COMPUTE ATLAS Supercomputer supply-chain graph
1153 systems 482 sites Sourced data

Supercomputers/Systems/xAI Colossus 2

USA · operational · ai training

xAI Colossus 2

Deep dive: xAI Colossus: fast to build, hard to count

Operated by xAI at xAI Colossus 2 (Whitehaven, Memphis) .

Phase 1 seed dataset, compiled by hand. These rows were built from public operator, laboratory and vendor sources. A mechanical second-reader pass has since fetched every cited source: 1092 of 1153 systems have a readable citation that names them, and 40 are genuinely weakly sourced. Every claim carries its source and a confidence tier. Treat anything below verified as a lead, not a citation.

Measured performance

Rmax
-
Rpeak
-
Rmax ÷ Rpeak
-
Cores
-
Power
-
Per watt
- GF/W

Figures are as last publicly reported for the configuration described below, not a live measurement. Where a system was upgraded in place, the post-upgrade configuration is shown and the earlier one appears in the timeline.

What this machine is made of

One row per supplier relationship. “Supplier at build” is the company that shipped the part at the time; where that company has since been acquired, the parent it rolls up to today is shown beside it. That distinction is what makes ticker-level aggregation possible across a thirty-year dataset.

Role Part Supplier at build Quantity Confidence Source
Accelerator NVIDIA GB200 NVL72
Accelerators
Arm CPU + CUDA GPU · TSMC 4NP
NVIDIA GB200/GB300-class GPUs; reported counts range from roughly 110,000 at first light (January 2026) to a later-reported 550,000, per Musk's public statements rather than an xAI technical disclosure, so no single quantity is recorded
NVIDIA - Reported actionnews5.comintrol.com

AI datacenter metrics

Figures beyond an accelerator count, each with who stated it. What can and cannot be compared across AI datacenters.

MetricValueAs ofBasisNoteSource
Accelerators 220,000 accelerators 2026-05 Regulatory filing S-1: 110,000 GB200 plus 110,000 GB300 clusters (sum is ours); S-1 expects at least 220,000 more GB300 when the next phase is operational, no date sec.gov
Stated investment 20 USD bn 2026-01 Operator stated Mississippi governor announced a $20 bn xAI investment at the Southaven site; S-1 counts Southaven within Colossus II; other coverage says more than $20 bn wlbt.com
Facility power 1,276 MW 2026-06-15 Third-party estimate Epoch AI modelled gross facility power (IT plus cooling and overhead) at this timeline date. epoch.ai
IT power 946 MW 2026-06-15 Third-party estimate Epoch AI modelled IT power from satellite imagery and permits at this timeline date; not operator-stated. Assumes the S-1 next phase is online; S-1 itself states 430 MW. epoch.ai
IT power 430 MW 2026-05 Regulatory filing S-1: about 210 MW plus 220 MW compute power (sum is ours), excluding cooling; S-1 expects over 400 MW more when the next phase is operational, no date sec.gov

How it is powered

SourceProviderMWStatusAs ofBasisNoteSource link
On-site gas xAI 1,300 operating 2026-05 Third-party estimate Cleanview counts about 59 temporary mobile turbines (about 1.3 GW) at Southaven in May 2026. S-1 confirms a self-built gigawatt-scale gas plant can run COLOSSUS II alone. cleanview.cosec.gov
On-site gas xAI 1,200 permitted 2026-03 Regulatory filing MDEQ permit 10 Mar 2026 for 41 turbines (17 PGM-130, 16 Titan 350, 8 PE6000); about 1.2 GW is MeasuredAI and Cleanview sum of unit ratings. Cleanview: not yet installed. selc.orgmeasuredai.substack.com
Battery storage Tesla - operating 2026-05 Operator stated S-1 cites the largest Megapack deployment integrated with its data centers; MeasuredAI cites over 375 million USD of Megapacks at Colossus 2. No MW given. sec.govmeasuredai.substack.com
Grid Memphis Light, Gas and Water - planned 2026-05 Operator stated S-1: COLOSSUS II to be supplemented over time by utility-funded grid capacity. MLGW (May 2025) had grid options of 260 MW to 1.1 GW under discussion, no final request. sec.govmlgw.com

How AI campuses are powered: capacities are what the operator, utility or filing states, not what has been delivered.

The read

A second, physically distinct xAI supercluster in the Memphis area, built at 5420 Tulane Road in the Whitehaven area of Memphis with associated gas turbine power generation across the state line in Southaven, Mississippi, separate from the original Colossus site (xai-colossus-1) at the former Electrolux factory in South Memphis. Mississippi's governor announced an expansion of the site in January 2026 quoting Elon Musk directly that combined Colossus training compute was approaching "~2GW", and Introl's infrastructure industry blog independently reports an initial GB200/GB300 cluster reaching roughly 1 GW for Colossus 2 alone, though the reported GPU count shifts across sources and over time (on the order of 110,000 GPUs at first light in January 2026, reported by other outlets as growing toward 550,000 later in the year) and, while the earliest roughly 110,000 figure is an outside analyst's estimate from cooling capacity, the larger figures rest on Musk's public statements rather than an xAI technical disclosure. Given that spread, no specific GPU count is recorded on this row; the accelerator component row below carries the range with confidence reported and flags it as an estimate. No FLOPS, interconnect vendor or cooling vendor is independently confirmed for this system specifically.

Timeline

Site & facility

xAI Colossus 2 (Whitehaven, Memphis)

Location
Memphis, TN, USA
Commissioned
2026

Change history

Source check

We fetched this system's own citations and recorded whether each page actually mentions it. This is a corroboration signal, not a fact check, and it is published so you can see how well the sourcing holds up rather than take it on trust.

Cited sourceResultFound on the page
introl.comnames system and partNVIDIA GB200 NVL72
actionnews5.comnames system only-

Checked 2026-10-09 by pnpm verify. Re-run it and the table changes with the web.

Related news

  • xAI's Colossus 2 reaches 200MW of cooling capacity in six months · SemiAnalysis
    The Memphis and Southaven build reached the milestone in about six months against roughly fifteen months for comparable Oracle, Crusoe and OpenAI projects, and is targeted at over 1GW of GB200 NVL72 capacity. It is a data point on how much faster a single well capitalized operator can now stand up gigawatt class capacity than the shared infrastructure neoclouds this dataset also tracks.

All news →

Further reading