Verified answer
What are the largest neoclouds by capacity?
The largest neocloud by live, operational power is CoreWeave (about 1,500 MW active), while the biggest announced buildout is Nebius (about 5,000 MW). Power is the comparable axis; live and announced are ranked separately so a plan is never mistaken for a running site.
Top of the list (live power)
- #1CoreWeave1,500 MW live · 4,200 MW planned
- #2YTL AI Cloud600 MW live
- #3Nebius170 MW live · 5,000 MW planned
Ranked by live (operational) power
| CoreWeave | North America | 1,500 MW | ~250,000 (last disclosed, end-2024) | NVIDIA H100 / H200 / GB200 NVL72 |
| YTL AI Cloud | Asia | 600 MW | NVIDIA GB200 NVL72 (count undisclosed) | — |
| Nebius | Europe | 170 MW | ~60,000 (reported) | NVIDIA H100 / H200 / B200 / GB200 |
| Sify Technologies | Asia | 129 MW | — | NVIDIA (H100 / H200 / GB200; deployed count undisclosed) |
| Voltage Park | North America | 60 MW | ~24,000 NVIDIA H100 (~35,000+ combined post-Lightning merger) | NVIDIA H100 (B200 / GB300 added) |
| Core42 | Middle East | 60 MW | up to 35,000 GB300-equivalent (US export-approved); 10,000 via Northern Data | NVIDIA H100 / H200 (+ AMD, Cerebras WSE-3) |
Power that is energized and running today. Figures are drawn from each operator’s sourced record.
Ranked by announced / planned power
| Nebius | Europe | 5,000 MW | ~60,000 (reported) | NVIDIA H100 / H200 / B200 / GB200 |
| Crusoe | North America | 4,900 MW | — | NVIDIA GB200 / B200 / H200 + AMD MI300X |
| CoreWeave | North America | 4,200 MW | ~250,000 (last disclosed, end-2024) | NVIDIA H100 / H200 / GB200 NVL72 |
| Lambda | North America | 3,000 MW | >1,000,000 target by 2030 | NVIDIA H100 / B200 / GB300 NVL72 |
| TensorWave | North America | 1,000 MW | 8,192 AMD Instinct MI325X (largest AMD cluster in North America) | AMD Instinct MI325X / MI300X / MI355X |
| IREN | North America | 750 MW | ~150,000 target | — |
| HUMAIN | Middle East | 500 MW | 18,000 NVIDIA GB300 (first phase) | NVIDIA GB300 (Grace Blackwell Ultra) |
| Fluidstack | North America | 360 MW | — | NVIDIA H100 / H200 / B200 / GB200 |
| Northern Data | Europe | 250 MW | ~22,000 NVIDIA H100 / H200 (>85% allocated) | NVIDIA H100 / H200 |
| Nscale | Europe | 240 MW | ~200,000 NVIDIA GB300 (Microsoft, phased 2026-27) | NVIDIA GB300 / GB200 / H100 + AMD MI300X |
| Jio Platforms | Asia | 120 MW | >200,000 H100-equivalent at full build (H100-equivalent, not raw count) | NVIDIA GB300 (Blackwell Ultra) |
| Sustainable Metal Cloud | Asia | 84 MW | ~54,000 NVIDIA GB300 (Project Southgate) | NVIDIA H100 (GB300 planned); immersion-cooled |
| Krutrim | Asia | 20 MW | — | NVIDIA GB200 NVL72 / H100 |
| GMI Cloud | Asia | 16 MW | 7,000 NVIDIA GB300 (Taiwan AI Factory, go-live ~Mar 2026) | NVIDIA GB300 NVL72 |
Contracted or targeted capacity being built. An announcement is not a running site; see the biggest AI data centers for the same distinction at campus level.
GPU fleet disclosed, power not
- Denvr DataworksNorth America
~1,024 NVIDIA H100 SXM5
- ScalewayEurope
~2,000+ NVIDIA H100 (Nabuchodonosor)
- Together AINorth America
36,000 NVIDIA GB200 NVL72 (co-build); access to 100,000+
- E2E NetworksAsia
~4,000+ (2x1,024 H200 + 1,024 B200 + ~1,000 H100)
~10,000+ live; ordered 16,000 H100/GH200
tens of thousands of NVIDIA H100 (ordered)
- Sakura InternetAsia
~4,000+ (H100 / H200 / Blackwell); ~10,800 planned
- NexGen CloudEurope
AI Supercloud ~20,000 H100 target (~$1B)
Common questions
What is the largest neocloud?
By live, operational data-center power, the largest neocloud DEPLOY tracks is CoreWeave at about 1,500 MW active. By announced or contracted capacity the leader is Nebius at about 5,000 MW. Live and announced are different facts, and both are shown so a plan is never presented as a running site.
Why rank by power instead of GPU count?
Because power is the axis operators actually disclose and the one that is comparable. Most large neoclouds have stopped publishing a single GPU number and now report megawatts, and the GPU figures that do exist mix verified deployments, announced orders, and H100-equivalent estimates that cannot be ranked against each other. Power in megawatts is the honest common denominator; GPU fleet is shown as a secondary column where a credible figure exists.
Which neocloud has the most GPUs?
It depends on whether you count deployed or announced, and the honest answer is that the very largest operators no longer disclose a hard total. Among verified fleets, Northern Data (~22,000 H100/H200) and Voltage Park (~24,000 H100) publish clear numbers; the biggest announced fleets (Lambda's 1,000,000 by 2030, HUMAIN's up to 600,000, Nscale's ~200,000 GB300) are targets, not running hardware.
Are these figures operational or announced?
Both, shown separately. The live table ranks power that is energized and running today; the announced/planned table ranks contracted or targeted capacity that is being built. Each figure is drawn from the operator's sourced registry record, with its status kept explicit rather than blended into one number.