Loading…
What is established
Operational. 300 MW announced (gross site power); 230,000 accelerators (NVIDIA H100/H200 + ~30,000 GB200); owned or developed by xAI.
What is not disclosed
Project capex, timeline, named tenant, power source, cooling technology — not disclosed by any source on this page. Recorded as unknown, never as zero.
Analyst note
Original Colossus at the former Electrolux plant, operational since 2024. 300 MW is facility power consumption for the completed Phase 2 buildout, not on-site generation (the site separately has ~150 MW of Megapack buffering and gas turbines). 2026 reporting says Colossus 1 now serves inference while training moved to Colossus 2. CORRECTION 2026-08-01: the 300 MW previously shown as confirmed was Tom's Hardware's Phase 1+2 consumption target, wrongly attributed to a Musk statement that covered only the GPU count. Documented power: 150 MW grid interconnection energized May 2025 (MLGW/TVA), a further 150 MW approved February 2026 (energization unconfirmed), and ~247 MW of permitted on-site gas generation. The 230,000-GPU figure is primary (Musk, July 2025) and stands. Recent reporting has Colossus 1 shifted to inference duty.
Physical work is under way and the evidence includes a primary source.
Power readiness
Capacity basis stated
The source distinguishes IT load from total site power or on-site generation.
Derived from the evidence on this page by a published rule, not a model. How this is calculated.
NVIDIA H100/H200 + ~30,000 GB200
35.0480, -90.1500
No companies linked to this project yet.
| Metric | Value | Confidence | As of | Basis / methodology | Source |
|---|---|---|---|---|---|
| Power | 300 MW | Medium | Not disclosed | Basis: site power. Phase 1+2 design target — not measured operating draw. 150 MW grid energized May 2025; ~247 MW on-site generation permitted. | xAI (Elon Musk) |
| GPU count | 230,000 GPUs | Confirmed | Not disclosed | Model: NVIDIA H100/H200 + ~30,000 GB200. | xAI (Elon Musk) |
xAI (Elon Musk) · Jul 22, 2025 · accessed Jul 28, 2026
230k GPUs, including 30k GB200s, are operational for training Grok @xAI in a single supercluster called Colossus 1
Reliability
95/100
Tom's Hardware · Not disclosed · accessed Jul 28, 2026
Phase 2 to consume 300 MW, enough to power 300,000 homes
Reliability
65/100
Original Colossus at the former Electrolux plant, operational since 2024. 300 MW is facility power consumption for the completed Phase 2 buildout, not on-site generation (the site separately has ~150 MW of Megapack buffering and gas turbines). 2026 reporting says Colossus 1 now serves inference while training moved to Colossus 2. CORRECTION 2026-08-01: the 300 MW previously shown as confirmed was Tom's Hardware's Phase 1+2 consumption target, wrongly attributed to a Musk statement that covered only the GPU count. Documented power: 150 MW grid interconnection energized May 2025 (MLGW/TVA), a further 150 MW approved February 2026 (energization unconfirmed), and ~247 MW of permitted on-site gas generation. The 230,000-GPU figure is primary (Musk, July 2025) and stands. Recent reporting has Colossus 1 shifted to inference duty.
Correction: 300 MW was a design target misrecorded as confirmed operating power; GPU count (230k) stands on Musk's statement
Aug 1, 2026
System
Oracle · Abilene, Texas, United States
Galaxy Digital · Dickens, Texas (Dickens County), United States
Nebius Group · Butler Township, Pennsylvania, United States
Applied Digital · Ellendale, North Dakota (Dickey County), United States
Meta · New Albany, Ohio, United States
Related Digital · Saline Township, Michigan, United States