Chapter 5 · The physical AI economy
Compute, concrete and copper.
Upstream of every chatbot are models, chips, buildings and power. Epoch AI tracks the first three; Census tracks construction. These are inputs. Capacity is not utilization, announced is not operational, and global is not U.S.
Training compute of notable AI models
Each dot is a model. Blue: models Epoch flags as frontier.
474 notable models with compute estimates since 2012. Training compute has grown by many orders of magnitude.
Caveat. Most compute figures are Epoch estimates from papers, hardware and training time; confidence varies by model. Training compute is not inference capacity.
Inspect the evidence
- Inference compute serving users.
- Economic value of any model.
- Models without a compute estimate.
- notable_ai_models.csv (CC BY 4.0)
- Rows with publication date and training compute; no imputation
- Our World in Data's training-compute charts adapt this Epoch table; they are the same evidence family, not independent confirmation.
Text description. 474 notable models with compute estimates since 2012. Training compute has grown by many orders of magnitude.
AI model releases per year, by modality
Language and multimodal models dominate; video and audio models grew after 2023.
Caveat. Counts reflect Epoch's inclusion criteria and collection lag; recent years are under-collected (science models especially). Fewer listed models is not fewer models.
Inspect the evidence
Text description. Language and multimodal models dominate; video and audio models grew after 2023.
Release tracker: frontier, video, music and audio models
Table of recent model releases filterable by modality.
| Date | Model | Organization | Modality | Epoch domains |
|---|---|---|---|---|
| 2026-11-17 | GPT-6.2 Astra | OpenAI | Language and multimodal | Language,Vision,Multimodal |
| 2026-11-17 | GPT-6.2 Sol | OpenAI | Language and multimodal | Language,Multimodal,Vision |
| 2026-11-12 | Gemini 4 Hydrogen | Google DeepMind | Video | Language,Multimodal,Vision,Audio,Video |
| 2026-11-10 | Gemini 4 Carbon | Google DeepMind | Language and multimodal | Language,Multimodal,Vision |
| 2026-11-09 | Claude Fable 5.2 | Anthropic | Language and multimodal | Language,Multimodal,Vision |
| 2026-11-09 | Muse Ember | Meta AI | Video | Language,Multimodal,Vision,Audio,Video |
| 2026-11-09 | Claude Mythos 5.2 | Anthropic | Language and multimodal | Language,Multimodal,Vision |
| 2026-11-05 | Inkling 2 Small | Thinking Machines | Language and multimodal | Language,Vision,Multimodal |
| 2026-11-04 | SWE-2 Flash | Cognition | Language and multimodal | Language |
| 2026-11-04 | Ripple-65B Preview | Ricursive Intelligence | Language and multimodal | Language |
| 2026-10-28 | Particle-1 | Particle | Language and multimodal | Language |
| 2026-10-27 | MAI-Image-2.7 | Microsoft | Image | Image generation |
| 2026-10-27 | Genie 4 | Google DeepMind | Video | 3D modeling,Language,Audio,Multimodal,Video |
| 2026-10-27 | MAI-Cyber-1.1-Flash | Microsoft | Language and multimodal | Cybersecurity,Language |
| 2026-10-26 | GPT-6.1 Terra | OpenAI | Language and multimodal | Language,Multimodal,Vision |
| 2026-10-23 | Kimi K3.1 Pro | Moonshot | Video | Language,Multimodal,Vision,Video |
| 2026-10-22 | Boba | Humans& (Humans and) | Language and multimodal | Language |
| 2026-10-22 | Canva Craft | Canva | Image | Image generation |
| 2026-10-21 | Mirendil-Base | Mirendil | Language and multimodal | Language |
| 2026-10-20 | Ampersand | Humans& (Humans and) | Language and multimodal | Language |
| 2026-10-16 | Mercury 3 | Inception Labs | Language and multimodal | Language |
| 2026-10-15 | Muse Spark Open | Meta AI | Video | Language,Multimodal,Vision,Audio,Video |
| 2026-10-15 | MAI-Base-2 | Microsoft | Language and multimodal | Language,Vision,Multimodal |
| 2026-10-15 | Kling 4.0 | Kuaishou Technology | Video | Image generation,Video,Audio |
| 2026-10-15 | GPT-Live-1.1 Quip | OpenAI | Audio, music, speech | Audio |
| 2026-10-15 | GPT-Live-1.1 | OpenAI | Audio, music, speech | Audio |
| 2026-10-15 | Bonsai 2 9B | PrismML | Language and multimodal | Language,Vision,Multimodal |
| 2026-10-14 | Grok Voice 3.0 | xAI | Audio, music, speech | Language,Audio,Speech |
| 2026-10-13 | Gemini 4 Argon | Google DeepMind | Language and multimodal | Language,Multimodal,Vision |
| 2026-10-12 | Muse Wonder | Meta AI | Video | Language,Vision,Video,Multimodal |
| 2026-10-12 | Muse Wonder Full | Meta AI | Video | Language,Vision,Video,Multimodal |
| 2026-10-09 | Qwen-Image-2.1-Pro | Alibaba | Image | Image generation |
| 2026-10-07 | Claude Haiku 5.5 | Anthropic | Language and multimodal | Language,Vision |
| 2026-10-06 | Mistral Large 4 | Mistral AI | Language and multimodal | Language,Multimodal,Vision |
| 2026-10-06 | Gemini 3.6 Flash Image (Nano Banana 2.1) | Google DeepMind | Image | Image generation,Language |
| 2026-10-05 | Beam 1.0 | Reflection | Language and multimodal | Language |
| 2026-10-01 | Grok Imagine Video 1.5 Lite | xAI | Video | Image generation,Video,Vision,Audio |
| 2026-10-01 | MAI-Voice-2.1-Flash | Microsoft | Audio, music, speech | Speech |
| 2026-10-01 | MAI-Voice-2.1 | Microsoft | Audio, music, speech | Speech |
| 2026-10-01 | MAI-Transcribe-2-Streaming | Microsoft | Audio, music, speech | Speech,Language,Audio |
| 2026-09-30 | Mercury Decide | Inception Labs | Language and multimodal | Language |
| 2026-09-29 | GPT-6.1 Sol | OpenAI | Language and multimodal | Multimodal,Language,Vision |
| 2026-09-29 | Mercury Voice | Inception Labs | Language and multimodal | Language |
| 2026-09-28 | Claude Sonnet 5.5 | Anthropic | Language and multimodal | Language,Multimodal,Vision |
| 2026-09-23 | Nemotron 3.5 Bolt E4B | Nvidia | Audio, music, speech | Language,Vision,Multimodal,Audio,Speech |
| 2026-09-22 | MiMo-V2.6 Pro | Xiaomi Corp | Language and multimodal | Language |
| 2026-09-22 | MiMo-V2.6 Flash | Xiaomi Corp | Language and multimodal | Language |
| 2026-09-22 | Claude Opus 5.5 | Anthropic | Language and multimodal | Language |
| 2026-09-22 | GPT-6 Sol | OpenAI | Language and multimodal | Multimodal,Language,Vision |
| 2026-09-22 | GPT-6 Luna | OpenAI | Language and multimodal | Multimodal,Language,Vision |
| 2026-09-22 | GPT-6 Luna Decisions | OpenAI | Language and multimodal | Language,Vision,Multimodal |
| 2026-09-22 | FLUX 3 Action | Black Forest Labs | Video | Robotics,Multimodal,Video,Vision,Language |
| 2026-09-21 | Grok 4.7 | xAI | Language and multimodal | Language |
| 2026-09-20 | Qwen-Image-2.1 | Alibaba | Image | Image generation |
| 2026-09-18 | Step 5 Preview | StepFun | Language and multimodal | Language,Vision,Multimodal |
| 2026-09-15 | Periodic Neon | Periodic Labs | Language and multimodal | Language,Materials science |
| 2026-09-15 | Jev | TypeSafe.ai | Language and multimodal | Language |
| 2026-09-10 | SWE-2 | Cognition | Language and multimodal | Language |
| 2026-09-10 | Persimmon | Humans& (Humans and) | Language and multimodal | Language |
| 2026-09-09 | DeepSeek V4.1 Flash | DeepSeek | Language and multimodal | Language,Vision,Multimodal |
60 releases since 1 Jul 2026 in Epoch's all-models table. Refresh with npm run data.
Caveat. A release is not adoption, and adoption is not impact. Modality is derived from Epoch's domain tags (first match: video, audio, image, language).
Inspect the evidence
Text description. Table of recent model releases filterable by modality.
Facilities
Frontier AI data centers in the United States.
Epoch tracks 77 large U.S. AI data centers with an estimated 12,640 MW of current power between them. Power is estimated from satellite imagery, permits and filings, not metered. State totals below aggregate Epoch's facility estimates; states without a tracked site are hatched.
Estimated current power of tracked AI data centers, by state
Largest tracked capacity: TX 1730 MW, TN 1373 MW, OH 1244 MW, IN 1215 MW.
Caveat. Capacity, not utilization. Only sites Epoch tracks; many smaller data centers are absent. Hosting a facility is not the same as local benefit.
Inspect the evidence
Text description. Largest tracked capacity: TX 1730 MW, TN 1373 MW, OH 1244 MW, IN 1215 MW.
| Site | State | Power (MW) | H100e | Owner |
|---|---|---|---|---|
| Anthropic-Amazon New Carlisle Users: Anthropic | IN | 1,037 | 818,428 | Amazon |
| Colossus 2 Users: Anthropic, Cursor, SpaceXAI | TN | 946 | 1,111,673 | SpaceXAI |
| OpenAI Stargate Abilene Users: OpenAI | TX | 824 | 1,018,696 | Oracle |
| Microsoft Fairwater Atlanta Users: OpenAI, Microsoft | GA | 636 | 768,769 | Microsoft |
| Meta Prometheus Users: Meta | OH | 471 | 536,343 | Meta |
| Microsoft Fairwater Wisconsin Users: OpenAI, Microsoft | WI | 369 | 445,680 | Microsoft |
| Google Pryor (North) Users: Google DeepMind | OK | 368 | 636,685 | |
| Colossus 1 Users: Anthropic | TN | 340 | 275,796 | SpaceXAI |
| Google New Albany Users: Google DeepMind | OH | 333 | 323,901 | |
| Google Columbus Users: Google DeepMind | OH | 303 | 331,986 | |
| Amazon Madison Mega Site Users: Anthropic | n/a | 284 | 214,348 | Amazon |
| CoreWeave Denton TX Users: OpenAI | TX | 262 | 252,653 | CoreWeave |
| QTS Richmond 1 | VA | 238 | 242,547 | n/a |
| Google Bristow Users: Google DeepMind | VA | 238 | 209,861 | |
| Google Council Bluffs (East) Users: Google DeepMind | IA | 237 | 255,685 | |
| Google Omaha Users: Google DeepMind | NE | 237 | 198,585 | |
| Google Papillion Users: Google DeepMind | n/a | 237 | 100,556 | |
| Amazon Ridgeland Users: Anthropic | n/a | 228 | 171,299 | Amazon |
| AWS New Albany | n/a | 213 | 201,112 | Amazon |
| Anthropic Lake Mariner Users: Anthropic | NY | 210 | 292,830 | AI XPV Platform |
| Microsoft Goodyear Users: OpenAI, Microsoft | AZ | 202 | 205,154 | Microsoft |
| Microsoft Project Osmium Users: OpenAI | IA | 190 | 155,634 | Microsoft |
| Google Mesa | n/a | 183 | 289,540 | |
| QTS Richmond 2 | VA | 180 | 182,921 | n/a |
| Meta-QTS Hillsboro 2 Users: Meta | OR | 180 | 123,839 | Meta |
| Meta Rosemount Users: Meta | MN | 178 | 215,675 | Meta |
| Meta Jeffersonville Users: Meta | IN | 178 | 203,133 | Meta |
| Meta Temple Users: Meta | TX | 178 | 202,473 | Meta |
| Google Storey County Users: Google DeepMind | n/a | 161 | 133,906 | |
| Google The Dalles | n/a | 154 | 175,341 | |
| Meta Montgomery Users: Meta | AL | 153 | 174,517 | Meta |
| Meta Kuna Users: Meta | n/a | 152 | 173,248 | Meta |
| Meta Cheyenne Users: Meta | WY | 152 | 172,815 | Meta |
| Meta Huntsville Users: Meta | n/a | 146 | 166,761 | Meta |
| QTS Richmond 3 | VA | 144 | 174,836 | n/a |
| Google Lincoln | NE | 141 | 197,575 | |
| Google Lancaster | OH | 137 | 130,369 | |
| Meta Eagle Mountain Users: Meta | UT | 133 | 151,354 | Meta |
| Coreweave Helios Users: OpenAI | TX | 132 | 159,552 | CoreWeave |
| Meta Aiken Users: Meta | SC | 129 | 147,044 | Meta |