🌍 AI Factories Top 20 — 2026 H1 Edition
Inspired by the TOP500 Supercomputer List, this list ranks the largest known or announced AI training infrastructures worldwide — often called “AI Factories”.
Figures are based on public announcements, credible leaks, and infrastructure analysis as of November 2025.
📦 AI Infrastructure Technical Cards
The following technical cards summarize the most important publicly known specifications for each project. Commercial AI infrastructure is often less transparent than traditional HPC systems, therefore some values are based on public announcements and conservative industry estimates.
🥇 1. Stargate
| Field | Value |
|---|
| 🏢 Operator | OpenAI · Oracle · SoftBank · MGX |
| 📍 Primary Sites | Abilene (Texas) plus additional U.S. campuses |
| 🖥️ Platform | Oracle Cloud AI Infrastructure |
| 🧠 CPUs | Grace CPU / x86 (expected mix) |
| 🚀 Accelerators | NVIDIA GB200 NVL72 → future Blackwell generations |
| 🔢 Estimated Scale | 100,000+ GPUs per major campus |
| 🌐 Network | NVIDIA Quantum InfiniBand + OCI RDMA (expected) |
| 💾 Storage | Enterprise AI storage (not disclosed) |
| ❄️ Cooling | Direct liquid cooling |
| ⚡ Target Power | Multi-GW (>9 GW long-term program) |
| 🧪 MLPerf | Not published |
| 📊 HPL-MxP | Not published |
| 🏁 Status | 🚧 Under construction |
| ⭐ Why it matters | Largest AI infrastructure investment announced globally. |
🥈 2. Colossus
| Field | Value |
|---|
| 🏢 Operator | xAI |
| 📍 Location | Memphis, Tennessee |
| 🖥️ Platform | NVIDIA SuperPOD-inspired architecture |
| 🧠 CPUs | AMD EPYC |
| 🚀 Accelerators | NVIDIA H100 expanding toward Blackwell |
| 🔢 Estimated Scale | ~200,000 GPUs with roadmap toward 1 million |
| 🌐 Network | NVIDIA Spectrum-X Ethernet |
| 💾 Storage | High-performance AI storage (not disclosed) |
| ❄️ Cooling | Liquid cooling |
| ⚡ Estimated Power | Hundreds of MW |
| 🧪 MLPerf | Not published |
| 📊 HPL-MxP | Not published |
| 🏁 Status | 🚀 Operational & expanding |
| ⭐ Why it matters | Largest production AI training cluster currently operating. |
🥉 3. Oracle OCI AI Superclusters
| Field | Value |
|---|
| 🏢 Operator | Oracle Cloud Infrastructure |
| 📍 Location | Multiple OCI regions |
| 🖥️ Platform | OCI AI Supercluster |
| 🧠 CPUs | Grace / x86 |
| 🚀 Accelerators | NVIDIA Blackwell |
| 🔢 Estimated Scale | 100k+ GPUs across regions |
| 🌐 Network | OCI RDMA |
| 💾 Storage | OCI High Performance Storage |
| ❄️ Cooling | Liquid |
| ⚡ Estimated Power | Multi-site |
| 🧪 MLPerf | Various OCI instances benchmarked |
| 📊 HPL-MxP | Not disclosed |
| 🏁 Status | 🚀 Expanding |
| ⭐ Why it matters | Core infrastructure partner for Stargate and frontier AI. |
4. Microsoft Azure AI Fabric
| Field | Value |
|---|
| 🏢 Operator | Microsoft |
| 📍 Location | Global Azure regions |
| 🖥️ Platform | Azure AI Fabric |
| 🧠 CPUs | Grace / Xeon / EPYC |
| 🚀 Accelerators | NVIDIA H100, H200, Blackwell |
| 🌐 Network | Azure AI Fabric |
| 💾 Storage | Azure Managed Storage |
| ❄️ Cooling | Liquid + hybrid |
| ⚡ Power | Global deployment |
| 🏁 Status | 🚀 Operational |
| ⭐ Why it matters | Production infrastructure for OpenAI and Microsoft AI services. |
5. Google AI Infrastructure
| Field | Value |
|---|
| 🏢 Operator | Google |
| 📍 Location | Global Google Cloud |
| 🖥️ Platform | TPU SuperPods |
| 🧠 CPUs | Google Axion + x86 |
| 🚀 Accelerators | TPU v5e, TPU v6, NVIDIA GPUs |
| 🌐 Network | Google Jupiter fabric |
| 💾 Storage | Colossus Distributed File System |
| ❄️ Cooling | Advanced liquid cooling |
| ⚡ Power | Global deployment |
| 🏁 Status | 🚀 Operational |
| ⭐ Why it matters | Industry leader in custom AI silicon and hyperscale networking. |
6. Amazon AI Infrastructure
| Field | Value |
|---|
| 🏢 Operator | AWS |
| 📍 Location | Global AWS Regions |
| 🖥️ Platform | EC2 UltraCluster |
| 🧠 CPUs | AWS Graviton |
| 🚀 Accelerators | Trainium2, Inferentia2, NVIDIA GPUs |
| 🌐 Network | Elastic Fabric Adapter (EFA) |
| 💾 Storage | Amazon FSx for Lustre |
| ❄️ Cooling | Liquid |
| ⚡ Power | Global deployment |
| 🏁 Status | 🚀 Operational |
| ⭐ Why it matters | Largest deployment of custom AI accelerators outside Google. |
7. CoreWeave AI Cloud
| Field | Value |
|---|
| 🏢 Operator | CoreWeave |
| 📍 Location | USA & Europe |
| 🖥️ Platform | NVIDIA DGX Cloud infrastructure |
| 🚀 Accelerators | H100, H200, Blackwell |
| 🌐 Network | NVIDIA InfiniBand |
| ❄️ Cooling | Liquid |
| ⚡ Estimated Power | Multi-campus |
| 🏁 Status | 🚀 Expanding |
| ⭐ Why it matters | Fastest-growing dedicated AI cloud provider. |
| Field | Value |
|---|
| 🏢 Operator | Meta |
| 📍 Location | USA |
| 🖥️ Platform | Hyperion AI Campus |
| 🚀 Accelerators | NVIDIA Blackwell generation |
| 🔢 Estimated Scale | 100k+ GPUs planned |
| ⚡ Target Power | Multi-GW |
| 🏁 Status | 🚧 Building |
| ⭐ Why it matters | Meta’s next-generation infrastructure for frontier open models. |
9. G42 AI Infrastructure
| Field | Value |
|---|
| 🏢 Operator | G42 |
| 📍 Location | Abu Dhabi |
| 🖥️ Platform | Condor Galaxy + UAE AI Campus |
| 🚀 Accelerators | Cerebras + NVIDIA |
| 🌐 Network | High-speed AI fabric |
| ⚡ Target Power | Multi-GW |
| 🏁 Status | 🚧 Expanding |
| ⭐ Why it matters | Foundation of the Middle East’s sovereign AI ecosystem. |
10. Chinese National AI Infrastructure
| Field | Value |
|---|
| 🏢 Operator | Multiple |
| 📍 Location | Multiple provinces |
| 🖥️ Platform | Mixed cloud infrastructure |
| 🚀 Accelerators | Huawei Ascend, NVIDIA H800, domestic NPUs |
| 🌐 Network | Not publicly disclosed |
| ⚡ Power | National deployment |
| 🏁 Status | 🚀 Expanding |
| ⭐ Why it matters | Largest coordinated sovereign AI expansion outside the United States. |
11. Lux
| Field | Value |
|---|
| 🏢 Operator | Oak Ridge National Laboratory |
| 🖥️ Platform | HPE Cray |
| 🧠 CPUs | AMD EPYC |
| 🚀 Accelerators | AMD Instinct MI355X |
| 🌐 Network | HPE Slingshot |
| ❄️ Cooling | Direct liquid cooling |
| 🏁 Status | 🚧 Building |
| ⭐ Why it matters | First dedicated U.S. DOE AI Factory. |
12. Discovery
| Field | Value |
|---|
| 🏢 Operator | Oak Ridge National Laboratory |
| 🖥️ Platform | HPE Cray |
| 🧠 CPUs | AMD EPYC (next generation) |
| 🚀 Accelerators | Future AMD Instinct generation |
| 🌐 Network | HPE Slingshot |
| 🏁 Status | 📅 Announced |
| ⭐ Why it matters | AI-first successor integrating simulation and foundation models. |
13. Solstice
| Field | Value |
|---|
| 🏢 Operator | Argonne National Laboratory |
| 🖥️ Platform | HPE Cray |
| 🚀 Accelerators | NVIDIA Blackwell |
| 🌐 Network | NVIDIA Quantum InfiniBand |
| 🏁 Status | 📅 Announced |
| ⭐ Why it matters | DOE flagship AI infrastructure for scientific discovery. |
14. Equinox
| Field | Value |
|---|
| 🏢 Operator | Argonne National Laboratory |
| 🖥️ Platform | HPE Cray |
| 🚀 Accelerators | NVIDIA Blackwell |
| 🌐 Network | NVIDIA Quantum InfiniBand |
| 🏁 Status | 🚧 Building |
| ⭐ Why it matters | First production deployment within the Solstice program. |
15. EuroHPC AI Factories
| Field | Value |
|---|
| 🏢 Operator | EuroHPC Joint Undertaking |
| 📍 Location | Multiple EU countries |
| 🖥️ Platform | Mixed national AI factories |
| 🚀 Accelerators | NVIDIA, AMD and European processors |
| 🌐 Network | Mixed HPC fabrics |
| ⚡ Estimated Scale | 19 AI Factories |
| 🏁 Status | 🚀 Rolling out |
| ⭐ Why it matters | World’s largest coordinated sovereign AI infrastructure initiative. |
16. JUPITER AI Factory
| Field | Value |
|---|
| 🏢 Operator | Forschungszentrum Jülich |
| 📍 Location | Germany |
| 🖥️ Platform | JUPITER Modular Supercomputer |
| 🧠 CPUs | NVIDIA Grace |
| 🚀 Accelerators | NVIDIA Grace Blackwell |
| 🌐 Network | NVIDIA Quantum InfiniBand |
| 💾 Storage | High-performance parallel storage |
| ❄️ Cooling | Direct liquid cooling |
| 🏁 Status | 🚧 Building |
| ⭐ Why it matters | Europe’s flagship exascale AI system. |
17. Alice Recoque AI Factory
| Field | Value |
|---|
| 🏢 Operator | GENCI / CEA |
| 📍 Location | France |
| 🏗️ System Integrator | Bull (formerly Eviden / Atos Advanced Computing) |
| 🖥️ Platform | BullSequana XH3500 |
| 🧠 CPUs | AMD EPYC “Venice” + SiPearl Rhea2 |
| 🚀 Accelerators | AMD Instinct MI430X |
| 🌐 Network | Bull BXI v3 (800 Gb/s) |
| 💾 Storage | DDN |
| ❄️ Cooling | Direct liquid cooling |
| ⚡ Estimated Power | ~12 MW |
| 🏁 Status | 🚧 Building |
| ⭐ Why it matters | France’s sovereign AI flagship and one of Europe’s first exascale AI factories built around the revived Bull HPC platform. |
18. ALPS
| Field | Value |
|---|
| 🏢 Operator | CSCS |
| 📍 Location | Switzerland |
| 🖥️ Platform | HPE Cray EX |
| 🧠 CPUs | NVIDIA Grace |
| 🚀 Accelerators | NVIDIA Grace Hopper |
| 🌐 Network | HPE Slingshot |
| ❄️ Cooling | Direct liquid cooling |
| 🏁 Status | 🚀 Operational |
| ⭐ Why it matters | One of the first production Grace Hopper supercomputers. |
19. Isambard-AI
| Field | Value |
|---|
| 🏢 Operator | UK Government |
| 📍 Location | Bristol |
| 🖥️ Platform | HPE Cray EX |
| 🧠 CPUs | NVIDIA Grace |
| 🚀 Accelerators | NVIDIA Grace Blackwell |
| 🌐 Network | NVIDIA Quantum InfiniBand |
| ❄️ Cooling | Direct liquid cooling |
| 🏁 Status | 🚧 Building |
| ⭐ Why it matters | Largest publicly funded AI system in the United Kingdom. |
20. ABCI 3.0
| Field | Value |
|---|
| 🏢 Operator | AIST |
| 📍 Location | Japan |
| 🖥️ Platform | ABCI Generation 3 |
| 🧠 CPUs | AMD EPYC |
| 🚀 Accelerators | NVIDIA H200 generation |
| 🌐 Network | High-speed InfiniBand |
| ❄️ Cooling | Hybrid liquid cooling |
| 🏁 Status | 🚧 Building |
| ⭐ Why it matters | Japan’s flagship open AI infrastructure serving academia and industry. |