Supercomputers/Systems/Microsoft Fairwater (Atlanta)
USA · operational · ai training
Microsoft Fairwater (Atlanta)
Deep dive: Microsoft Fairwater: two sites, few numbers
Operated by Microsoft at Microsoft Atlanta AI datacenter site .
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 Blackwell GB200 and GB300 GPUs, up to 72 per rack | NVIDIA | - | Reported | blogs.microsoft.com |
AI datacenter metrics
Figures beyond an accelerator count, each with who stated it. What can and cannot be compared across AI datacenters.
| Metric | Value | As of | Basis | Note | Source |
|---|---|---|---|---|---|
| Accelerators | 304,279 accelerators | 2026-06-14 | Third-party estimate | Epoch AI chip estimate, latest per chip type (B200); some rows are company disclosures, others modelled. | epoch.ai |
| Stated investment | 24.1 USD bn | 2026-06-14 | Third-party estimate | Epoch AI modelled build cost, compute plus construction, 2025 USD, scaled from IT power; not a disclosed spend. | epoch.ai |
| Facility power | 859 MW | 2026-06-14 | Third-party estimate | Epoch AI modelled gross facility power (IT plus cooling and overhead) at this timeline date. | epoch.ai |
| IT power | 636 MW | 2026-06-14 | Third-party estimate | Epoch AI modelled IT power from satellite imagery and permits at this timeline date; not operator-stated. | epoch.ai |
| Rack power | 140 kW | 2025-11 | Operator stated | Microsoft Fairwater design: about 140 kW per rack and 1,360 kW per row, up to 72 Blackwell GPUs per rack, as described for the Atlanta site | blogs.microsoft.com |
How it is powered
| Source | Provider | MW | Status | As of | Basis | Note | Source link |
|---|---|---|---|---|---|---|---|
| Grid | Georgia Power | - | operating | 2025-11 | Operator stated | Microsoft: highly available utility power lets Atlanta skip on-site generation and UPS. It does not name the utility; Epoch lists Georgia Power. No MW stated. | blogs.microsoft.comepoch.ai |
| Battery storage | Microsoft | - | planned | 2025-11 | Operator stated | Microsoft describes an on-site energy storage solution to mask power oscillations from large training jobs; capacity not stated. | blogs.microsoft.com |
How AI campuses are powered: capacities are what the operator, utility or filing states, not what has been delivered.
The read
Microsoft describes an Atlanta AI datacenter of NVIDIA Blackwell GB200 and GB300 GPUs at up to 72 per rack, about 140 kW per rack, on a two-tier 800 Gbps Ethernet fabric and a dedicated AI wide-area network that links it to the Wisconsin site. Microsoft gives no GPU count and no facility power, and trade press repeats Microsoft's release rather than corroborating it, so this stays reported. The network vendor is a live disagreement (Microsoft's blog describes SONiC-based Ethernet, one trade article says Spectrum-X), so no interconnect row is written.
Timeline
- 2025-11 Announced source
Site & facility
Microsoft Atlanta AI datacenter site
- Location
- Atlanta, GA, USA
- Cooling
- Closed-loop liquid cooling
- Commissioned
- 2025
Where sources disagree
We record conflicts instead of picking a winner quietly. All open questions.
- Scale-out network: SONiC-based Ethernet (Microsoft) against NVIDIA Spectrum-X (trade press). No interconnect edge is recorded until one source resolves it.
Change history
- 2026-10-04 AI datacenter metrics: Microsoft Fairwater (Atlanta)
- 2026-09-19 Microsoft Fairwater Atlanta added
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 source | Result | Found on the page |
|---|---|---|
| blogs.microsoft.com | names system and part | NVIDIA GB200 NVL72 |
Checked 2026-10-09 by pnpm verify. Re-run it and the table changes with the web.
Related news
- Microsoft links Fairwater Atlanta and Wisconsin into one "AI superfactory" · MicrosoftThe two Fairwater sites are joined over a dedicated AI WAN fiber backbone so training jobs can span both facilities as one system rather than two separate clusters, a different scaling strategy than the single site gigawatt campuses xAI and Meta are building. Each rack carries up to 72 NVIDIA Blackwell GPUs; Microsoft has not disclosed a total GPU count for either site.