100 companies, 847 moves

The Dublin-incorporated professional-services firm — the world's largest by revenue and headcount (~$74.2bn in fiscal 2026 and ~814,000 people, company) — that decided in 2023 to make generative AI the centre of its pitch, its training budget and its internal operations. Accenture sells no frontier model and runs no data centre; it earns money by helping large companies and governments choose, int

The San Francisco lab founded in 2022 by former OpenAI and Google Brain research leader David Luan, with transformer co-authors Ashish Vaswani and Niki Parmar, to build AI that takes actions in software rather than only producing text — two years before "computer-use agents" became a mainstream product. Adept raised ~$415m (reported), showed the ACT-1 agent demo (September 2022) and released the P

The San Jose creative and marketing software company (Photoshop, Illustrator, Premiere, Acrobat, Experience Cloud) that met generative AI as its biggest opportunity and its biggest threat at once. Adobe's answer, from March 2023, was Firefly: image (later video, audio and vector) models trained on Adobe Stock, openly licensed and public-domain content and sold as "commercially safe", with intellec

The Tel Aviv lab founded in 2017 by the Stanford AI professor Yoav Shoham, the entrepreneur Ori Goshen and the Mobileye founder Amnon Shashua. It was one of the first companies outside the US big labs to train and sell a GPT-3-scale language model (Jurassic-1, August 2021), and it ran a popular consumer writing assistant (Wordtune, October 2020). It became a unicorn with a $208m Series C at a $1.4

The Heidelberg model lab, founded in 2019 by former Apple AI researcher Jonas Andrulis, that was cast as Germany's answer to OpenAI, announced Europe's largest AI round of 2023 ("more than half a billion US dollars", November 2023), and then retreated from the frontier-model race to sell PhariaAI, a sovereign AI software platform for German public bodies and regulated industries. The retreat did n

The Hangzhou e-commerce group whose cloud arm, Alibaba Cloud (Aliyun), founded in 2009, became China's largest cloud provider, and whose Qwen (Tongyi Qianwen) model family became the most widely used open-weight model family in the world by 2025–26. Alibaba's AI strategy is a cloud strategy: give away strong models under permissive licences so that developers and companies build on Qwen, then sell

The search-and-advertising giant whose researchers invented much of modern AI (the transformer, large-scale deep learning on custom chips, AlphaGo and AlphaFold through Google DeepMind), which was caught flat-footed by ChatGPT in late 2022, stumbled publicly through 2023–24 (the Bard demo error, the Gemini image-generation pause, glue-on-pizza AI Overviews), and then recovered to become, by 2026,

The Seattle retailer whose cloud arm, Amazon Web Services, invented rentable computing in 2006 and became the largest cloud in the world, then entered the generative-AI era looking late (no frontier model of its own, a voice assistant that had stalled, and a rival in Microsoft that had locked up OpenAI) and answered with a different strategy: be the neutral marketplace for other people's models (B

The Santa Clara chip designer that spent forty years as Intel's second source and nearly went bankrupt in 2015, then under Lisa Su bet the company on one new CPU core (Zen) and one outside foundry (TSMC), took a large share of the server processor market from Intel, bought Xilinx and ZT Systems, and became the only merchant GPU maker with a credible challenge to Nvidia in AI data centres. AMD's re

The San Francisco model lab founded in 2021 by Dario and Daniela Amodei and a group of former OpenAI researchers who wanted to build frontier AI with safety research at the centre of the company. It makes the Claude family of models and sells them mainly to businesses and developers, through its own API, through Amazon Bedrock, Google Vertex AI and Microsoft's platforms, and through subscription p

The San Francisco start-up, founded in 2022 by four MIT students, whose AI code editor Cursor, a fork of Microsoft's open-source VS Code with AI built into every part of the editing loop, became the fastest-growing software product of the generative-AI era by its own and investors' accounts. Annualised revenue went from ~$100m (January 2025) to ~$1bn (November 2025) and ~$2bn (February 2026), then

The world's most valuable consumer-hardware company, owner of the iPhone and of the ~2.35bn-device installed base (company, January 2025) through which most people in rich countries reach AI, which bought Siri in 2010, let its assistant stall for a decade, and then spent 2024–26 visibly behind in generative AI. Apple announced "Apple Intelligence" in June 2024 with a privacy-first design (small mo

The Santa Clara networking company, founded in 2004 by Sun Microsystems co-founder Andy Bechtolsheim, Stanford professor David Cheriton and engineer Ken Duda, and run since 2008 by former Cisco executive Jayshree Ullal, that took the high-speed data-centre switching market from Cisco by selling fast, software-defined Ethernet switches to the largest cloud companies — and then became the main indep

The Cambridge, England chip-architecture company whose instruction set runs in nearly every smartphone in the world and, increasingly, in the CPUs at the heart of AI data centres — Nvidia's Grace and Vera, Amazon's Graviton, Google's Axion, Microsoft's Cobalt. For 35 years Arm made money by licensing designs and collecting a royalty of cents to a few dollars on each chip its customers shipped, nev

The Dutch maker of the lithography machines that print the patterns on computer chips, and the only company in the world that can build extreme-ultraviolet (EUV) scanners, without which no leading-edge AI chip can be made at reasonable cost. Founded in 1984 as a joint venture between Philips and ASM International, ASML spent two decades and billions of euros, part of it paid by its own customers,

The Beijing search company founded by Robin Li (Li Yanhong) in 2000 that became China's dominant search engine and then, from 2013, its earliest large-scale corporate bet on AI: a deep-learning institute, Andrew Ng as chief scientist, an open autonomous-driving platform (Apollo, 2017), its own AI chips (Kunlun, 2018) and the first major Chinese ChatGPT rival (Ernie Bot, March 2023). Baidu was firs

The semiconductor and infrastructure-software conglomerate that Hock Tan built by acquisition from Avago Technologies, and which became the second great winner of the AI chip boom after Nvidia. It does not sell a general-purpose AI chip. Instead it co-designs custom accelerators ("XPUs") with the largest buyers — Google's TPUs above all, plus Meta, ByteDance (reported), OpenAI and, through Google'

The London start-up, founded by Sachin Dev Duggal in 2016 as Engineer.ai, that sold custom app development to businesses as if it were an AI product. Customers described what they wanted to an AI assistant called "Natasha", and Builder.ai said its platform assembled most of the software from reusable components, with human developers doing the rest. It raised ~$450m or more (reported) from investo

The Redwood City enterprise-AI software company founded in 2009 by Tom Siebel, the CRM pioneer who sold Siebel Systems to Oracle, that tried to be the platform on which large industrial and government organisations build AI applications — predictive maintenance, supply-chain and fraud models, and later generative-AI search and agents. C3.ai took the ticker AI at its December 2020 IPO, and its shar

The Sydney-headquartered online design platform, founded in 2012 by Melanie Perkins, Cliff Obrecht and Cameron Adams, that made graphic design a free, browser-based activity for non-designers and then turned generative AI from a threat into its main growth story. Canva bundled AI features under Magic Studio (October 2023) with an indemnity (Canva Shield) and a $200m creator compensation fund, boug

The Silicon Valley chipmaker that refused to cut the wafer: instead of dicing a 300mm silicon wafer into hundreds of small chips, Cerebras makes the whole wafer one processor, the Wafer-Scale Engine (WSE), with ~900,000 cores and 44GB of on-chip memory in its third generation (company, March 2024). For most of its life it was a niche supplier of supercomputers to national labs and drug companies,

The companion-chat app founded in 2021 by two Google researchers, Noam Shazeer (a co-author of the transformer paper) and Daniel De Freitas, after Google declined to launch their chatbot. It let anyone create and talk to AI "characters" and by 2024 was one of the most-used AI apps by time spent, especially among teenagers and young adults (~20m monthly users, company, 2024). It never found a busin

The New York facial-recognition company, founded in 2017 by Hoan Ton-That and Richard Schwartz, that built "a search engine for faces" (Seventh Circuit, July 2026) by scraping billions of photographs from social media and the open web without consent, and sold searches mainly to police and government agencies. A single upload of a face returns other photographs of the same person and links to wher

The San Francisco company, founded in 2023 by competitive programmers Scott Wu, Steven Hao and Walden Yan, that coined the "AI software engineer" with Devin, an autonomous coding agent unveiled on 12 March 2024 in a demo that made it famous and drew accusations of exaggeration. Cognition does not train frontier models; it builds the agent harness, its own smaller SWE models and the product around

The Toronto model lab, co-founded in 2019 by transformer co-author Aidan Gomez, that chose from the start to sell large language models only to businesses and governments rather than to build a consumer chatbot. Cohere competes with OpenAI, Anthropic and Mistral AI not on frontier benchmarks but on private deployment (its models run in a customer's own cloud or data centre), retrieval and search c

The New Jersey GPU cloud that began in 2017 as Atlantic Crypto, an Ethereum-mining operation run by three former commodity traders, turned its mining GPUs into a rental business in 2019, and became the largest and most closely watched "neocloud" of the generative-AI boom. It rents large clusters of the newest Nvidia accelerators, mostly on multi-year take-or-pay contracts, to a handful of very lar

The San Francisco self-driving start-up, founded in 2013 by Kyle Vogt and Dan Kan, that General Motors bought in 2016 for a reported ~$1bn. It became the best-funded challenger to Waymo and then the clearest failure of the robotaxi era. Backed by GM, SoftBank's Vision Fund, Honda, Microsoft, Walmart and T. Rowe Price, Cruise reached a ~$30bn valuation (January 2021). It started paid driverless rid

The Denver company founded in 2018 to burn natural gas that oil producers would otherwise flare, using it to power bitcoin-mining containers on the well pad, which turned that habit of going where the power is into an AI-infrastructure business: first a GPU cloud (Crusoe Cloud, from 2022), then the developer of the first Stargate campus in Abilene, Texas, built for Oracle and OpenAI (~1.2 GW plann

The San Francisco data and AI platform founded in 2013 by the University of California, Berkeley researchers who created Apache Spark. It sells the "lakehouse", a design that keeps enterprise data in cheap cloud object storage in open formats (Delta Lake, and since its 2024 purchase of Tabular, Apache Iceberg) and runs data engineering, SQL warehousing, machine learning and now AI agents on top of

The San Francisco customer-support agent company founded in August 2023 by Jesse Zhang (CEO) and Ashwin Sreenivas (CTO), two repeat founders who had each sold a previous start-up, which builds AI "concierge" agents that resolve customer inquiries across chat, email, SMS and voice for consumer and technology brands. Decagon is Sierra's closest AI-native rival but took the opposite side of the prici

The Hangzhou model lab that Liang Wenfeng spun out of his quantitative hedge fund High-Flyer in July 2023. It was funded for almost three years by the fund's profits and its stock of Nvidia GPUs, many bought before US export controls. It became the most famous Chinese AI company in the world in January 2025, when its open-weight reasoning model R1 and a claim that its V3 base model's final trainin

The Montreal AI company, co-founded in October 2016 by serial entrepreneur Jean-François Gagné and deep-learning pioneer Yoshua Bengio with Real Ventures, that was meant to be Canada's answer to the US AI giants: a "research-to-product" firm that would build AI for large enterprises. It raised two of the largest venture rounds in Canadian history — ~US$102m (C$137.5m) in its June 2017 Series A and

The voice-AI company founded in 2022 by two Polish childhood friends, Mati Staniszewski and Piotr Dąbkowski, that turned frustration with flat, single-voice film dubbing in Poland into the most widely used independent text-to-speech, voice-cloning and dubbing platform, and then into a platform for voice agents. ElevenLabs competes with OpenAI, Google and Microsoft, whose general models also speak,

The Redwood City, California data-centre company, founded in 1998 as a neutral place where networks could exchange internet traffic, that grew into the world's largest operator of interconnection-focused colocation: ~268 data centres in ~74 metros (end-2024, company) where carriers, clouds and enterprises connect directly to each other through more than 480,000 cross-connects. It converted to a re

The Abu Dhabi technology group, chaired by the UAE's national security adviser Sheikh Tahnoon bin Zayed Al Nahyan, that became the operating arm of the Emirates' bid to be the world's third AI centre after the US and China. G42 began in 2018 as a little-known analytics firm, grew through pandemic testing and health contracts, and then turned itself into a holding group for cloud (Core42), data cen

The Palo Alto enterprise-search company founded in 2019 by former Google search engineer and Rubrik co-founder Arvind Jain, which indexes a company's documents, messages and tickets across dozens of software suites and puts an AI assistant and, from 2025, AI agents on top. Glean's bet is neutrality: a large company that uses Microsoft, Google, Salesforce, Atlassian, Slack and ServiceNow at once ne

The London research lab founded in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman to "solve intelligence", bought by Google in January 2014 for a price reported at ~$400–650m, which became famous for science rather than products (AlphaGo beating Lee Sedol in March 2016; AlphaFold predicting protein structures in 2020 and winning Hassabis and John Jumper half of the 2024 Nobel Prize in Che

The inference-chip start-up founded in 2016 by Jonathan Ross, who had started Google's Tensor Processing Unit as a side project, which built a deterministic, SRAM-based processor it later named the Language Processing Unit (LPU), struggled for years to sell it as hardware, went viral in February 2024 when its public chatbot generated text at ~500 tokens per second, and rebuilt itself as GroqCloud,

The San Francisco legal-AI company, founded in 2022 by a litigation associate (Winston Weinberg) and an AI research scientist (Gabe Pereyra), that was seeded by the OpenAI Startup Fund, launched with a firm-wide deployment at Allen & Overy before most lawyers had heard of GPT-4, and became the best-funded application company in professional services. Harvey sells a platform of assistants, document

The New York finance-research AI company, founded in 2020 by Stanford PhD dropout George Sivulka, whose Matrix interface — a spreadsheet in which every row is a document, every column a question and every cell a model-generated answer linked to its source — became one of the best-known AI tools on Wall Street. Hebbia sells to asset managers, private-equity and credit funds, investment banks and la

The New York- and Paris-based company that began in 2016 as a chatbot app for teenagers, open-sourced a small library for running Google's BERT model in 2018, and turned it into the default home of open machine learning: the transformers library, the Hugging Face Hub (more than 3 million public models, 500,000 datasets and 1 million demo applications by September 2026, per Nvidia's announcement) a

Saudi Arabia's national AI company, launched by the Public Investment Fund (PIF) on 12 May 2025, the eve of President Trump's visit to Riyadh, and chaired by Crown Prince Mohammed bin Salman. HUMAIN (styled in capitals by the company; folder Humain) was given a full-stack mandate: data centres, cloud, Arabic models (ALLaM), consumer and enterprise apps (HUMAIN Chat, HUMAIN One) and investment. It

The San Francisco hardware start-up founded in 2018 by former Apple designers Imran Chaudhri and Bethany Bongiorno that spent ~five years and ~$230m (reported) building the Ai Pin, a screenless, clip-on AI assistant with a camera and a laser that projected text onto the user's palm, launched it in November 2023 at $699 plus $24 a month, and sold only ~10,000 units by August 2024 (reported) against

The 115-year-old Armonk computing company whose *Jeopardy!*-winning Watson system made it the most famous name in commercial AI in the 2010s. It then became the industry's best-known cautionary tale when Watson Health was sold off in 2022. Since 2020 IBM has rebuilt its AI strategy around three ideas: hybrid cloud as the platform, centred on the ~$34bn Red Hat acquisition (2019); small, open, gove

The Palo Alto lab founded in March 2022 by Mustafa Suleyman (a co-founder of DeepMind), Reid Hoffman (LinkedIn co-founder and Microsoft board member) and Karén Simonyan (a leading DeepMind researcher) to build "personal AI" — a kind, conversational assistant called Pi. It raised ~$1.5bn in under fifteen months, including $1.3bn at ~$4bn in June 2023 from Microsoft, Nvidia, Hoffman, Bill Gates and

The Bengaluru IT-services company, founded in 1981 by N. R. Narayana Murthy and six colleagues, that helped build India's offshore outsourcing industry and is now one of the clearest test cases of what generative AI does to labour-priced services. Infosys earns ~$20bn a year (FY2026, company) mainly by maintaining, modernising and running software and business processes for large Western companies

The AI company that Karim Beguir and Zohra Slim started in Tunis in 2014 with two laptops and about $2,000 (as the founders tell it), which moved its headquarters to London in 2015, sold "decision-making AI" (optimisation and reinforcement learning) to railways, logistics firms and chip designers, and became Africa's best-known AI exit when BioNTech bought it in 2023 for up to ~£562m (~£362m up fr

The company that put "Silicon" in Silicon Valley and dominated PCs and servers for four decades, then missed the smartphone, lost manufacturing leadership to TSMC, failed repeatedly in AI accelerators while Nvidia took the market, and bet the firm on becoming a foundry for others. After a record ~$18.8bn net loss in 2024 and the exit of chief executive Pat Gelsinger, Intel was partly rescued by it

The Austin, Texas start-up that turned OpenAI's GPT-3 into the first breakout AI writing business — launched in early 2021 as Conversion.ai, renamed Jarvis and then Jasper — and became the AI-application era's defining cautionary tale. It reached a reported ~$45m of revenue in 2021 and raised a $125m Series A at a ~$1.5bn valuation in October 2022 (Insight Partners lead; reported), six weeks befor

The San Jose GPU cloud founded in 2012 by twin brothers Stephen and Michael Balaban, which began as a tiny AI-software start-up, became a seller of deep-learning workstations and servers after an Amazon Web Services bill nearly sank it, and turned into one of the largest private "neoclouds" renting Nvidia GPUs to AI researchers, start-ups and, from 2025, to Microsoft, Nvidia itself and Anthropic.

The Menlo Park owner of Facebook, Instagram, WhatsApp and Messenger (~3.5bn daily users across its apps, company), which funds AI from the most profitable advertising machine outside Google, built one of the field's great research labs (FAIR, 2013) and its most-used framework (PyTorch), and between 2023 and 2025 made open-weight models (Llama) the industry's default alternative to closed labs. Aft

The Boise, Idaho memory maker — the only large US producer of DRAM and the smallest of the three companies that make nearly all of the world's DRAM — whose AI strategy is a bet on high-bandwidth memory (HBM), the stacked DRAM that sits beside every Nvidia, AMD and custom AI accelerator. Micron skipped HBM3, came from behind with HBM3E for Nvidia's H200 in February 2024, and by Q2 2026 held ~18% of

The Redmond software company that made the earliest and largest bet of any incumbent on generative AI, investing ~$13bn in OpenAI from 2019 to 2023 in return for exclusive cloud hosting, rights to OpenAI's models and a share of its profits, and then putting those models into Bing, GitHub, Office and Windows under the "Copilot" brand. The bet turned Azure into the fastest-growing large cloud (+48%

The San Francisco image-generation lab founded by Leap Motion co-founder David Holz that launched inside a Discord server in 2022, charged subscriptions from the start and never took venture capital. Midjourney became the best-known paid image generator on the strength of a distinctive aesthetic, reported revenue above ~$200m in 2023 with a team of a few dozen people (Holz statement reported by Th

The Paris model lab founded in April 2023 by three French researchers from Google DeepMind and Meta, which made its name by releasing small, efficient open-weight models under the permissive Apache 2.0 licence (Mistral 7B, September 2023; Mixtral, December 2023) and then built a business selling commercial models, an assistant (Le Chat), on-premises deployments and, from 2025, its own compute to E

The Jerusalem company, founded in 1999 by Hebrew University professor Amnon Shashua and businessman Ziv Aviram, that bet a single camera and a purpose-built vision chip (EyeQ) could do most of what cars needed to see, and turned that bet into the world's dominant supplier of advanced driver-assistance systems (ADAS). Mobileye sells chips and software to carmakers through their Tier 1 suppliers, no

The Beijing model lab behind the Kimi assistant, founded in March 2023 by the Tsinghua- and Carnegie Mellon-trained researcher Yang Zhilin and named after Pink Floyd's *The Dark Side of the Moon* (reported). Moonshot first made its name in China with very long context windows (200,000 Chinese characters in October 2023, 2 million announced in March 2024) and an expensive consumer advertising campa

The Amsterdam-headquartered AI cloud that is what remained of Yandex N.V., Russia's largest internet company, after it sold all of its Russian businesses in July 2024 for 475bn roubles (~$5.4bn) and kept a Finnish data centre, a few thousand engineers outside Russia, cash and a handful of side businesses. Led by Yandex's co-founder Arkady Volozh, it relaunched on Nasdaq as NBIS in October 2024 and

The Santa Clara chip designer that began in 1993 making 3D graphics cards for PC games, bet from 2006 that its graphics processors could become general-purpose parallel computers (CUDA), was in the right place when deep learning took off in 2012, and became the main supplier of the computing power behind the generative-AI boom. It sells GPUs, but increasingly whole systems: chips, networking, rack

The San Francisco lab founded as a nonprofit in December 2015 to build artificial general intelligence "for the benefit of humanity". It became a capped-profit company funded by Microsoft in 2019, launched ChatGPT in November 2022 and so started the generative-AI boom, and in October 2025 became a public benefit corporation controlled by a renamed nonprofit foundation. It has the largest consumer

The database company Larry Ellison co-founded in 1977, which arrived late and small to cloud computing, rebuilt its infrastructure cloud (OCI) from scratch in 2016–18, and then turned it into the fastest-growing AI compute supplier of 2024–26 by renting giant GPU clusters to model labs, above all OpenAI. A reported ~$300bn, roughly five-year OpenAI contract (September 2025) lifted Oracle's remaini

The Denver-based data-integration and AI software company, co-founded in 2003 by Peter Thiel and Alex Karp with early backing from the CIA's venture arm In-Q-Tel, that spent its first decade and a half as a secretive, loss-making contractor to intelligence agencies and militaries and then became the clearest public-market winner of the enterprise-AI boom. Palantir does not train frontier models. I

The San Francisco "answer engine", founded in August 2022 by former OpenAI researcher Aravind Srinivas with Denis Yarats, Johnny Ho and Andy Konwinski, that answers questions with a short, cited summary of live web results instead of a list of links. Perplexity does not train frontier models; it combines its own search index and Sonar models (post-trained on open-weight bases) with models rented f

The New York–headquartered, Israeli-founded company that created the "vector database" as a managed cloud product, founded in 2019 by former Amazon AI Labs head Edo Liberty. Vector databases store the numerical representations (embeddings) that language models use to find relevant documents, and in 2023 Pinecone became the best-known piece of the retrieval-augmented generation (RAG) stack: it rais

The coding-model lab founded in 2023 by former GitHub CTO Jason Warner and software entrepreneur Eiso Kant on the thesis that the shortest route to more general AI runs through software: train models on code, then improve them with reinforcement learning from code execution (letting a model write code, run it, and learn from whether it works). Poolside chose to build its own foundation models rath

The San Diego chip and patent company that built the wireless standards inside almost every smartphone and is now trying to turn that position into an AI business twice over: first by making on-device AI (models that run on the phone, PC or car rather than in a data centre) a selling point for its Snapdragon chips, and second by re-entering the data-centre market it had twice abandoned, with the A

The small Santa Monica start-up led by Chinese-born entrepreneur Jesse Lyu that became the surprise hit of CES in January 2024 with the R1, a $199 orange handheld designed with Teenage Engineering that promised a "large action model" (LAM) able to use apps on the owner's behalf, took ~100,000 orders in about three months (company claim; ~130,000 units sold by June 2024, Forbes) — and then became,

The San Francisco online-community platform (NYSE: RDDT), founded in 2005 by Steve Huffman and Alexis Ohanian, that discovered in the generative-AI era that its two decades of human conversation were among the most valuable training and grounding data on the web. Reddit turned that corpus into a business line by charging for its API (2023), signing data-licensing deals with Google (~$60m a year, F

A small multimodal model lab, founded in 2022 by researchers from Google DeepMind, Google Brain and Meta, that trains its own models to understand text, images, video and audio and sells them, and agent products built on them, to enterprises. Reka is the counter-example to the multi-billion-dollar "reputation round": it emerged from stealth with ~$58m (June 2023), turned down or lost a reported >$

The browser-based coding platform, founded in 2016 by Amjad Masad, Haya Odeh and Faris Masad, that spent eight years as a free learning tool for students and hobbyists with ~$2.8m of annual revenue, then cut ~20% of staff (May 2024), turned away from professional developers, and launched Replit Agent (September 2024) — an AI that builds, runs and deploys whole applications from a plain-English des

The New York generative-video company, founded in 2018 by three NYU art-school graduates, that went from packaging research models for artists to co-authoring the latent-diffusion work behind Stable Diffusion, then to selling the first widely available text-to-video product (Gen-2, 2023) and, from December 2025, to calling itself "the world model company". Runway competes with OpenAI's Sora, Googl

The research lab Ilya Sutskever founded in June 2024, a month after leaving OpenAI, with Daniel Gross and Daniel Levy, to build "safe superintelligence" in "a straight shot" with "one product" and no commercial distractions (company, 19 June 2024). Without releasing a product, a model or a research paper describing its approach, it raised $1bn at a reported ~$5bn valuation (September 2024) and ~$2

The Tokyo research lab founded in 2023 by David Ha and transformer co-author Llion Jones, with former diplomat and Mercari executive Ren Ito, that bet it could win without the frontier labs' budgets by borrowing ideas from nature (evolution, collective intelligence) and by building AI for Japanese language, culture and industry. Its name means "fish" in Japanese, a reference to schools of small fi

The San Francisco company that invented cloud-delivered business software (customer relationship management, or CRM, sold as a subscription from 1999) and, after a decade of bolting predictive AI onto it under the Einstein brand, bet the company from 2024 on Agentforce — AI agents that work inside Salesforce's own data and workflows. It is the clearest case in this knowledge base of an application

The Bengaluru start-up, founded in July 2023 by Vivek Raghavan and Pratyush Kumar, that set out to build a "full-stack" generative AI platform for India's languages and became the government's first pick to build a sovereign Indian foundation model under the IndiaAI Mission (April 2025). Sarvam's path shows the sovereign-model dilemma in miniature. Its first widely promoted large model, Sarvam-M (

The San Francisco data company, founded in 2016 by Alexandr Wang and Lucy Guo, that built the largest business in AI's least glamorous layer: turning human judgement into training data. Scale labelled camera and lidar frames for self-driving programmes, then became the main outside supplier of expert-written examples, preference ratings and evaluations for frontier labs' post-training (section 09)

The Santa Clara workflow-software company, founded in 2004 by Fred Luddy as an IT help-desk ("IT service management") platform, that became the system of record for how large organisations route requests, incidents and approvals — and then tried to turn that position into the control layer for enterprise AI agents. ServiceNow sells its generative-AI features under the Now Assist brand as premium "

The New York stock-media marketplace, founded in 2003 by Jon Oringer, that pays contributors royalties to license photos, video and music to businesses — and that, from 2022, became the AI industry's best-known seller of licensed training data. It partnered with OpenAI on DALL-E image generation (October 2022), signed a six-year data deal with OpenAI (July 2023) and licensed content to Meta, Alpha

The San Francisco customer-experience agent company founded in February 2023 by Bret Taylor (former Salesforce co-CEO, chair of OpenAI's board since November 2023) and Clay Bavor (18 years at Google), which sells large consumer brands branded AI agents that answer customers by chat, email and voice and take actions such as returns, card replacements and subscription changes. Sierra is the best-kno

The South Korean memory maker that was a creditor-owned near-bankrupt in 2001, the perennial number two to Samsung for two decades, and then, by betting early on stacked high-bandwidth memory (HBM) and becoming the lead supplier to Nvidia, the most profitable memory company in history. SK Hynix posted a 2023 operating loss of ~7.73tn won, its largest ever (reported), then 2025 operating profit of

The cloud data warehouse founded in 2012 by two Oracle architects and a Dutch database researcher, which separated storage from compute, charged by consumption, and in September 2020 completed the largest software IPO to that date (~$3.4bn raised at $120 a share; shares closed ~112% higher on the first day). It became the archetype of usage-based software: product revenue grew from ~$554m (fiscal

The London-founded generative-AI company that funded and released Stable Diffusion in August 2022, the open-weights image model that put text-to-image generation on millions of ordinary computers and became the first mass-market proof that open models could rival closed ones. It raised $101m at a $1bn valuation two months later (October 2022), spent far faster than it earned (2023 revenue reported

The Cambridge, Massachusetts, music-generation company founded in 2022 by four alumni of the AI start-up Kensho, whose models turn a text prompt into a complete song with vocals, lyrics and instruments. Suno made AI music a mass consumer product — more than 100m people had used it and ~2m paid for it by June 2026 (reported) — and raised at ~$500m (May 2024), $2.45bn (November 2025) and ~$5.4bn (Ju

The San Jose server maker, founded in 1993 by Charles Liang, that builds computers from interchangeable "building blocks" and, by being first to ship each new Nvidia GPU generation in volume, became the best-known listed pure play on AI servers. Its revenue rose from ~$7.1bn in fiscal 2023 to ~$15bn in fiscal 2024, ~$22.0bn in fiscal 2025 and $39.1bn in fiscal 2026 (year to June 2026, company), an

The London company, founded in 2017 by Victor Riparbelli, Steffen Tjerrild and computer-vision professors Matthias Niessner and Lourdes Agapito, that sells AI-generated presenter videos — a realistic digital "avatar" reading a script in any of well over a hundred languages — to corporate training, internal-communications and marketing teams. Synthesia chose early to be an enterprise product with s

The Austin-based electric-car and energy company that, under Elon Musk, has staked most of its valuation on artificial intelligence: a camera-only driver-assistance system sold to consumers (Autopilot from 2014, "Full Self-Driving" from 2016), a fleet of millions of cars that supplies driving video for training, its own in-car AI chips, a training supercomputer (Dojo, shut down in August 2025 and

The San Francisco model lab founded by former OpenAI chief technology officer Mira Murati, which in July 2025 raised a ~$2bn seed round at a ~$12bn valuation — the largest seed on record at the time (reported) — with no product, then shipped a fine-tuning API (Tinker, October 2025) and a ~975bn-parameter open-weight model (Inkling, July 2026, reported). It is the clearest case of the 2024–26 "repu

The San Francisco "AI-native cloud" founded in June 2022 by Vipul Ved Prakash (founder of Topsy, which Apple bought in 2013), Ce Zhang (then of ETH Zurich), Chris Ré and Percy Liang (both of Stanford), with FlashAttention's author Tri Dao as chief scientist from July 2023. It began with a plan for a decentralised cloud of volunteered GPUs and open research (RedPajama, 2023), then became one of the

The Taiwanese contract chipmaker founded by Morris Chang in 1987 on an idea the industry first dismissed: a "pure-play" foundry that makes chips for others and never sells chips of its own. Almost four decades later it makes almost every leading-edge AI accelerator in the world, for Nvidia, AMD, Broadcom's custom-chip clients and the hyperscalers, as well as Apple's iPhone and Mac processors. It a

The robotic process automation (RPA) company, founded in Bucharest in 2005 by Daniel Dines and Marius Tîrcă, that built software "robots" to click through legacy screens on behalf of office workers, rode a 2018–21 venture boom to a ~$35bn private valuation, listed on the New York Stock Exchange in April 2021 and then lost most of that value as growth slowed. Since 2024 it has tried to turn its ins

The San Francisco cloud platform, founded as ZEIT in 2015 by Argentine-born developer Guillermo Rauch, that turned an open-source React framework, Next.js, into the main sales channel for a hosting business, then re-cast itself as the "AI Cloud": the place where AI apps — and, increasingly, software written by AI agents — are built and deployed. Its AI products are the open-source AI SDK (June 202

The Ohio-based maker of the power and cooling equipment that sits between the electricity grid and the AI chip — uninterruptible power supplies, switchgear, busways, precision air conditioning, chillers, and since 2023 the coolant distribution units and liquid-cooling systems that frontier GPU racks need. Vertiv is the former Emerson Network Power, sold to private equity in 2016, listed through a

Alphabet's self-driving company. It began in January 2009 as Google's self-driving car project and was spun out under its own name in December 2016. It is the only Western company that runs fully driverless, paid robotaxi service at scale. Waymo spent most of its first decade as a symbol of the self-driving overpromise: its timelines slipped, its valuation was marked down from a sell-side ~$175bn

The London self-driving company, founded in Cambridge in 2017 by Alex Kendall and Amar Shah, that bet on "end-to-end" learning — a single neural network trained on driving data that goes from camera pixels to steering and speed, instead of the hand-built modules and detailed maps of the first robotaxi generation. Wayve does not build cars or run a fleet. It sells software, the "Wayve AI Driver", t

The San Francisco developer-tools company, founded in 2017 by Lukas Biewald, Chris Van Pelt and Shawn Lewis, whose experiment-tracking software became the default "lab notebook" for deep-learning research, including at OpenAI and most other frontier labs (company claims). Weights & Biases (W&B) sold to the people who train models rather than to the people who use them: a free tier for researchers

The AI coding company that began in 2021 as Exafunction, a GPU-virtualisation start-up, pivoted in mid-2022 to Codeium, a free AI autocomplete tool with a self-hosted enterprise edition, and in November 2024 launched the Windsurf editor, an AI-native fork of VS Code built around the Cascade agent. Renamed Windsurf in April 2025, it claimed ~$100m of annual recurring revenue (ARR) that month (repor

The San Francisco enterprise generative-AI company (styled WRITER from 2025) founded by May Habib and Waseem Alshikh, who turned their earlier localisation start-up Qordoba into a writing assistant for corporate brand and style in 2020 and then, unusually for an application company, trained its own family of large language models (Palmyra) to sell a "full-stack" platform — models, retrieval, guard

Elon Musk's AI company, incorporated in March 2023 and announced in July 2023 as a "maximally truth-seeking" alternative to an OpenAI he had co-founded and come to oppose, which reached the frontier faster than any lab before it by spending on compute at extraordinary speed: the Colossus cluster in Memphis was built in 122 days (company claim, September 2024) and powered at first by unpermitted ga

The Beijing model lab spun out of Tsinghua University's Knowledge Engineering Group (KEG) in 2019, whose GLM ("General Language Model") family made it one of China's earliest builders of large language models and, from January 2026, the first independent large-model developer anywhere to list its shares (Hong Kong, stock code 2513). For most of its life Zhipu was the Chinese lab most tied to the s

The algorithmic home-buying ("iBuying") programme that Zillow Group, the most visited US property website, ran from April 2018 to November 2021. Zillow used pricing models descended from its Zestimate home-value estimate to make cash offers on houses, did light repairs and resold them within months, aiming to earn a fee and a spread and to move Zillow from selling advertising to agents to owning t