Power-led AI infrastructureEngineered for accelerated compute

AI infrastructure,
from site to cluster.

Midas combines a controlled West Texas development site with a repeatable, liquid-cooled platform engineered for rack-scale AI.

Midas-controlled development platform
Controlled West Texas development platform
20MW → 120MW → 1.32GW Campus power development sequence
4 Scalable Units per module
GB300 NVL72 Current-generation design point

01 / The platform

One development stack.
Three integrated layers.

Midas coordinates site control, power, cooling and compute as one integrated development programme.

01

Site

Site control, diligence and phased transmission planning in Howard County.

  • Land
  • Permitting
  • Interconnection
02

Infrastructure

Modular power, liquid cooling and resilient distribution engineered as a repeatable block.

  • Power
  • Cooling
  • Controls
03

Compute

GB300 NVL72 rack infrastructure with high-speed fabric and a defined path to Vera Rubin density classes.

  • GB300 NVL72
  • Quantum-X800
  • Vera Rubin

02 / Howard County

A controlled site in Howard County.
Built for phased scale.

Adjacent to transmission infrastructure, the site combines an established air-permitting pathway for onsite generation with a 20 MW to 120 MW to 1.32 GW campus power roadmap.

Control

Midas-controlled development site in Howard County, Texas.

Permitting

Established air-permitting pathway for phased onsite generation.

Scale

20 MW, 120 MW and 1.32 GW campus power development stages.

HOWARD COUNTY / TX
Dark Fibre Corridor Midas Development Corridor
I-20 US 87
Campus power development sequence
20 MW120 MW1.32 GW

Map data © OpenStreetMap contributors · ODbL

03 / AI Infrastructure Module

The repeatable
tenant block.

Four Scalable Units combine 32 GB300 NVL72 racks, hybrid liquid cooling, independent A/B power and high-speed fabric. The baseline separates 4.8 MW of reference IT demand from a 6.144 MW rack-distribution envelope.

Engineering visualization of the Midas AI infrastructure module alternating between fully panelled and service-access states
GB300 NVL72 reference configuration Panels installed ↔ service access
Platform scale32 racks

2,304 Blackwell Ultra GPUs across four Scalable Units.

Reference IT TDP4.8 MW

Published-reference demand across the four-SU tenant block.

Distribution envelope6.144 MW

192 kW per rack for coordinated power distribution.

Cooling architecture2 zones

Two SUs per zone, with a ≥2.5 MW usable-duty design target per zone.

Start a conversation

Configure the first block
around your workload.

For capacity, site partnerships and strategic capital discussions.

hosting@midascompute.com

Investors and strategic partners

Investors
Strategic partners