Companies
Last reviewed
May 9, 2026
Sources
20 citations
Review status
Source-backed
Revision
v2 ยท 4,893 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 9, 2026
Sources
20 citations
Review status
Source-backed
Revision
v2 ยท 4,893 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: Organizations, Humanoid robot manufacturers, Foundation models
Companies in the field of artificial intelligence span foundation model developers, cloud platforms, semiconductor manufacturers, vector database vendors, developer tooling startups, image and video generation studios, consumer chatbot makers, robotics firms, and vertical specialists serving regulated industries such as defense, biotech, and law. This gateway article indexes the most active firms across each segment as of 2026.
The modern AI industry began to take its current shape in the mid-2010s, when deep learning breakthroughs in image recognition and machine translation drew large investments to a small set of research labs. The 2017 publication of the Transformer architecture by Google researchers, the 2018 release of BERT, and the 2019 announcement of GPT-2 by OpenAI shifted the center of gravity from academic labs to industrial labs with access to large compute clusters. The November 2022 launch of ChatGPT accelerated commercial activity dramatically. By 2026 the global AI sector counts dozens of unicorns, several hundred well-funded startups, and a handful of publicly traded firms whose valuations are dominated by AI revenue or AI-related capital expenditure.
This article groups firms by their primary line of business. Many large companies appear in multiple segments because their AI activity spans research, infrastructure, and applications. For example, Google operates a foundation model lab through Google DeepMind, an AI cloud through Google Cloud, and consumer products such as Gemini. Where a parent firm has a distinct AI subsidiary, the subsidiary appears in the relevant section.
Foundation model labs train large general-purpose neural networks, typically large language models, and license the resulting systems through APIs, partnerships, or open weights. The category is dominated by a small group of well-capitalized firms, several of which have raised more than ten billion dollars in cumulative funding. Most of these labs operate at a loss while spending billions per year on compute.
| Company | Founded | Headquarters | Focus | Notable models |
|---|---|---|---|---|
| OpenAI | 2015 | San Francisco, California | Frontier general models, ChatGPT, API | GPT-4, GPT-5, o1, o3, Sora |
| Anthropic | 2021 | San Francisco, California | Constitutional AI, safety research, Claude | Claude 3, Claude 3.5 Sonnet, Claude 4, Claude Opus 4 |
| Google DeepMind | 2010 (DeepMind), merged 2023 | London, United Kingdom | Frontier research, Gemini, AlphaFold | Gemini, Gemini 2.5, AlphaFold, AlphaGo |
| Meta AI | 2013 (FAIR) | Menlo Park, California | Open weights LLMs, computer vision, AR | Llama 2, Llama 3, Llama 4, SAM |
| xAI | 2023 | Palo Alto, California | Grok chatbot, X integration, Colossus cluster | Grok 1, Grok 2, Grok 3, Grok 4 |
| Mistral AI | 2023 | Paris, France | Open weight LLMs, European sovereignty | Mistral 7B, Mixtral 8x7B, Mistral Large |
| DeepSeek | 2023 | Hangzhou, China | Cost-efficient open weight LLMs, reasoning | DeepSeek-V3, DeepSeek-R1 |
| Cohere | 2019 | Toronto, Canada | Enterprise LLMs, retrieval, embeddings | Command R, Command R+, Embed |
| AI21 Labs | 2017 | Tel Aviv, Israel | Jurassic LLMs, Wordtune, enterprise | Jamba, Jurassic-2 |
| Inflection AI | 2022 | Palo Alto, California | Pi assistant, enterprise pivot in 2024 | Inflection-1, Inflection-2.5 |
| Reka AI | 2022 | San Francisco, California | Multimodal frontier models | Reka Core, Reka Flash |
| Adept AI | 2022 | San Francisco, California | Action models, software automation | Fuyu, Persimmon |
| Liquid AI | 2023 | Cambridge, Massachusetts | Liquid neural networks, MIT spinoff | LFM-1B, LFM-3B, LFM-40B |
| Sakana AI | 2023 | Tokyo, Japan | Evolutionary model merging, Japanese LLMs | EvoLLM-JP, Sakana AI Scientist |
| Aleph Alpha | 2019 | Heidelberg, Germany | European sovereign LLMs, multilingual | Luminous, Pharia |
| Zhipu AI | 2019 | Beijing, China | GLM family, Tsinghua spinoff | GLM-4, ChatGLM, CogVideoX |
| Moonshot AI | 2023 | Beijing, China | Long context, Kimi chat | Kimi K1, K1.5, K2 |
| Baichuan Intelligence | 2023 | Beijing, China | Chinese-English bilingual LLMs | Baichuan-7B, Baichuan-13B, Baichuan2 |
| 01.AI | 2023 | Beijing, China | Yi family, founded by Kai-Fu Lee | Yi-34B, Yi-Large, Yi-Vision |
| MiniMax | 2021 | Shanghai, China | Conversational and creative AI | abab series, Talkie, Hailuo |
| StepFun | 2023 | Shanghai, China | Trillion-parameter dense models | Step-1, Step-2, Step-Video |
| EleutherAI | 2020 | Distributed | Open research collective, GPT-Neo, Pythia | GPT-Neo, GPT-J, Pythia |
| Hugging Face | 2016 | New York City | Open model hub, datasets, libraries | BLOOM (with BigScience), HuggingChat |
Many of these labs publish frontier research alongside their commercial APIs. OpenAI reached an estimated valuation around $300 billion in late 2024 and crossed $500 billion in 2025 secondary share transactions, while Anthropic closed a series F in 2025 at a $61.5 billion valuation. Google DeepMind operates as the consolidated AI research arm of Alphabet following the 2023 merger of Google Brain and DeepMind. xAI, founded by Elon Musk, raised more than $11 billion across 2024 and 2025 and runs the Colossus supercomputer in Memphis, Tennessee, which by mid-2025 contained roughly 200,000 NVIDIA H100 and H200 accelerators.
The Chinese cluster of DeepSeek, Zhipu AI, Moonshot AI, Baichuan, 01.AI, and MiniMax is sometimes called the "AI Tigers" of China. DeepSeek drew international attention in January 2026 when its DeepSeek-R1 reasoning model demonstrated competitive benchmark performance at a fraction of the training cost of comparable Western models.
The AI cloud and platform layer hosts foundation models, provides managed inference, exposes data and analytics integrations, and sells GPU capacity. The hyperscalers dominate by revenue, but a layer of "neoclouds" specializes in dense GPU rentals at lower prices.
| Company | Founded | Headquarters | AI offerings |
|---|---|---|---|
| Microsoft | 1975 | Redmond, Washington | Azure OpenAI Service, Microsoft Copilot, Microsoft 365 Copilot, Azure AI Foundry |
| 1998 | Mountain View, California | Vertex AI, Gemini API, Google AI Studio, TPUs | |
| Amazon Web Services | 2006 | Seattle, Washington | Amazon Bedrock, Amazon SageMaker, Amazon Q, Trainium |
| Oracle Cloud | 2016 | Austin, Texas | OCI GenAI, exclusive Stargate hosting |
| Alibaba Cloud | 2009 | Hangzhou, China | Model Studio, Qwen family |
| Tencent | 1998 | Shenzhen, China | Tencent Cloud TI, Hunyuan |
| Huawei AI | 2019 | Shenzhen, China | Pangu, Ascend NPU cloud |
| Baidu AI | 2014 | Beijing, China | ERNIE Bot, Qianfan |
| Naver Cloud | 2017 | Seongnam, South Korea | HyperCLOVA X |
| IBM watsonx | 2023 | Armonk, New York | watsonx.ai, Granite models |
| Salesforce AI | 2014 | San Francisco, California | Agentforce, Einstein |
| Databricks | 2013 | San Francisco, California | Mosaic AI, DBRX, Lakebase |
| Snowflake AI | 2012 | Bozeman, Montana | Cortex, Snowflake Arctic |
| Nebius | 2023 | Amsterdam, Netherlands | GPU cloud (ex-Yandex) |
| CoreWeave | 2017 | Roseland, New Jersey | Specialized GPU cloud |
| Crusoe | 2018 | Denver, Colorado | AI data centers, Stargate sites |
| Lambda Labs | 2012 | San Jose, California | Lambda Cloud GPU |
| RunPod | 2022 | Wilmington, Delaware | Serverless GPU pods |
| Replicate | 2019 | San Francisco, California | Hosted open model APIs |
| Fireworks AI | 2022 | Redwood City, California | Optimized open model inference |
| DeepInfra | 2022 | Palo Alto, California | Open model serving |
| Modal | 2021 | New York City | Serverless Python, GPU bursting |
| Anyscale | 2019 | San Francisco, California | Ray distributed AI |
| Baseten | 2019 | San Francisco, California | Model serving |
The three Western hyperscalers, Microsoft Azure, Google Cloud, and Amazon Web Services, each generated more than $80 billion in 2025 cloud revenue, with AI workloads cited by all three as the leading driver of growth. Microsoft holds a major equity and exclusivity arrangement with OpenAI extending through 2030, although Microsoft's 2025 settlement with OpenAI removed its right of first refusal on new compute and lifted certain capacity caps. Amazon committed up to $8 billion in Anthropic by late 2024 and built dedicated Trainium clusters for Anthropic's training runs. Google operates its own TPU hardware and runs frontier training on TPU v5p and TPU v6 "Trillium" pods.
The neocloud cohort, including CoreWeave, Crusoe, Lambda Labs, Nebius, and RunPod, grew rapidly during the 2023 to 2025 GPU shortage. CoreWeave went public on Nasdaq in March 2025 with a debut market capitalization above $20 billion. Crusoe partnered with Oracle and OpenAI on the first Stargate sites in Abilene, Texas. Nebius emerged from the post-sanction reorganization of Yandex and listed on Nasdaq in late 2024.
AI hardware spans GPUs, custom accelerators, AI-specific ASICs, neuromorphic chips, and the foundries that fabricate them. Nvidia holds a dominant position in training accelerators, while several startups target inference workloads with novel architectures.
| Company | Founded | Headquarters | AI hardware focus |
|---|---|---|---|
| Nvidia | 1993 | Santa Clara, California | Hopper H100/H200, Blackwell B200/GB200/B300, Rubin generation, CUDA |
| AMD | 1969 | Santa Clara, California | Instinct MI300X, MI325X, MI355X, ROCm |
| Intel | 1968 | Santa Clara, California | Gaudi 3, Xeon AI extensions, IFS foundry |
| TSMC | 1987 | Hsinchu, Taiwan | Foundry for Nvidia, AMD, Apple, on N4P, N3, N2 |
| Apple | 1976 | Cupertino, California | Apple Silicon, Apple Neural Engine, Private Cloud Compute |
| Qualcomm | 1985 | San Diego, California | Snapdragon X Elite, Hexagon NPU, AI200/AI250 data center chips |
| Samsung | 1969 (electronics) | Suwon, South Korea | HBM3E and HBM4 memory, Exynos NPUs, foundry |
| SK hynix | 1983 | Icheon, South Korea | HBM3E, HBM4 memory |
| Cerebras Systems | 2015 | Sunnyvale, California | Wafer scale WSE-3 |
| Groq | 2016 | Mountain View, California | LPU deterministic inference |
| SambaNova Systems | 2017 | Palo Alto, California | RDU, SN40L |
| Tenstorrent | 2016 | Toronto, Canada | RISC-V AI accelerators |
| Cambricon | 2016 | Beijing, China | MLU chips |
| Huawei HiSilicon | 1991 | Shenzhen, China | Ascend 910B, 910C |
| Graphcore | 2016 | Bristol, UK | IPU, SoftBank-owned |
| Etched | 2022 | Cupertino, California | Sohu transformer ASIC |
| Rebellions | 2020 | Seongnam, South Korea | ATOM, REBEL |
| Foxconn | 1974 | New Taipei, Taiwan | AI server manufacturing |
Nvidia is the largest AI hardware company by revenue and by market capitalization. Following the launches of H100, H200, B200, and the GB200 NVL72 and GB300 rack-scale systems, the company crossed a $4 trillion market capitalization in mid-2025, the first publicly traded firm to do so. Its data center revenue alone exceeded $115 billion in fiscal 2025 according to filings with the U.S. Securities and Exchange Commission.
AMD closed the gap somewhat with the MI300X, MI325X, and MI355X accelerators and a series of multi-billion dollar wins at Microsoft, Meta, and Oracle. In October 2025 OpenAI announced an agreement to deploy roughly 6 gigawatts of AMD Instinct GPUs starting in 2026, paired with warrants for up to 10 percent of AMD shares. TSMC fabricates the leading-edge silicon for nearly every Western AI hardware vendor, with Apple usually receiving first access to each new node such as N3 in 2023 and N2 in 2025.
The inference startup cohort, including Cerebras, Groq, SambaNova, and Tenstorrent, competes on tokens per second per watt rather than raw FLOPS. Cerebras's WSE-3 integrates 4 trillion transistors on a single wafer. Groq's LPU reaches very low first-token latency for transformer inference. Tenstorrent, led by Jim Keller, licenses RISC-V cores and ships Wormhole and Blackhole boards.
Vector databases store high-dimensional embeddings and serve approximate nearest neighbor queries, the core operation behind retrieval-augmented generation and semantic search. The category emerged around 2019 and consolidated quickly as enterprises adopted RAG patterns to ground LLM outputs in proprietary data.
Standalone vector databases: Pinecone (managed serverless), Weaviate (open source, hybrid search), Qdrant (Rust engine), Chroma, Milvus (by Zilliz), LanceDB (Lance columnar format), Turbopuffer (object-storage backed).
Embedding models: Voyage AI, acquired by MongoDB in February 2025 for approximately $220 million.
Database extensions: pgvector (PostgreSQL), MongoDB Atlas Vector Search, Elasticsearch dense_vector. Most large enterprises now run vector search inside an existing database. The standalone vendors compete on scale, recall accuracy, multi-tenancy, and integrated metadata filtering. Pinecone launched its serverless tier in early 2024 and reported tens of thousands of paying customers by 2025.
AI developer tools include code editors with embedded models, agentic coding agents, observability platforms, evaluation frameworks, prompt engineering tools, and orchestration libraries. The segment grew explosively after the 2021 launch of GitHub Copilot and the 2022 launch of ChatGPT.
Code editors and IDEs: Anysphere (maker of Cursor), Windsurf (formerly Codeium, now part of Cognition AI), Replit, Augment Code.
Autonomous coding agents: Cognition AI (Devin), Factory (AI company) (Droids), Poolside AI (custom code models).
Frameworks and libraries: Hugging Face (Transformers, Datasets, model hub), LangChain (LangChain, LangGraph), LlamaIndex (RAG framework), Vellum.
Observability and evaluation: LangSmith, Langfuse, Helicone, Patronus AI, Arize Phoenix, Cleanlab.
Agent tooling and memory: Mem0, Composio, Toolhouse, Letta (MemGPT).
Enterprise search: Glean (company).
In 2025 the agentic coding subsegment crossed several billion dollars in annualized revenue. Anysphere, the maker of Cursor, reported more than $500 million in annual recurring revenue by mid-2025 and raised at a valuation of $9.9 billion in May, then a higher valuation later in the year. Cognition AI, creator of the Devin agent, acquired Windsurf (formerly Codeium) in July 2025 in a deal that closed after Google hired Windsurf's CEO and several engineers. Replit reached a $3 billion valuation in late 2024 and approximately $1 billion ARR by mid 2025 after pivoting around its Agent product.
Generative media spans image, video, music, voice, and 3D synthesis. The segment is highly fragmented because models for each modality have different cost and quality frontiers, and consumer demand favors specialized apps over general platforms.
Image generation: Midjourney, Stability AI (Stable Diffusion), Black Forest Labs (FLUX.1), Krea AI, Ideogram, Recraft AI, Magnific AI (Freepik-owned), Leonardo.AI (Canva-owned), Adobe Firefly.
Video generation: Runway (company) (Gen-3, Gen-4), Pika (video generation), Luma AI (Dream Machine), Higgsfield AI, Kling (video generation) (Kuaishou), Sora (OpenAI).
Music and voice: ElevenLabs (voice cloning), Cartesia (Sonic real-time voice), Suno (music), Udio (music), AssemblyAI (speech-to-text).
Avatars and video translation: HeyGen, Synthesia, D-ID, Tavus, Colossyan.
Midjourney, notable for being self-funded and never having taken venture capital, reportedly reached more than $200 million in revenue by 2024 with a small team. Black Forest Labs was founded in 2024 by several alumni of Stability AI, including Robin Rombach, and shipped FLUX.1 image models that have been adopted across many downstream apps. Runway raised at a $3 billion valuation in 2025 after the launch of Gen-4. ElevenLabs reached an $11 billion valuation in 2025 secondaries.
Music generation became a flashpoint in 2024 when the major record labels filed copyright lawsuits against Suno and Udio over training data sourced from copyrighted recordings. By late 2025 several labels had begun licensing negotiations with both companies.
Consumer AI applications include chatbots, search engines, AI companions, browsers, and on-device assistants. Many of these companies sit atop foundation models from third parties; others train their own.
Chatbots and assistants: ChatGPT (OpenAI), Claude.ai (Anthropic), Gemini (Google), Microsoft Copilot, Grok (xAI), Meta AI (Meta).
Answer engines and AI search: Perplexity (Comet browser), You.com.
AI companions: Character.AI, Replika, Inflection AI (Pi, now B2B), Friend (AI device), Sesame (AI company).
Writing and marketing: QuillBot, Jasper AI, Writer (AI company), Copy.ai, Writesonic, Lindy.
AI hardware devices: Humane AI Pin (sold to HP and discontinued in 2025), Rabbit R1.
ChatGPT is the largest consumer AI service by user base, reporting more than 700 million weekly active users by mid 2025. Perplexity closed a Series F at an $18 billion valuation in late 2025 after launching its Comet browser. Character.AI reached a non-exclusive licensing deal with Google in August 2024 in which Google paid for the right to use the Character technology and hired co-founders Noam Shazeer and Daniel De Freitas, leaving the company in operation under new leadership.
The consumer hardware bets have been mixed. Humane sold its remaining business to HP in February 2025 and discontinued the AI Pin. Rabbit shipped firmware updates for the R1 through 2025 but has not reached the volumes its early order book suggested. Smart glasses, in contrast, have grown quickly: Meta reported more than two million units of Ray-Ban Meta sold by mid 2025.
Robotics firms developing AI-driven systems include classical industrial robotics vendors, autonomous vehicle developers, drone makers, surgical robotics specialists, and the rapidly expanding humanoid cohort. The humanoid wave grew sharply between 2022 and 2026 thanks to better foundation models, cheaper compute, and improved hardware.
| Company | Founded | HQ | Platform |
|---|---|---|---|
| Boston Dynamics | 1992 | Waltham, Massachusetts | Atlas, Spot, Stretch (Hyundai-owned) |
| Tesla (robotics) | 2003 | Palo Alto, California | Optimus Gen 2, Gen 3 |
| Figure AI | 2022 | Sunnyvale, California | Figure 02, Figure 03 |
| Agility Robotics | 2015 | Tangent, Oregon | Digit |
| 1X Technologies | 2014 | Moss, Norway | NEO, EVE |
| Sanctuary AI | 2018 | Vancouver, Canada | Phoenix |
| Apptronik | 2016 | Austin, Texas | Apollo |
| Unitree Robotics | 2016 | Hangzhou, China | H1, G1 |
| AgiBot | 2023 | Shanghai, China | A2, A2-W |
| XPENG Robotics | 2020 | Guangzhou, China | Iron |
| Kepler Robotics | 2023 | Shanghai, China | Forerunner K2 |
| Fourier Intelligence | 2015 | Shanghai, China | GR-1, GR-2 |
| LimX Dynamics | 2022 | Shenzhen, China | CL-1, P1 |
| UBTECH Robotics | 2012 | Shenzhen, China | Walker, Walker S |
| EngineAI | 2023 | Shenzhen, China | SE01, SA01 |
| Booster Robotics | 2023 | Beijing, China | T1, T2 |
| Robotera | 2023 | Beijing, China | XBot-L, Star1 |
| MagicLab | 2023 | Suzhou, China | MagicBot |
| PNDbotics | 2023 | Hefei, China | Adam |
| Mentee Robotics | 2022 | Herzliya, Israel | Menteebot |
| Pollen Robotics | 2016 | Bordeaux, France | Reachy (Hugging Face-owned) |
| Engineered Arts | 2004 | Penryn, UK | Ameca |
| Hanson Robotics | 2013 | Hong Kong | Sophia |
| NEURA Robotics | 2019 | Metzingen, Germany | 4NE-1 |
| Humanoid (company) | 2024 | London, UK | HMND 01 |
| Cartwheel Robotics | 2023 | San Francisco, California | Yogi |
| Foundation | 2023 | Vancouver, Canada | Phantom MK1 |
| Skild AI | 2023 | Pittsburgh, Pennsylvania | Robot foundation model |
| Physical Intelligence | 2024 | San Francisco, California | Pi-zero |
Quadrupeds and inspection: DEEP Robotics, ANYbotics, Unitree.
Drones: DJI (consumer and enterprise drones), Skydio (autonomous inspection).
Defense autonomy: Anduril (Lattice OS), Shield AI (V-BAT), Saronic (autonomous vessels).
Autonomous vehicles and trucking: Waymo, Zoox (Amazon-owned), Cruise (GM-owned), Wayve, Tesla (FSD, Cybercab), Aurora Innovation, Kodiak Robotics, Plus.
Warehouse, service, and manufacturing robotics: Covariant (company), Symbotic, Amazon Robotics, Pudu Robotics, KEENON Robotics, SoftBank Robotics, Hyundai Motor Group (owns Boston Dynamics), Toyota Research Institute, Honda (robotics), Doosan Robotics, Rainbow Robotics, Naver Labs.
The humanoid segment received unprecedented funding between 2024 and 2026. Figure AI raised more than $675 million in a Series B in February 2024 led by Microsoft and including OpenAI, Nvidia, Jeff Bezos, and others, then closed a much larger round at a $39.5 billion valuation in late 2025. 1X Technologies, backed by OpenAI and Tiger Global, unveiled its NEO humanoid for home use in 2025. Tesla demonstrated Optimus Gen 3 in late 2025 with new actuators and a redesigned hand. Chinese vendors led by AgiBot, Unitree, UBTECH, XPENG Robotics, Booster Robotics, and EngineAI published rapid hardware iterations and competed at events such as the World Robot Conference and the World Humanoid Robot Games held in Beijing in August 2025.
In autonomous vehicles, Waymo operates fully driverless rides in Phoenix, San Francisco, Los Angeles, Austin, Atlanta, Miami, Washington, and other cities, with weekly trip volume exceeding 250,000 by mid 2025. Tesla launched its first Robotaxi pilot in Austin in June 2025. Cruise, majority owned by General Motors, wound down its commercial robotaxi service in late 2024 following safety incidents and was reorganized as part of GM's broader autonomy program. Zoox began limited public riders in Las Vegas and Austin in 2025.
Vertical AI firms apply machine learning to a single domain, often using proprietary data and tightly regulated workflows. Many compete with horizontal foundation model providers by offering data, fine tuning, and enterprise software for specific industries.
Defense and dual-use: Anduril (Lattice OS, Roadrunner), Palantir (AIP for defense and commercial), Shield AI (autonomous V-BAT and Hivemind).
Drug discovery and healthcare: Recursion Pharmaceuticals (merged with Exscientia in 2024), Insitro, Isomorphic Labs (DeepMind spinoff), Hippocratic AI (patient-facing agents), Tempus (precision medicine), Neuralink (brain-computer interface).
Legal and finance: Harvey, Hebbia, Clio, Klarity.
Customer service and support agents: Sierra AI (founded by Bret Taylor), Decagon, Moveworks (acquired by ServiceNow in 2025), Ada Support, Cresta.
Voice agents: Bland AI, Vapi, Retell AI, PolyAI, Hume AI.
Workflow and HR: Gumloop (no-code agents), Mercor (AI hiring), Glean (company) (enterprise search).
Industrial and computer vision: Hexagon (company), Hikvision, SenseTime, Megvii, Yitu Technology, CloudWalk Technology, iFlytek.
Defense and dual-use AI saw very large funding rounds in 2024 and 2025. Anduril raised $2.5 billion at a $30.5 billion valuation in August 2025. Palantir crossed a $400 billion market capitalization in late 2025 on strong U.S. Department of Defense and commercial AIP adoption. The biotech AI cluster benefited from milestone payments and improved AlphaFold inspired methods.
A handful of venture firms, sovereign wealth funds, and corporate investors recur across most of the rounds in the segments above. Andreessen Horowitz, Sequoia Capital, Tiger Global, Lightspeed, Founders Fund, Khosla Ventures, and General Catalyst lead the venture side. SoftBank Group Vision Fund, Saudi PIF, and Mubadala participate as sovereign and corporate balance sheet investors. Strategic investors include Microsoft (OpenAI, Mistral, Figure), Nvidia (Figure, Cohere, CoreWeave, Wayve, Recursion, Reka, Mistral), Google (Anthropic, Runway, Character.AI license), and Amazon (Anthropic, Hugging Face).
The U.S. Stargate program announced in January 2025 by OpenAI, SoftBank, Oracle, and MGX targets up to $500 billion of AI infrastructure spending over four years, with initial sites in Abilene, Texas built by Crusoe. Microsoft and other hyperscalers committed approximately $325 billion of combined capital expenditure in 2025 alone, most of it for AI compute. The Saudi PIF launched the Humain AI company in 2025 and signed compute deals with Nvidia, AMD, and Qualcomm.
The period saw heavy consolidation, partial acquihires that left companies operating, and several novel licensing deals.
| Year | Event |
|---|---|
| 2023 | Google Brain and DeepMind merge into Google DeepMind. |
| 2024 | Microsoft hires Mustafa Suleyman and most of Inflection AI's staff, licenses technology, leaves Inflection operating as a B2B company. |
| 2024 | Amazon hires founders and licenses technology from Adept AI. |
| 2024 | Google signs licensing deal with Character.AI, hires founders Noam Shazeer and Daniel De Freitas. |
| 2024 | Canva acquires Leonardo.AI. Freepik acquires Magnific AI. SoftBank acquires Graphcore. Recursion and Exscientia merge. |
| 2025 | Cognition AI acquires Windsurf (formerly Codeium) after Google hires Windsurf's CEO. |
| 2025 | MongoDB acquires Voyage AI. Hugging Face acquires Pollen Robotics. ServiceNow acquires Moveworks. Salesforce acquires Informatica. HP acquires Humane and discontinues the AI Pin. |
| 2025 | CoreWeave lists on Nasdaq, market debut above $20 billion. |
| 2025 | Microsoft and OpenAI revise their partnership; OpenAI restructures into a for-profit public benefit corporation while preserving the OpenAI Foundation as majority owner. |
While the United States, particularly the San Francisco Bay Area, hosts most of the highest valued AI startups, the industry is global. London hosts Google DeepMind, Stability AI, Wayve, ElevenLabs, Synthesia, and Isomorphic Labs. Paris hosts Mistral AI, Hugging Face, and Poolside AI. Germany hosts Qdrant, Aleph Alpha, Black Forest Labs, and NEURA Robotics. Israel hosts AI21 Labs, D-ID, and Mentee Robotics. Toronto hosts Cohere, Tenstorrent, and Ideogram. Vancouver hosts Sanctuary AI, Foundation, and Higgsfield AI.
In China, Beijing hosts Zhipu AI, Moonshot AI, Baichuan, 01.AI, Cambricon, and Booster Robotics. Shanghai hosts MiniMax, StepFun, AgiBot, Fourier Intelligence, and Kepler Robotics. Hangzhou hosts DeepSeek, Alibaba, Hikvision, and Unitree. Shenzhen hosts Tencent, Huawei AI, DJI, LimX Dynamics, UBTECH, XPENG, and EngineAI. South Korea hosts Naver Cloud, Naver Labs, Samsung AI, Doosan Robotics, and Rebellions. Japan hosts Sakana AI, SoftBank Group, Tokyo Robotics, Honda, and Toyota. The United Arab Emirates hosts G42 and the Mubadala AI portfolio. Saudi Arabia backs the Humain AI company through Saudi PIF. India hosts Sarvam AI, Krutrim, and Writesonic.
DeepSeek's January 2026 model release sharpened debate over export controls and the relative compute efficiency of Chinese versus American labs. The U.S. Commerce Department continues to administer export controls on advanced GPUs to China under the Foreign Direct Product Rule, while Nvidia, AMD, and TSMC operate special compliance programs.
Most frontier AI labs and many infrastructure firms remain privately held. The publicly traded set has grown to include hardware leaders Nvidia (NVDA), AMD (AMD), Intel (INTC), TSMC (TSM), and Samsung Electronics; hyperscalers Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), Oracle (ORCL); platform vendors Snowflake (SNOW), Salesforce (CRM), ServiceNow (NOW), Palantir (PLTR); and pure-play AI cloud providers CoreWeave (CRWV) and Nebius (NBIS). Smaller listed names include SoundHound (SOUN), C3.ai (AI), and BigBear.ai (BBAI).
Several cross-cutting trends shape the company landscape in 2026.
Compute concentration. Most frontier training runs in 2026 occur on clusters of more than 100,000 Nvidia Hopper or Blackwell GPUs, plus emerging clusters built around AMD Instinct or Google TPUs. The capital requirement limits frontier training to a small set of firms.
Open weights versus closed weights. Meta, Mistral AI, DeepSeek, Alibaba (Qwen), and Zhipu AI release open weight models. OpenAI released gpt-oss in August 2025, its first open weight LLM since GPT-2.
Agentic systems. Coding agents from Anysphere, Cognition AI, Windsurf, Replit, and Factory lead the developer agent segment by revenue. Customer service agents from Sierra AI, Decagon, and Ada Support compete for enterprise contact center workflows.
Robotics foundation models. Physical Intelligence, Skild AI, Covariant, Field AI, and Generative Bionics train general purpose robot foundation models on diverse manipulation and navigation data.
On-device AI. Apple, Qualcomm, Samsung, and Huawei push on-device inference using NPUs. Apple Intelligence launched on iPhone, iPad, and Mac through 2024 and 2025 with Private Cloud Compute as a fallback.
Sovereign AI. Governments invest in domestic compute and labs. France backs Mistral AI. Germany backs Aleph Alpha and Black Forest Labs. The UAE funds G42 and Falcon. Saudi Arabia backs Humain. Japan supports Sakana AI. South Korea funds Naver, LG AI Research, and Rebellions.
Energy and siting. AI data centers consume substantial electricity. Operators have signed long-term power purchase agreements covering nuclear (Microsoft with Constellation at Three Mile Island), natural gas, geothermal, and renewables.