Top 50 AI Companies: 2026 Edition

NVIDIA designs graphics processing units (GPUs), networking and systems software, and sells them as complete data-center platforms. Revenue for fiscal 2026, which ended on 25 January 2026, was US$215.9 billion, of which the Data Center segment contributed US$193.7 billion. The company describes its Blackwell generation as "data-center-scale" products in which chips, networking, systems and software are co-designed so that hundreds of thousands of GPUs can operate as one computer. During fiscal 2026 it unveiled the Rubin platform, which it expects to begin production shipments in the second half of fiscal 2027 and which it says reduces cost per token by up to ten times compared with Blackwell. NVIDIA had about 42,000 employees in 38 countries at the end of the fiscal year. Its chips are also a supply input for many other companies on this list. The principal limitation is concentration: the data-center business depends on a small number of very large cloud and AI buyers, and on export-control rules that restrict sales to some countries.

Anthropic, a public benefit corporation founded in 2021 and headquartered in San Francisco, develops the Claude family of large language models (LLMs), which are neural networks trained to generate and reason over text, code and images. Its S-1 registration statement reports revenue of US$41 million in 2023, US$386 million in 2024 and US$4.6 billion in 2025, then US$33.2 billion in the nine months to 30 September 2026; run-rate revenue (a recent period's revenue annualised) was about US$81 billion at 31 October 2026. Claude is sold through an API, through applications such as Claude Code and Claude Cowork, and through Amazon Web Services, Google Cloud and Microsoft Azure. The company trains and serves models on AWS Trainium, Google TPUs, NVIDIA GPUs and AMD GPUs. On 10 November 2026 it launched the roadshow for an initial public offering of 825 million Class A shares at US$70 to US$78 each, to trade on Nasdaq as ANT. The filing lists customer concentration in large accounts and very high compute commitments among its risks.

OpenAI develops the GPT series of models and the ChatGPT assistant, which it distributes to consumers through subscriptions, to businesses through enterprise plans and to developers through an API. In a January 2026 blog post, chief financial officer Sarah Friar stated that revenue more than tripled in 2025 to over US$20 billion, from US$6 billion in 2024, and attributed the growth mainly to compute capacity, which rose from 0.2 gigawatts (GW) in 2023 to 0.6 GW in 2024 and about 1.9 GW in 2025. Its compute arrangements appear in suppliers' filings: CoreWeave reports an order form under which OpenAI committed to pay up to about US$6.5 billion through May 2031, and Cerebras reports a multi-year agreement valued at more than US$20 billion covering 750 megawatts of inference capacity. OpenAI does not publish audited accounts, so its figures cannot be compared line by line with those of listed companies. Its main limitation is the scale of the compute obligations it has signed relative to its current revenue.

Databricks sells a cloud data platform that combines data engineering, analytics and model development, and positions it as the base on which companies build AI applications and agents. On 13 August 2026 it reported a revenue run-rate above US$7 billion and year-on-year growth above 80% in its second quarter, together with a US$5 billion strategic funding round at a US$190 billion valuation led by Coatue. The release states that more than 20,000 organisations use the platform and gives the headquarters as San Francisco. Databricks sits in the data and tooling layer: it does not train frontier models for sale, but it hosts customers' models and connects them to governed enterprise data. Revenue is consumption-based, rising with customers' compute usage, which makes it sensitive to customers' budgets for data and AI projects. The company remains private and publishes no audited accounts.

CoreWeave rents GPU capacity built for AI workloads, a model often called a "neocloud" to distinguish it from general-purpose hyperscale clouds. Revenue in 2025 was US$5.13 billion. Its 10-K states that about 67% of 2025 revenue came from a single customer, Microsoft, and that OpenAI signed an order form in September 2025 committing to pay up to about US$6.5 billion through 31 May 2031. The company has bought software businesses to move up the stack, naming Weights & Biases, OpenPipe, Marimo and Monolith AI among its acquisitions. It is headquartered in Livingston, New Jersey, and had 2,189 employees at the end of 2025. Customer concentration is the clearest risk the company itself discloses: the loss or reduction of one or two contracts would materially change its revenue.

Palantir sells four software platforms: Gotham, Foundry, Apollo and its Artificial Intelligence Platform (AIP), which it describes as its generative AI platform. Revenue in 2025 was US$4.48 billion, made up of US$2.40 billion from government customers and US$2.07 billion from commercial customers. Foundry supplies the data layer, the Palantir Ontology, which maps an organisation's data, logic and actions; AIP lets customers connect large language models to that layer so that models act on operational data rather than on documents alone. Palantir was founded in 2003 and began by building software for the US intelligence community. Its 10-K lists principal executive offices in Aventura, Florida, and 4,429 full-time employees at the end of 2025. The company depends heavily on public-sector contracts, which are subject to budget cycles and political scrutiny.

xAI develops the Grok models and assistant. SpaceX acquired X.AI Holdings Corp. effective 2 February 2026, and xAI had itself acquired X Holdings, owner of the X social network, effective 28 March 2025. SpaceX now reports an AI segment comprising AI compute, Grok and X; in 2025 that segment generated revenue of US$3.20 billion and an operating loss of US$6.36 billion, and in the first quarter of 2026 revenue of US$818 million. Capital expenditure attributed to the AI segment was US$12.7 billion in 2025. SpaceX reports about 550 million monthly active AI users across Grok and X as of March 2026. In 2023 and 2024 nearly all AI-segment revenue came from X advertising, subscriptions and data licensing, so the segment figure is not a measure of model revenue alone. SpaceX also acquired Cursor, the AI coding editor, in August 2026.

UiPath sells an automation platform that began with robotic process automation (software that repeats rule-based tasks in user interfaces) and is now marketed as agentic automation, in which AI agents plan and carry out multi-step work under human controls. Revenue for the fiscal year ended 31 January 2026 was US$1.61 billion. The company was established in Bucharest, Romania, in 2005, incorporated in Delaware in 2015, and has its principal executive offices in New York; it had 3,981 full-time employees at fiscal year-end. Its customers are mostly large enterprises and public bodies that already run UiPath robots, which gives it an installed base into which agent products can be sold. The limitation is competitive: the same agent capabilities are offered by model developers and by large software vendors that sell to the same buyers.

Tempus AI sells genomic and diagnostic tests to clinicians and licenses de-identified clinical and molecular data to pharmaceutical companies, using machine-learning models to match patients to therapies and trials. Revenue in 2025 was US$1.27 billion. The company was founded in 2015, renamed Tempus Labs in 2016 and Tempus AI in 2023, and reincorporated from Delaware to Nevada on 7 August 2025; its principal executive offices are in Chicago. Its revenue therefore comes mostly from laboratory testing and data licensing rather than from software subscriptions, which places it at the applied end of the AI industry. The 10-K highlights its limited operating history and rapid growth as factors that make its prospects difficult to evaluate.

Upstart operates what its annual report calls an AI lending marketplace: its models assess borrowers' credit risk and route loans to partner banks, credit unions and capital-markets buyers. Revenue in 2025 was US$1.04 billion. The company's principal executive offices are in San Mateo, California, and it had 1,405 full-time employees at the end of 2025. Upstart is included here because its product is the output of proprietary machine-learning models; it is a lending business, not a model developer. Its 10-K lists the ability to keep improving the AI models, to increase loan volume through the marketplace and to maintain diversified loan funding among the factors that drive results. Its revenue depends on credit conditions and on the willingness of funding partners to buy loans.

Cognition builds Devin, a software agent that plans and writes code, runs tests and opens pull requests with limited human input. On 25 September 2026 the company announced that it had passed US$1 billion in annualised revenue run-rate, less than two years after Devin became generally available, naming GE Aerospace and Rivian among customers. Its valuation rose quickly: US$10.2 billion in September 2025, US$26 billion in a round of over US$1 billion led by Lux Capital, General Catalyst and 8VC on 27 May 2026, and US$48 billion in a Series E of over US$2 billion led by Andreessen Horowitz and Accel on 8 September 2026. It signed an agreement to acquire the Windsurf coding editor in July 2025 and has since added teams including The Interaction Company and Corridor. Coding agents are the most contested application category, with model developers selling competing tools directly.

SenseTime began as a computer-vision company and now sells generative AI models, AI infrastructure and vision applications, mainly in mainland China. Revenue for 2025 rose 32.9% to RMB5.01 billion, its fastest growth in three years; the loss for the year narrowed 58.6% to RMB1.78 billion. At the European Central Bank cross rate on 31 December 2025 the revenue equals about US$716 million. Gross margin was 41.0% of revenue. The company reports that second-half 2025 EBITDA (earnings before interest, tax, depreciation and amortisation) turned positive, although full-year EBITDA was still negative. Like other Chinese developers, it operates under US export controls on advanced AI chips, which constrain the compute it can deploy.

Nebius Group, incorporated in the Netherlands with principal offices at Schiphol, operates a full-stack AI cloud: GPU clusters, storage and software for training and inference. Revenue in 2025 was US$530 million. The company released Nebius Token Factory, a managed inference service for deploying and fine-tuning open-source and custom models, in November 2025. Its 20-F states that new sites in the United States, France, the United Kingdom and Israel bring total contracted power to more than 2 GW, most of it in data centres the company owns and designs. Nebius had about 1,500 employees at the end of 2025, most of them engineers. The scale of planned capacity relative to current revenue means results depend on signing large, long-term customer contracts.

Cerebras designs wafer-scale processors, chips the size of a full silicon wafer, and sells them as hardware systems and as a cloud inference service. Revenue in 2025 was US$510.0 million, up from US$290.3 million in 2024; hardware contributed US$358.4 million and cloud and other services US$151.6 million. Founded in 2016 and based in Sunnyvale, California, it had 708 staff at the end of 2025. In January 2026 it announced a multi-year agreement with OpenAI valued at more than US$20 billion for 750 megawatts of inference compute, and in March 2026 it began a partnership with AWS. Its registration statement set a price range of US$150 to US$160 per share for a Nasdaq listing as CBRS. The filing discloses that a significant majority of 2025 revenue came from customers headquartered in the United Arab Emirates, notably G42 and MBZUAI.

ElevenLabs develops speech-synthesis, transcription and voice-cloning models and sells them through an API, a creator product and ElevenAgents, a platform for conversational voice agents. The company reported annual recurring revenue (ARR) of US$350 million at the end of 2025 and more than US$500 million within the first four months of 2026. It raised US$500 million in a Series D at an US$11 billion valuation on 4 February 2026, and on 30 September 2026 announced a US$300 million employee tender offer, led by Wellington and T. Rowe Price, at a US$22 billion valuation. Investors named in May 2026 include BlackRock, NVIDIA's NVentures and Deutsche Telekom's T.Capital. Enterprise customers account for 55% of revenue and the company has more than 800 staff. Voice cloning carries misuse risk, and competing speech models are offered by the largest labs.

Sierra builds AI agents that handle customer-service conversations for large companies and charges on an outcome basis, so that customers pay for resolved conversations rather than for model usage. On 30 October 2026 the company reported more than US$400 million in ARR, having added more than US$150 million in the third quarter alone; ARR grew from US$100 million to US$200 million in six months and from US$200 million to US$400 million in five. It says it works with almost half of the Fortune 50. A May 2026 post referenced by the company described a US$950 million raise at a valuation above US$15 billion. On 5 November 2026 it announced the acquisition of Anon. Sierra competes with contact-centre software vendors that are adding similar agents to existing contracts.

Glean indexes a company's documents, tickets, messages and applications and lets employees query them through search, an assistant and agents, with access controls carried over from source systems. The company announced on 28 May 2026 that it had surpassed US$300 million in ARR. Its most recent priced round on its own site is a US$150 million Series F at a US$7.2 billion valuation, led by Wellington Management and announced on 10 June 2025, when ARR had passed US$100 million. Glean belongs to a group of application companies that rely on third-party models and differentiate on data connectors and permissions. Its limitation is that cloud-suite vendors bundle comparable assistants into existing licences.

Innodata describes itself as a data-engineering and AI-systems services company supporting the development, training, post-training, evaluation and deployment of AI systems. Revenue in 2025 was US$251.7 million. Its 10-K states that leading AI laboratories and large technology companies, including five of the so-called Magnificent Seven, engage it for training and post-training data, alignment and preference optimisation, and capability and safety evaluations. Founded in 1988 and based in Ridgefield Park, New Jersey, it employed 10,107 people at the end of 2025, most of them full-time. Innodata's revenue depends on a small number of very large technology customers, so changes in one lab's data programme can move results sharply.

C3.ai sells the C3 Agentic AI Platform, a library of C3 AI Applications for uses such as predictive maintenance and supply-chain optimisation, and C3 Generative AI. Revenue for the fiscal year ended 30 April 2026 was US$250.3 million. The company is based in Redwood City, California, and had 764 full-time employees at fiscal year-end, 582 of them in the United States. C3.ai was one of the first companies to list on a US exchange as an "enterprise AI" vendor, and it sells through direct sales and alliances with cloud providers. Its annual report identifies achieving and sustaining profitability as an open question and notes pending litigation among its risks.

Heartflow provides software and AI that analyse coronary CT angiography scans to assess coronary artery disease without an invasive procedure. Revenue in 2025 was US$176.0 million. The company states that its platform has been used to assess more than 600,000 patients, including 219,000 in 2025 alone. It is based in Mountain View, California, and had 843 full-time employees at the end of 2025. Heartflow is an example of AI sold as a regulated medical product, priced per analysis and paid for largely by health insurers. Its 10-K names healthcare reimbursement policy as a risk to its ability to grow revenue.

SoundHound AI licenses voice-recognition and conversational AI to manufacturers of cars, televisions and connected devices, and sells customer-service products such as Smart Answering, Smart Ordering and Employee Assist. Revenue in 2025 was US$168.9 million. Developers can access its technology through Houndify, an open-access platform with more than 100 content domains. The company is based in Santa Clara, California, and had about 954 full-time employees at the end of 2025, 56% of them in the United States. SoundHound's products overlap with voice features that large model developers and device makers can build themselves, which limits pricing power with the largest customers.

BigBear.ai sells AI and analytics software and services to US defence, intelligence and homeland-security customers. Revenue in 2025 was US$127.7 million. The company is based in McLean, Virginia, and had 579 employees at the end of 2025, the large majority in the United States and holding security clearances; about 67% of its workforce are software engineers, data scientists and related specialists. Its 10-K describes a balance-sheet change after year-end that substantially reduced debt, which it says lets it invest from a stronger position. Revenue depends on government contract awards and their timing, and the company competes with Palantir and large defence contractors for the same programmes.

Zhipu, whose listed entity is Knowledge Atlas Technology Joint Stock Company, develops the GLM family of large language models and sells them through cloud APIs and on-premises deployments to enterprises, developers and end users. Revenue in 2025 rose 131.9% to RMB724.3 million, about US$103 million at the ECB cross rate for 31 December 2025. The loss for the year widened 59.5% to RMB4.72 billion, driven by research and development expenses of RMB3.18 billion. Gross profit was RMB296.7 million. Zhipu trades on the Hong Kong Stock Exchange under code 2513. Its main constraint is the gap between research spending and revenue, together with US export controls that limit access to the most advanced accelerators.

Skild AI develops a foundation model for robotics, a single model intended to control many types of robot body across tasks. On 14 January 2026 it announced a US$1.4 billion Series C led by SoftBank at a valuation of more than US$14 billion. On 10 September 2026 it reported passing US$100 million in ARR about ten months after its first commercial deployment. Its earlier US$300 million Series A in July 2024 was backed by Lightspeed, Coatue, SoftBank and Jeff Bezos. Skild belongs to the physical-AI segment, where models must work in uncontrolled real-world settings and where data collection is slower and more expensive than for text models.

Lovable lets users describe an application in plain language and generates the code, database and hosting, a practice often called "vibe coding". The company announced US$100 million in ARR on 23 July 2025, its most recent company-stated revenue figure. On 12 August 2026 it raised a US$400 million Series C at a US$13.3 billion valuation, led by Menlo Ventures and co-led by the Scaleup Europe Fund managed by EQT; it says users have created more than 60 million projects and that Lovable-built apps receive over 900 million visits a month. It keeps its "centre of gravity" in Stockholm and plans to grow to about 450 staff in 2026. It has acquired Molnett (November 2025) and Sutro (September 2026). Its dependence on third-party models exposes margins to model pricing.

WeRide develops level-4 autonomous driving systems (vehicles that drive themselves within a defined area without a human driver) and sells autonomous vehicles and services, including robotaxis, robobuses, robosweepers and robovans. Revenue in 2025 was RMB684.6 million, up from RMB361.1 million in 2024, about US$98 million at the ECB cross rate for 31 December 2025. Product sales were 52.6% of 2025 revenue and services 47.4%. The company commenced operations in 2017, launched paid robotaxi services in China in 2019, and had 3,801 employees at the end of 2025. It is listed in the United States and Hong Kong. Commercial scale depends on regulatory permits city by city.

Veritone describes itself as a provider of AI computing solutions and as the creator of one of the first AI operating systems, aiWARE, which orchestrates multiple AI models over audio, video and text. Revenue in 2025 was US$92.2 million. The company is based in Irvine, California, and had 446 employees at the end of 2025, 437 of them full-time, with staff in the United States, Israel and other countries. It sells to media and entertainment companies, public-sector agencies and AI developers that license its data. Veritone is a small, loss-making company whose scale limits its ability to compete with larger AI platform vendors.

Pony AI develops autonomous-driving software and operates robotaxi and robotruck services, mainly in China. Revenue in 2025 was US$90.0 million. Its principal executive offices are in the Nansha district of Guangzhou, and it had 1,669 employees at the end of 2025, a substantial majority in China and the rest in Fremont, California. The company lists the timeline and results of large-scale commercialisation of its robotaxi and robotruck services among the main uncertainties in its forward-looking statements. Like WeRide, it depends on municipal permits and on the cost of vehicle hardware falling as fleets grow.

WhiteFiber describes itself as a provider of AI infrastructure solutions: it owns high-performance computing data centres, provides cloud GPU services for AI application and machine-learning developers, and offers hosting and colocation in Tier-3 facilities. Revenue in 2025 was US$79.2 million. The company has principal executive offices in New York and employed 83 full-time staff at the end of 2025. It is one of several smaller neoclouds that compete with CoreWeave, Nebius and Crusoe for GPU leases, usually with fewer and smaller customers. Its cloud services support generative-AI training and inference, and its data centres also host other companies' equipment under colocation contracts. Small scale and customer concentration are its main limitations.

MiniMax develops foundation models across text, video, image, speech and music and sells them through an open platform for developers and through consumer applications. Its interim results announcement states that total revenue for 2025 was US$79.0 million. In the six months to 30 June 2026, revenue rose 283.1% to US$116.6 million, already above the full-year 2025 figure, and revenue from the Open Platform and other AI-based enterprise services rose from US$9.2 million to US$73.9 million, or 63.4% of the total. The loss for the half-year was US$358.0 million. MiniMax trades in Hong Kong under stock code 0100. Its model releases in 2026 include MiniMax M3 for text and H3 for video and images. Losses remain several times revenue.

Recursion uses automated laboratories and machine-learning models to identify drug candidates, and earns revenue mainly from discovery partnerships with large pharmaceutical companies. Revenue in 2025 was US$74.7 million. The company is based in Salt Lake City, Utah. It combined with Exscientia, another AI drug-discovery company, in November 2024, and its 2025 filing reflects the integration and disposals of some Exscientia assets. Recursion's value rests mainly on its clinical pipeline rather than on current revenue, so results depend on trial outcomes that are uncertain and slow to arrive. The Exscientia combination joined two of the earliest AI-first drug-discovery companies in one listed entity.

Rezolve AI states that it operates a global, AI-driven commerce platform that combines product discovery, conversation, checkout, payments and fulfilment for merchants, built on proprietary AI and machine-learning models. Revenue in 2025 was US$46.8 million. Its principal executive offices are at 21 Sackville Street, London, and it had 573 employees and 230 contractors at the end of 2025. Part of its growth has come from acquisitions of established software businesses, including Crownpeak, a digital-experience management platform founded in 2001. Rezolve is incorporated in the United Kingdom and files annual reports with the SEC as a foreign private issuer. Acquisition-led growth makes underlying organic growth harder to measure.

Nscale, incorporated in May 2024, builds and operates data centres and an AI cloud in low-cost power markets including Norway, Portugal, Iceland and the United States. Revenue was US$33.0 million in 2025 and US$19.1 million in 2024, then US$140.6 million in the first half of 2026. A Portuguese campus will support an expanded collaboration with Microsoft through more than 66,000 NVIDIA Vera Rubin NVL72 GPUs from late 2027. On 9 March 2026 it raised a US$2 billion Series C at a US$14.6 billion valuation, led by Aker ASA and 8090 Industries. It agreed on 28 July 2026 to acquire Anyscale, the company behind the Ray distributed-computing framework, and filed in September 2026 to list on the NYSE as NSCL. Its very short operating history is the principal risk.

Kodiak AI describes itself as a provider of physical AI focused on AI-powered autonomous-vehicle technology for trucking. Revenue in 2025 was US$3.8 million. Legacy Kodiak was founded in 2018 and began driverless commercial operations in December 2024. Its first driverless deployments have been in off-road and industrial settings, where routes are controlled, ahead of broader highway operation. Physical AI is the industry term for AI systems that perceive and act in the physical world, such as vehicles and robots. The company argues that driverless trucks can improve road safety and truck utilisation while reducing costs. Revenue is still very small relative to costs, and the 10-K highlights its limited operating history.

Aurora develops self-driving technology, with commercial driverless trucking as its first product. Revenue in 2025 was US$3.0 million. The company was founded in 2017 by Chris Urmson, Sterling Anderson and Drew Bagnell, is based in Pittsburgh, Pennsylvania, and had about 1,900 employees at the end of 2025. Its annual report sets out the expected benefits of driverless trucks, including lower fuel use through eco-driving and off-peak deployment. Aurora's revenue is at an early stage relative to its research spending, so it ranks near the bottom of the revenue tier despite its scale of operations.

Figure AI builds humanoid robots and the Helix vision-language-action models that control them. On 16 September 2025 it announced more than US$1 billion in committed Series C capital at a US$39 billion post-money valuation, led by Parkway Venture Capital with Brookfield Asset Management, NVIDIA, Intel Capital, Qualcomm Ventures and others. Figure 03, its third-generation robot, was introduced on 9 October 2025; Helix 02, described as full-body autonomy, followed on 27 January 2026, and Helix 2.5, which the company says generalises zero-shot across 30 homes, on 17 September 2026. It reported ramping Figure 03 production in April 2026 and delivering F.03 robots to BMW in June 2026. Figure has disclosed no revenue, and humanoid robots remain unproven at commercial scale.

Crusoe develops power and data-centre campuses (which it calls AI factories) and runs Crusoe Cloud on them. On 17 September 2026 it announced the initial closing of a US$3.9 billion Series F at a US$30.9 billion post-money valuation, co-led by Atreides Management, Mubadala Capital and Valor Equity Partners, with Founders Fund, GIC and NVIDIA participating. The company reports more than US$140 billion of total contracted value, more than 6 GW of gross contracted capacity with 1 GW delivered and operational, and more than US$100 million in ARR for its Crusoe Managed Inference service. It has no company-wide revenue figure on its own pages, so it is ranked by valuation. The release is datelined Denver. Building at this scale requires very large debt financing and long-term power contracts.

Scale AI supplies labelled and expert-generated data for training and evaluating AI models, and builds AI applications for governments. On 12 June 2025 it announced a significant new investment from Meta Platforms that values the company at more than US$29 billion; founder Alexandr Wang joined Meta to work on its AI efforts while remaining on Scale's board, and chief strategy officer Jason Droege became interim chief executive. Recent company posts cover AI-literacy programmes and evaluation benchmarks such as a legal-reasoning benchmark with Singapore's IMDA. Scale has not published revenue. Meta's stake creates a commercial risk with other model developers that may not want a competitor-linked supplier handling their data.

Mistral AI develops open-weight models (whose trained parameters are published for others to run) alongside commercial models and a compute service. On 8 September 2026 it raised €3 billion in a Series D at a post-money valuation of more than €21 billion, which it called the largest equity round completed by a European technology company; Samsung Electronics led, with the EQT-managed Scaleup Europe Fund and PSG Equity as co-leads. At the ECB reference rate for that day, €21 billion equals about US$24.4 billion. Its previous round, €1.7 billion in September 2025, was led by ASML. The company operates in 20 countries and has announced partnerships with NVIDIA in March and October 2026 and an in-region inference and compute offering in August 2026. Mistral has disclosed no revenue and competes against labs with far larger compute budgets.

Mercor recruits domain experts, such as lawyers, doctors and engineers, to produce training data and evaluations for AI developers, and is expanding into evaluation and enterprise agent infrastructure. On 10 November 2026 it announced a US$750 million Series D at a US$20 billion valuation, with participation from Benchmark, Felicis, NVIDIA, Salesforce and General Catalyst; the company says it spends three times as much on inference as on employee salaries. Its earlier Series C valued it at US$10 billion. The company lists San Francisco in its site footer and has opened offices in New York and London. Mercor has not published revenue. Its business depends on a small number of model developers whose data needs can change quickly.

Harvey builds AI tools for legal work, including research, drafting, due diligence and workflow agents, sold on enterprise subscriptions to law firms and in-house legal departments. On 9 September 2026 it announced a US$550 million round at a US$15.5 billion valuation, co-led by Diffusion and Lightspeed Venture Partners. The company states that 80% of Am Law 100 firms use Harvey and that five of the Fortune 10 are in-house clients, and it planned offices in Tokyo and Boston. In November 2026 it added Danish legislation through a partnership with Djøf Forlag. Harvey has not published revenue. Its main risks are professional liability for incorrect outputs and competition from legal publishers that own primary-law databases.

Applied Intuition sells simulation, vehicle operating-system and autonomy software to car manufacturers, trucking fleets, mining and construction operators and defence organisations. On 17 June 2025 it announced a Series F, led by BlackRock and Kleiner Perkins with Fidelity Management & Research and Lux Capital, that valued it at US$15 billion, more than double its valuation a little over a year earlier. In 2026 it announced selection by the US Department of War for an enterprise autonomy pipeline (14 May), the launch of Dana, an agentic platform (21 July), and selection by the US Navy for field AI perception (22 October). The company has not disclosed revenue or the size of its Series F. Defence work subjects it to procurement timing and export rules.

Replit offers a cloud development environment in which its Agent writes, runs and deploys applications from natural-language instructions. In March 2026 it raised US$400 million in a Series D at a US$9 billion valuation, led by Georgian with Andreessen Horowitz, Coatue and the Qatar Investment Authority, up from US$3 billion about six months earlier. The company, based in Foster City, California, reports millions of users and more than 500,000 professional users; it did not disclose revenue but said it was on track for US$1 billion in ARR by the end of 2026. Because that is a target rather than a reported figure, Replit is ranked by valuation. It shares the margin exposure of other app builders that pay third-party model providers for each request.

Cohere develops language and retrieval models for enterprises and governments, emphasising private deployment in customers' own clouds or data centres. Its most recent priced round on its own site is US$500 million at a US$6.8 billion valuation, announced on 14 August 2025. In 2026 it signed an agreement with Germany's Aleph Alpha (16 September), signed a letter of intent to acquire AISLE for cyber-resilient sovereign AI (15 October) and advanced a C$4 billion commitment to Canadian AI capacity and workforce expansion (30 October). Cohere has not published revenue on its newsroom. Its focus on sovereign deployments narrows its market to buyers that will pay for control over data and infrastructure.

Suno develops models and an app that generate songs, including vocals, from text prompts. On 3 June 2026 it announced more than US$400 million in Series D funding at a US$5.4 billion post-money valuation, led by Bond Capital with IVP, Forerunner, Union Square Ventures and others. Its blog lists a partnership with Warner Music Group (25 November 2025), a global partnership with BMG (12 August 2026) and one with Believe and TuneCore (8 September 2026), as well as the acquisition of WavTool in June 2025. Suno has not disclosed revenue. Licensing terms with rights holders determine both its cost base and which music its models can learn from.

Decagon builds AI agents that handle customer-support conversations across chat, email and voice. On 27 January 2026 it announced a US$250 million Series D at a US$4.5 billion valuation, roughly triple its valuation six months earlier, led by new investors Coatue Management and Index Ventures. Returning investors include Andreessen Horowitz, Accel, Bain Capital Ventures and Ribbit Capital. The company reported more than 100 new enterprise customers in its past fiscal year, naming Avis Budget Group, Block and Deutsche Telekom, along with Oura Health, Affirm and Chime. Decagon competes directly with Sierra and with established customer-service software vendors, and has not disclosed revenue.

Synthesia lets companies produce videos with AI-generated presenters (avatars) and voices from a script, and is used mainly for training and internal communications. On 26 January 2026 it raised US$200 million in a Series E at a US$4 billion valuation, led by GV (Google Ventures) with Evantic and Hedosophia. The company, headquartered in London, says it is used by more than 90% of the Fortune 100 and has more than 2 million users. Synthesia has not disclosed revenue. Its use of digital likenesses of real people requires consent and moderation controls, which limit what customers can produce. General-purpose video models from large labs are a competitive threat to its avatar-based product.

Hippocratic AI builds AI agents for patient-facing healthcare tasks such as pre-operative calls, follow-up and chronic-care check-ins, limited to non-diagnostic work. On 3 November 2025 it announced a US$126 million Series C at a US$3.5 billion valuation, led by Avenir Growth. That followed a US$141 million Series B led by Kleiner Perkins at a US$1.64 billion valuation on 9 January 2025, and a US$17 million investment from NVIDIA's NVentures in September 2024. Its investors therefore include both NVIDIA's venture arm and the two firms that led successive rounds, Kleiner Perkins and Avenir Growth. The company has not disclosed revenue. Clinical-safety validation and health-system procurement cycles slow its deployments.

Together AI operates a cloud for inference, fine-tuning and training of open-source models, sold through an API and dedicated GPU clusters. On 20 February 2025 it announced a US$305 million Series B led by General Catalyst and co-led by Prosperity7, at a valuation of US$3.3 billion. In October 2026 it expanded its enterprise inference capacity with a dedicated B300 cluster with IBM Cloud and NVIDIA, and it has partnered with Y Combinator on a GPU cluster for YC companies. Together has not published revenue. Its valuation is the oldest in this tier, and it competes on price with hyperscalers and with model developers that host their own open models.

Sakana AI is a Tokyo-based research company developing AI models using methods inspired by natural evolution and collective intelligence, with a commercial focus on Japanese enterprises, finance and government. On 17 November 2025 it announced a Series B of about ¥32 billion at a post-money valuation of about ¥432 billion, with investors including MUFG, Khosla Ventures, Factorial, Macquarie Capital, Mitsubishi Electric, Google and In-Q-Tel. At the ECB cross rate for that day, ¥432 billion equals about US$2.8 billion. In early 2026 it announced a partnership with Google (23 January) and strategic investments from Salesforce Ventures (19 February) and Citi (24 February). Sakana has not published revenue, and its scale is small compared with US and Chinese labs.