cleanNo panels, no cabinets in the rack room
Assembled

One system, one floor.

Heat is managed before it reaches a GPU, power is kept upstairs, and the compute floor stays clean. The modules lock onto a single floor assembly and ship as one data hall.

05
In days, not months

From truck to live.

The slow part of a data center is building it on site. Landman does the two slow things at the same time: the site is prepared while the modules are fabricated, so the day the trucks arrive is the week the hall goes live.

Timelapse. Site work at Landman, day and night, while the modules are built in the factory.
Site
Site preparation
Factory
Module fabrication and testing
Road
Ship
Slab
Set and connect
Live
Power on
ParallelDays
06
The finished hall

Fully assembled. Fully yours.

Every module you just walked through, locked onto one floor and set on land we own, fed by power we control. One tenant, one hall, ready for B300, GB300 and Vera Rubin the week it lands. Project Landman is the ground floor of GPU One, run by the team that runs regulated environments.

NVIDIA B300GB300Vera RubinSingle-tenant
Reserve Capacity ↗GPU One · Accelerate It
The finished Project Landman data hall: vestibule, two GPU halls, cooling and electrical modules on one floor
The finished hall. Five modules, one floor, one tenant. Built in the factory, live in days.
Let’s talk

Ready for what’s next?

400MW is being deployed now and the campus scales to 1.6GW. Talk to an AI infrastructure architect about density, timing and terms.

Schedule a Consultation ↗Read the Odessa Story