As of October 2026, this list tracks 128 AI labs, from frontier labs that pre-train models from scratch to applied labs that build models into their products. Each row was checked against company announcements, filings and news reporting, and any figure we could not confirm is shown as a dash.
The frontier is narrow, and the money is narrower still. By June 2025, OpenAI, Google DeepMind and Anthropic had built 58% of known models trained with more than 10^25 FLOP (floating-point operations, the standard measure of training compute). Today Anthropic and OpenAI hold 84% of the combined value of the 38 labs with a closed funding round or market cap.
Behind them sit xAI (now part of SpaceX), Meta Superintelligence Labs, Mistral, and a Chinese bloc led by DeepSeek, Moonshot AI and Zhipu's Z.ai.
Below: a quick top 10, the numbers behind the list, what each term means, then ten tables from frontier labs to coding labs.
Top 10 AI labs in 2026: the quick answer
These ten build the general-purpose models most people use. They are grouped, not ranked: US leaders first, then China and Europe.
Anthropic (San Francisco): Claude models and Claude Code; $965B valuation (May 2026).
OpenAI (San Francisco): GPT-6 Astra, ChatGPT and Codex; $852B valuation (March 2026).
Google DeepMind (London): Gemini 4 and open-weight Gemma 4; a unit of Alphabet.
xAI (Palo Alto): Grok 4.7; bought by SpaceX in February 2026 at a reported $250B.
Meta Superintelligence Labs (Menlo Park): Muse Wonder and Muse Spark models; a unit of Meta.
DeepSeek (Hangzhou): open-weight DeepSeek-V4; first outside round of $7.4B at $50B+ (June 2026).
Qwen, Alibaba Tongyi Lab (Hangzhou): the open-weight Qwen3.8 family; a unit of Alibaba.
Moonshot AI (Beijing): open-weight Kimi models, which Cursor and Cognition build their coding models on; $20B valuation.
Z.ai, formerly Zhipu AI (Beijing): open-weight GLM-5.3; listed in Hong Kong in January 2026.
Mistral AI (Paris): Mistral Large 4 and Le Chat; €21B+ valuation (September 2026), Europe's largest lab.
The full list of 128 follows, sorted into ten tables by what each lab builds.
Key numbers
The list is young, American and top-heavy.
59% American. 76 of 128 labs are US-based, 59 of them in the San Francisco Bay Area; the UK (14) and China (11) follow.
Post-ChatGPT. 72 of the 105 labs with a known founding year launched in 2023 or later.
Half open. 57 of the 114 labs with a known open-weights status publish at least some open weights.
Two labs, most of the money. Anthropic and OpenAI hold 84% of the combined value of the 38 labs with a closed funding round or market cap.
LeadWave verification, Oct 21, 2026 · 105 of 128 labs with a known founding or launch year; a few years come from company sites only
LeadWave verification, Oct 21, 2026 · last closed equity round or market cap; Mistral converted from €21B+ at about $1.17/€; AMI Labs is pre-moneyurge AI, AfterQuery, Mechanize and Deeptune sell
What counts as an AI lab, an AI research lab and a frontier AI lab
An AI lab builds or adapts its own models; a frontier lab trains general-purpose models at the edge of capability, which regulators now proxy with training compute above 10^25 FLOP.
Term | Working definition | The evidence behind it |
|---|---|---|
AI lab | An organization that builds or adapts its own AI models as a core part of its work, from frontier labs to applied labs that tune models inside a product | The test used for every row below |
AI research lab | An AI lab where research itself is the output: papers, open models, benchmarks. Often a nonprofit, an academic group or a big-tech unit | Ai2, Biohub and the Swiss AI Initiative are examples in this list |
Frontier AI lab | Trains general-purpose models near the capability frontier | The EU AI Act, Article 51 presumes high-impact capabilities once training compute exceeds 10^25 FLOP. Epoch AI counted 33 known models past that line by June 2025, built by just 12 developers |
Neolab | A research-first startup, often founded by researchers who left established labs and valued in the billions before it ships a product | Radical Ventures counts 29 neolabs with $23B raised, and fewer than 35% have a commercial product (Dealroom, Oct 18, 2026) |
In Epoch AI's June 2025 count, OpenAI (8 models), Google DeepMind (6) and Anthropic (5) built 19 of the 33 models above 10^25 FLOP.
A second roster is self-declared. On May 21, 2024, 16 companies signed the Frontier AI Safety Commitments at the AI Seoul Summit: Amazon, Anthropic, Cohere, Google DeepMind, G42, IBM, Inflection AI, Meta, Microsoft, Mistral AI, Naver, OpenAI, Samsung Electronics, Technology Innovation Institute, xAI and Zhipu AI.
List of frontier AI labs
Twelve labs outside China build general-purpose models at or near the frontier: five are units of big tech, one now sits inside SpaceX, and six are independent, led by Anthropic at a $965B valuation.
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
San Francisco, US | 2021 | Claude models (Opus 5.5, Fable 5.1), Claude Code | $125B raised · $965B val. May 2026 | |
San Francisco, US | 2015 | GPT-6 Astra, ChatGPT, Codex; gpt-oss (open weights) | $122B round · $852B val. Mar 2026 | |
London, UK | 2010 | Gemini 4 Argon, Gemma 4 (open weights) | Unit of Alphabet | |
Palo Alto, US | 2023 | Grok 4.7 | Acquired by SpaceX, Feb 2026 (at $250B, reported) | |
Menlo Park, US | 2025 | Muse Wonder, Muse Spark 1.1; earlier Llama (open weights) | Unit of Meta | |
Redmond, US | 2024 | MAI-Thinking-1, MAI-Code-1.1-Flash, MAI-Voice-2.1 | Unit of Microsoft | |
Seattle, US | 2023 | Nova 2 models, Nova Act agent | Unit of Amazon | |
Santa Clara, US | 1993 (NVIDIA) | Nemotron 3 Ultra (open weights) | Unit of NVIDIA | |
Paris, France | 2023 | Mistral Large 4 (weights due end of Oct 2026), Le Chat | €3B Series D · €21B+ val. Sep 2026 | |
Palo Alto, US | 2024 | No model released; aims straight at superintelligence | $1B+ raised · $32B val. Apr 2025 (reported) | |
San Francisco, US | 2025 | Inkling 975B mixture-of-experts model (open weights), Tinker fine-tuning API | $2B seed · $12B val. Jul 2025 (reported) | |
New York, US | 2024 | Beam 501B mixture-of-experts model (open weights, Apache 2.0) | $2B raised (company, 2025) |
Funding sources: OpenAI, Anthropic, xAI, Mistral, SSI, Thinking Machines, Reflection. Eight of the twelve publish at least some open weights; Anthropic, Microsoft AI and Amazon publish none, and SSI has released nothing.
Open-source AI research labs and enterprise LLM labs
These ten labs build general or task-specific language models below the frontier. Nine publish at least some open weights. Cohere, the largest at a $7B valuation, agreed in September 2026 to merge with Germany's Aleph Alpha.
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
Toronto, Canada | 2019 | Command A+ (open weights), North agent platform | $1.6B raised · $7B val. Sep 2025 | |
Cambridge (MA), US | 2023 | LFM2.5 efficient models (open weights) | $2B val. Dec 2024 | |
US | 2023 | Trinity models (open weights) | $1B pre-money val. Sep 2026 | |
US | 2022 | Hermes models (open weights), Psyche distributed training | $90M raised (company, Oct 2026) | |
Tel Aviv, Israel | 2017 | Jamba models; custom post-training for enterprises | $336M raised | |
Seattle, US | 2014 | Olmo 3, Molmo 2: fully open models and training data | Nonprofit | |
Sunnyvale, US | — | Reka Flash 3.1, Reka Edge multimodal models | $170M+ raised | |
San Francisco, US | 2024 | Cogito v2.1 (open weights) | — | |
Redwood City, US | — | Mercury 3 diffusion LLMs | $50M seed (Nov 2025) | |
San Francisco, US | — | Task-specific developer model (OCR, scraping, extraction); 1.1 Lite open weights | Y Combinator-backed (2026) |
AI21 Labs is the cautionary row. Reported sale talks with Nvidia (which AI21 denied) and Nebius came to nothing, and in 2026 it cut more than 60% of staff and stopped selling standalone models.
Chinese AI labs
China's labs trail the US frontier narrowly and lead it on openness. Eight of these nine release at least some open weights. In Stanford's 2026 AI Index, the top US model led the top Chinese model on the Arena leaderboard by 39 points (1,503 vs 1,464).
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
Hangzhou, China | 2023 | DeepSeek-V4, V4.1-Flash (open weights) | $7.4B first outside round · $50B+ val. Jun 2026 | |
Beijing, China | 2019 | GLM-5.3 (open weights) | HK-listed Jan 2026 · $35.8B mkt cap Oct 2026 | |
Beijing, China | 2023 | Kimi K3 (open weights), Kimi chatbot | $20B val. May 2026 | |
Shanghai, China | 2022 | MiniMax M3 LLM (some open weights), Hailuo video, Speech 2.8 | HK-listed Jan 2026 · $9.2B mkt cap Oct 2026 | |
Hangzhou, China | 2023 | Qwen3.8 (open weights), Qwen3.6-Max, Wan video | Unit of Alibaba | |
Beijing, China | 2023 | Seed2.1 LLM, Seedance 2.5 video, Seedream 5.0 image (some open research models) | Unit of ByteDance | |
Shenzhen, China | 2023 | Hy4-preview (open weights), Hunyuan3D | Unit of Tencent | |
Shanghai, China | 2023 | Step-5-Preview (open weights), StepAudio 3 | — | |
Beijing, China | 2024 | Kling 4.0 video model | Unit of Kuaishou |
The money gap is far wider than the model gap: the same AI Index puts 2025 private AI investment at $285.9B in the US and $12.4B in China, though state-channelled funding is not captured. Public markets have been volatile too: Z.ai's market cap peaked near $128B in June 2026, then fell to about $36B as lock-ups expired.
European, Asian and sovereign AI labs
Sovereign labs build models for one language, country or regulatory regime, and they run on a fraction of frontier budgets: Sakana AI has raised about $412M, roughly 1/300th of Anthropic's total.
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
Tokyo, Japan | 2023 | Fugu orchestration, Namazu Japanese LLM, AI Scientist | $412M raised · $2.7B val. Apr 2026 | |
San Sebastian, Spain | 2019 | CompactifAI compressed LLMs | $1.7B val. (talks) Feb 2026 | |
Heidelberg, Germany | 2019 | Kolibri sovereign model, PhariaAI suite | Merging with Cohere (signed Sep 2026) | |
Bengaluru, India | 2023 | Sarvam 105B and 30B (open weights), Bulbul voice | $41M raised | |
Paris, France | 2023 | Moshi speech dialogue model (open weights) | Nonprofit (donor-funded) | |
Lausanne / Zurich, Switzerland | 2023 | Apertus fully open multilingual LLMs | Public research funding | |
Milan, Italy | — | Domyn Large and Small sovereign LLMs | — | |
Seoul, South Korea | 2025 | Motif-3 315B (open weights) | — | |
Bangkok, Thailand | 2023 | Thai-language models (open weights) | Unit of SCBX Group |
Mistral, Europe's largest lab, sits in the frontier table above. Not every sovereign lab pre-trains from scratch: Typhoon, for one, mostly post-trains open base models for Thai.
AI neolabs: research-first labs, many led by big-lab alumni
Neolabs sell a founder's track record before they sell a product: the 16 below with a verified total have raised $5.6B, and five valued at $1B or more have released no model at all.
Lab | HQ | Founded | Founders | Focus | Funding (latest verified) |
|---|---|---|---|---|---|
San Jose, US | 2025 | Brett Adcock (Figure AI) | Hark Pro assistant, Handoff computer-use model | $800M raised · $6B val. May 2026 | |
London, UK | 2025 | David Silver (ex-DeepMind) | RL 'superlearner' that learns from experience; no model yet | $1.1B raised · $5.1B val. Apr 2026 | |
London / San Francisco | 2025 | Richard Socher, Tim Rocktäschel | Automated AI research; no product yet | $650M raised · $4.7B val. May 2026 | |
San Francisco, US | 2026 | Andi Peng, Georges Harik, Eric Zelikman, Noah Goodman | Ampersand model (API), Persimmon user model | $480M raised · $4.5B val. Jan 2026 | |
Palo Alto, US | 2025 | Anna Goldie, Azalia Mirhoseini (AlphaChip) | AI for chip design; no product yet | $335M raised · $4B val. Jan 2026 | |
Paris, France | 2025 | Yann LeCun, Alexandre LeBrun | World models; no model yet | $1B raised · $3.5B pre-money Mar 2026 | |
Palo Alto, US | 2026 | Joon Sung Park, Michael Bernstein, Percy Liang | Simulations of human behavior | $300M raised · $2B val. Jul 2026 | |
San Francisco, US | 2024 | Eric Ho, Tom McGrath | Interpretability: Silico agent | $200M raised · $1.2B val. Feb 2026 | |
San Francisco, US | 2026 | Behnam Neyshabur, Harsh Mehta (ex-Anthropic) | AI that automates AI R&D; no model yet | $200M raised · $1B val. Jun 2026 | |
Menlo Park, US | 2025 | Liam Fedus (ex-OpenAI), Ekin Dogus Cubuk | Science models plus autonomous labs | $300M raised | |
Palo Alto, US | 2020 | Zuzanna Stamirowska, Jan Chorowski, Adrian Kosowski | BDH, a post-transformer architecture | $30M raised · $500M val. Aug 2026 | |
San Francisco, US | — | Justin Dangel, Alex Whedon | SubQ long-context models | $29M raised · $500M val. (talks) May 2026 | |
San Francisco, US | 2024 | Galen Mead, Devansh Pandey | FDM-1 computer-action model | $75M round (Apr 2026) | |
Palo Alto, US | 2026 | Andrew Dai (ex-DeepMind) | Glimpse visual-reasoning model (preview) | $55M raised · $300M val. Apr 2026 | |
Mountain View, US | 2025 | Shumeet Baluja, Ian Fischer | Reasoning system on top of other labs' models (ARC-AGI-2 results) | $46M raised | |
San Francisco, US | 2024 | Diogo Almeida (ex-OpenAI) | Jev decision model (early access) | $40M raised | |
Pasadena, US | — | Babak Hassibi (Caltech) | Bonsai 1-bit and ternary compressed models (open weights) | $22M raised | |
London, UK | 2025 | Edward Hughes, Louis Kirsch | Faraday AI-scientist agent | — | |
US | — | Eve Bodnia | Kona reasoning model, Aleph formal prover | — | |
London, UK | — | Ex-DeepMind team (per company) | Stealth European frontier lab; nothing public yet | — | |
Beijing, China | — | — | Self-evolving science models (waitlist); ASI-Bench benchmark | — |
Two earlier neolabs, Safe Superintelligence and Thinking Machines Lab, have grown into the frontier table above.
AI-for-science labs
Science and engineering labs train models on proteins, molecules, materials, physics and math. One dwarfs the rest: Jeff Bezos's Prometheus has raised $18.2B, more than three times the ten others here with a disclosed total combined.
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
San Francisco, US | 2025 | Jeff Bezos and Vik Bajaj; AI for physical engineering, no product yet | $18.2B raised · $41B val. Jun 2026 | |
Cambridge (MA), US | 2023 | Flagship Pioneering spinout; AI Science Factory autonomous labs | $550M raised · $8.5B val. (talks) Jun 2026 | |
San Francisco, US | 2024 | Chai-3 molecular design model | $630M raised · $3.8B val. Jul 2026 | |
London, UK | 2021 | Demis Hassabis; IsoDDE drug design engine, AlphaFold 3 | $2.7B raised | |
London, UK | — | Large Physics Models for engineering | $467M+ raised · $2.4B val. Jun 2026 | |
Palo Alto, US | 2023 | Vlad Tenev, Tudor Achim; Aristotle math reasoning model | $295M raised · $1.4B val. Nov 2025 | |
London, UK | 2019 | EDEN biology foundation models | $225M raised | |
San Francisco, US | 2025 | Kosmos AI scientist | $70M raised | |
London, UK | — | Simon Kohl (ex-AlphaFold); Latent-X2 antibody design | $50M raised | |
London, UK | 2022 | Orb atomistic simulation model | $50M Series B (May 2026) | |
Palo Alto, US | — | AI-designed chips and chip-design models | $24M raised | |
London, UK | 2022 | Generative physics model | $18M raised | |
Palo Alto, US | 2026 | Jeff Dean, Sanjay Ghemawat, Quoc Le, Oriol Vinyals; automated discovery | Initial round, size undisclosed (Aug 2026) | |
Redwood City, US | 2016 | ESM protein models (absorbed EvolutionaryScale, Nov 2025) | Nonprofit | |
San Francisco, US | 2024 | Odyssey protein models | — | |
Freiburg, Germany | 2024 | TabPFN tabular foundation model | Acquired by SAP (Jul 2026) |
London is the second science hub after the Bay Area: six of these 16 labs are based there.
Robotics and world-model labs
Physical-AI labs build robot brains and world models. Investors now value the leaders in the billions: Skild AI closed at $14B in January 2026, and AMD agreed in September to buy Fei-Fei Li's World Labs for about $8.2B.
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
Pittsburgh, US | 2023 | Deepak Pathak, Abhinav Gupta; Skild Brain general robot model | $1.8B raised · $14B val. Jan 2026 | |
San Francisco, US | 2024 | Karol Hausman, Sergey Levine, Chelsea Finn; π0.7 robot model | $1.1B raised · $11B val. (talks) Mar 2026 | |
Irvine, US | 2023 | Field Foundation Models for robots | $405M raised · $10B val. (talks) Oct 2026 | |
San Francisco, US | 2024 | Fei-Fei Li; Marble world model | $1.2B raised · AMD deal at $8.2B (pending) | |
New York, US | 2025 | MIRA playable world model | $674M raised · $6.2B val. Sep 2026 | |
Tel Aviv / San Francisco | 2023 | Real-time video and world models (Lucy, Oasis) | $450M raised · $4B val. May 2026 | |
Palo Alto, US | 2025 | Rivian spinout (RJ Scaringe); industrial robot models | $1B raised · $3B val. May 2026 | |
Palo Alto, US | 2023 | Odyssey-3 world model | $337M raised · $1.4B val. Jun 2026 | |
Redwood City, US | 2024 | Dyna-2.1 robot model | $144M raised · $600M val. (talks) Sep 2025 | |
San Mateo, US | — | GEN-1.5 embodied foundation model | $400M round (Jun 2026); $500M+ raised | |
Stuttgart, Germany | 2021 | Cortex robot foundation model | $116M Series B (Jul 2026) | |
Munich, Germany | 2020 | Roman Hölzl; Alfie autonomous industrial robot, RobFlow software | $100M Series C (Feb 2026) | |
Zurich, Switzerland | — | ETH Zurich spinout; video-action models for robot hands | $18M raised | |
Bellevue, US | 2024 | Ex-Meta FAIR team; Isaac embodied model (open weights) | — | |
Munich, Germany | 2025 | Humanoid robot (in development) | — | |
Munich, Germany | — | Real-world data and models for embodied AI | — | |
Toronto, Canada | — | World models that simulate physical reality | — |
Covariant, often listed here, is left out: Amazon licensed its models and hired its founders and about a quarter of staff in 2024.
Image, video and 3D labs
Generative-media companies build or serve image, video and 3D models. The most famous never raised a dollar: Midjourney has no outside investors, while Runway has raised about $860M at a $5.3B valuation.
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
San Jose, US | 2023 | Firefly image, video and vector models inside Photoshop and Express | Unit of Adobe | |
New York, US | 2018 | Gen-4.5 video, GWM-1 world model | $860M raised · $5.3B val. Feb 2026 | |
San Francisco, US | — | Inference platform serving 600+ image, video, audio and 3D models | ~$250M round · $4B+ val. Oct 2025 (reported) | |
Palo Alto, US | 2021 | Ray3.2 video model, Dream Machine | $900M Series C · $4B val. Nov 2025 | |
Freiburg, Germany | 2024 | FLUX 3 image and video models (some open weights) | $331M+ raised · $3.25B val. Dec 2025 | |
Bay Area, US | 2023 | Pika 2.5 video model | $135M raised | |
Los Angeles, US | 2023 | Marey video model (licensed training data) | ~$124M raised (May 2025) | |
San Francisco, US | 2021 | Marengo and Pegasus video understanding (not generation) | $100M Series B (Jul 2026) | |
Toronto, Canada | 2022 | Ideogram 4.5 image model; 4.0 open weights | $96M raised | |
San Francisco, US | 2022 | Krea 3 image model | $83M raised | |
London, UK | 2019 | Stable Diffusion 3.5, Stable Audio 3.0 (open weights) | $76M Series B (Aug 2026) | |
London, UK | 2022 | Recraft V4 image and vector model | $42M raised | |
San Francisco, US | 2020 | Phoenix-4.5 video avatar models | $18M Series A (Mar 2024) | |
San Francisco, US | — | Arrow 2 SVG models | $8.3M raised | |
San Francisco, US | 2021 | Midjourney V8.2 image model, V1 video | $0 (never raised outside capital) | |
Palo Alto, US | — | Reve Image model | — | |
Sunnyvale, US | 2023 | Meshy 7 3D generation | — | |
Beijing, China | 2023 | Tripo 3D generation | — | |
— | — | 3D model generation and 3D-printing platform | — |
Kling (Kuaishou), MiniMax's Hailuo and Alibaba's Wan, three widely used video models, appear in the Chinese labs table above.
Voice and audio labs
Voice labs train text-to-speech (TTS), speech-to-text (STT) and sound models. ElevenLabs has pulled far ahead: a $22B valuation after a September 2026 share sale, on $781M raised.
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
London, UK | 2022 | Eleven v3 voice, Eleven Music | $781M raised · $22B val. Sep 2026 | |
Paris, France | 2025 | Real-time speech models (Kyutai spinout) | $100M raised | |
San Francisco, US | 2023 | Sonic-3 TTS, Ink-2 STT | $91M+ raised | |
US | — | Enterprise phone-call voice agents | $65M raised through Series B (Feb 2025) | |
Palo Alto, US | 2025 | Fish Audio S2.1 TTS | $52M seed (Jul 2026) | |
Berlin, Germany | 2023 | Mirelo SFX: sound effects and music for video | $44M raised | |
San Francisco, US | 2022 | Arcana and Mist TTS | $29.5M+ raised | |
San Francisco, US | 2023 | Lightning TTS, Pulse STT | $21M raised | |
Mountain View, US | 2021 | Realtime TTS-2 voice models | $500M val. Aug 2023 (last reported) | |
Istanbul / San Francisco | 2025 | Freya TTS for financial voice agents | — | |
Foster City, US | — | Soniox v5 speech-to-text | — |
Coding labs
Four coding labs build their own models, in two different ways. Poolside and Magic pre-train from scratch; Cursor and Cognition post-train Moonshot's open Kimi models inside their products, and are worth far more.
Lab | HQ | Founded | Known for | Funding (latest verified) |
|---|---|---|---|---|
San Francisco, US | 2022 | Cursor editor and agents; Composer 2.5 built on Kimi K2.5 | Acquired by SpaceX, Aug 2026 ($60B in stock) | |
San Francisco, US | 2023 | Devin agent and Windsurf IDE; SWE-2 model post-trained on Kimi K3 | $2B+ Series E · $48B val. Sep 2026 | |
San Francisco, US | 2023 | Laguna coding models (open weights) | $626M raised · $12B val. (talks) Oct 2025 | |
San Francisco, US | — | LTM-2-mini, 100M-token-context code model | $768M raised |
How this list was built
Every row was checked on October 21, 2026. We used the lab's own announcements, regulatory filings and news reporting (Reuters, Bloomberg, FT, TechCrunch, CNBC, SCMP and peers), and treated trackers and aggregators only as leads. Five rows (Hi3D, MicroAGI, Veeda AI, Cursive and Apex Intelligence) are documented only by their own websites, so most of their cells are dashes.
Test for inclusion: the organization builds or adapts its own AI models as a core part of its work. That covers frontier labs that pre-train from scratch, research labs, and applied labs that post-train models or serve them inside a product, such as Cursor, Cognition, Adobe Firefly and fal.
Starting point: 138 AI companies from the LeadWave list of SaaS companies. Products were rolled into their parent lab (Claude into Anthropic, Gemini into Google DeepMind, Grok into xAI), and 14 labs missing from the database were added, such as ByteDance Seed, Skild AI and Periodic Labs.
Funding column: the latest figure we could confirm, either a total or the latest round when no total is public. "Talks" means reported but not closed; listed companies show market cap; a dash means nothing passed our source bar.
FAQ
What is a frontier AI lab? A lab that trains general-purpose models near the edge of capability. The EU AI Act uses 10^25 FLOP (floating-point operations) of training compute as its threshold. By June 2025 only 12 developers had crossed it, led by OpenAI, Google DeepMind and Anthropic.
How many AI labs are there? This list counts 128 organizations that build or adapt their own AI models, as of October 2026. A broader automated census by Keenable SELECT found 90 labs founded between 2022 and September 2026 alone.
What is the biggest AI lab? By last closed valuation, Anthropic at $965B (May 2026), ahead of OpenAI at $852B (March 2026). Google DeepMind is not in that ranking because, as a unit of Alphabet, it has no separate valuation.
Which AI labs release open weights? 57 of the 114 labs here with a known open-weights status publish at least some open weights. Chinese labs lead: 8 of the 9 in our China table, including DeepSeek, Qwen, Moonshot and Z.ai.
Which country has the most AI labs? The US, with 76 of 128 (59%), and 59 of those sit in the San Francisco Bay Area. The UK follows with 14 and China with 11.
Which AI labs are in Europe? 30 of the 128, led by Mistral AI and AMI Labs (Paris), Google DeepMind, ElevenLabs and Ineffable Intelligence (London), and Black Forest Labs (Freiburg).
What is an AI neolab? A research-first startup, often founded by researchers who left established labs, such as Ineffable Intelligence, AMI Labs or Recursive. Many raise at billion-dollar valuations before shipping a product.
What is an AI research lab? An AI lab whose main output is research: papers, open models and benchmarks. Ai2, Kyutai and the Swiss AI Initiative are nonprofit or public examples; Google DeepMind and ByteDance Seed are corporate ones.
What are the biggest AI labs in 2026? By last closed valuation or market cap, the five largest independent labs are Anthropic ($965B), OpenAI ($852B), DeepSeek ($50B), Cognition ($48B) and Prometheus ($41B).
About this list
Curated by Khrystyna Oleksiienko at LeadWave, which maintains a hand-researched database of SaaS and AI companies.
