We wanted to map out which large enterprises are using which of the top emerging AI platforms in 2025. The goal: help InnoLead’s corporate members understand which of their peers are already beginning to weave these new AI tools — not Microsoft Copilot or Google Gemini, from already entrenched enterprise software providers — into their business.
LLM Deep Research to the rescue. (More about our process for this research in the “Methodology” section below.) We created a list of the top 25 “unicorn” startups, with valuations greater than $1 billion, and then hunted for articles, case studies, and customer pages touting some of their early adopters.
As we did this, a clear structure emerged. Frontier model makers like OpenAI and Mistral operate as the “brains” offering general-purpose intelligence. Application-layer startups like Moveworks or Harvey wrap those brains in jobs-to-done like legal assistants, IT agents, customer service assistants, and so on. And infrastructure firms like Scale or Snorkel are quietly refining the data plumbing, so the models can do more than hallucinate.
In terms of the enterprise adopters (both direct clients and strategic partners), I found these four patterns interesting.
- Application Trumps Model. Legal teams don’t buy Claude or ChatGPT. They buy Harvey. Marketing teams choose Writer, not Mistral. The unit of enterprise adoption is workflow, not architecture. The exceptions to this are the very large enterprises tuning their own models who then purchase services from the likes of Cohere, Glean, H20, and Instabase.
- Finance and Telecom Lead. Firms in finance and telecom like JPMorgan, Deutsche Telekom, Citi, and Vodafone show up across the board. Documentation and compliance-heavy environments are proving fertile ground for AI deployment, and big valuations for AI/ML doc extraction unicorns like Snorkel.
- Large Consultancies Act as Middleware. Accenture, PwC, Cognizant aren’t just clients, they’re connectors and implementers. They embed GenAI into large-scale digital transformations, often becoming the recommendation and delivery engine.
- Text Still Dominates. Despite the sizzle of video, voice, and image generation, text remains the dominant modality in real-world enterprise deployments. Contracts, tickets, chats, and emails remain the enterprise’s raw material.
Of course, we know that some of the more established tech providers, like Microsoft, Google, Amazon, and Salesforce, have a deeper roster of enterprise customers than the 25 unicorn startups below. Their enterprise customer list includes the likes of Mondelēz, Accenture, Vodafone (Microsoft CoPilot), Deutsche Telecom, Citigroup, Samsung (Google Gemini), United Airlines, Salesforce, Accenture (Amazon Bedrock/Nova) and General Mills, PepsiCo, Mercedes-Benz (Salesforce/Agentforce)
In summary, this is clearly the most dynamic space of economic activity in tech, and what we are seeing is a time of rapid experimentation and learning.
Who’s winning and who will win? From this view, there’s no one “winning” at the model level yet, despite ChatGPT having the most mindshare, user growth, and annual recurring revenue. (ChatGPT created the image above, of unicorns and elephants dancing.) Enterprises are not converging on a single “stack” of tools and platforms; they’re all still trying to find out what works for their organization.
The table below is available in spreadsheet form for InnoLead members.
Provider (Ranked by Valuation) | Description | Enterprise Customers |
---|---|---|
OpenAI (ChatGPT) | Pioneers in multimodal frontier AI, OpenAI powers both enterprise and consumer experiences through its ChatGPT platform. | PwC (PricewaterhouseCoopers) ~101,000 employees (US+UK) Morgan Stanley ~82,000 employees Also Block (11,000), Carlyle (2,200), and Estée Lauder (63,000) among ChatGPT Enterprise customers |
Anthropic (Claude) | Anthropic’s Claude is a frontier multimodal model designed for safe, steerable AI interactions. Deployed by leading enterprises, it supports productivity, knowledge work, and agent-based use cases with a focus on alignment and controllability. | Deloitte ~460,000 employees (global) SK Telecom ~23,000 employees (telecom) Rakuten ~28,000 employees Lotte Homeshopping ~10,000 employees Thomson Reuters ~25,000 employees Amazon (AmazonQ) ~1.M employees Boston Consulting Group (BCG) ~25,000 employees |
X.AI | X.AI develops advanced AI systems including the Grok conversational AI platform, focusing on enterprise-grade AI solutions and partnerships with major technology and financial services organizations. | Walt Disney Company ~173,250 employees Nike ~79,400 employees Coca-Cola ~69,700 employees BlackRock ~20,000 employees And partnerships with Oracle ~164,000 employees (for cloud) and NVIDIA ~26,000 employees (for infrastructure) |
Scale AI | Scale AI provides critical infrastructure for enterprise and government AI initiatives, specializing in data labeling, model evaluation, and synthetic data generation. It plays a key role in enabling large-scale, multimodal AI deployments across commercial and defense sectors. | Microsoft ~221,000 employees Meta ~71,000 employees Cisco ~80,000 employees Fox Corporation ~10,000 employees Koch Industries ~120,000 employees U.S. Army (18th Airborne Corps) ~90,000 personnel |
Perplexity AI | Perplexity AI delivers an AI-native search experience by combining real-time web retrieval with conversational responses. Positioned at the application layer, it powers information access for enterprise users through integrations and strategic deployments. | HP ~58,000 employees SAP ~45,000 employees Stripe ~8,000 employees (included due to strategic relevance) Zoom ~8,400 employees (included due to strategic relevance) |
Glean | Glean provides enterprise AI search by connecting and indexing knowledge across workplace apps. Its semantic search and retrieval capabilities help employees find information, people, and resources with contextual precision across large-scale organizations. | Booking.com ~14,000 employees Deutsche Telekom ~207,000 employees Wealthsimple ~1,000 employees Pure Storage ~4,000 employees |
Mistral | Mistral AI develops high-performance open-weight language models optimized for enterprise deployment and developer accessibility. With a focus on European institutions and global enterprises, Mistral is rapidly gaining adoption in finance, automotive, and telecom sectors. | Cisco ~84,900 employees Stellantis ~300,000 employees BNP Paribas ~190,000 employees Orange ~130,000 employees Making Mistral models available to enterprise customers: IBM ~311,000 SAP ~105,000 employees |
Cohere | Cohere develops frontier language models focused on enterprise needs, offering tools for retrieval-augmented generation, classification, and search. Its models are used across industries including finance, electronics, and IT, often embedded via APIs or partner platforms. | Salesforce ~79,000 employees SAP ~105,000 employees Oracle ~143,000 employees Royal Bank of Canada ~92,000 employees LG ~140,000 employees Fujitsu ~124,000 employees |
Abridge | Abridge uses AI to transform clinical conversations into structured notes, streamlining medical documentation across major health systems. Its ambient AI solutions are deployed enterprise-wide in leading hospitals to reduce provider burnout and enhance EHR workflows. | Johns Hopkins Medicine ~40,000 employees (deploying ambient AI system for 6,700 clinicians) Mayo Clinic ~73,000 employees (scaling Abridge enterprise-wide) Inova Health System ~20,000 employees (using Abridge for documentation) Duke Health ~25,000 employees Emory Healthcare ~24,000 employees |
Hugging Face | As the hub of open-source AI, Hugging Face supports researchers and developers with multimodal tools and model hosting. | IBM ~311,000 employees (partnered on IBM watsonx platform) Amazon/AWS ~1,500,000 employees (cloud partnership) Dell Technologies ~133,000 employees (Dell-HF AI Hub) (Also supported by many others; e.g. ServiceNow and NVIDIA are investors/partners.) |
AlphaSense | AlphaSense powers enterprise market intelligence by applying AI to search, summarize, and extract insights from financial documents, earnings calls, and internal content. It is widely adopted across sectors for decision support and competitive research. | Siemens ~311,000 employees (uses AlphaSense for market intel) Schneider Electric ~128,000 employees Vale ~74,000 employees NetApp ~11,800 employees (reported that 85% of the S&P 100 use AlphaSense) |
Runway | Runway ML brings generative video capabilities to creative industries, offering AI-powered tools for editing, synthesis, and content creation. It serves enterprise media, entertainment, and advertising clients through both partnerships and production-level deployments. | Paramount (CBS) ~24,000 employees (The Late Show uses Runway) New Balance ~7,500 employees (design use case) Omnicom ~70,000 employees Lionsgate ~7,000 employees (included for enterprise-scale deployment) |
ElevenLabs | ElevenLabs delivers advanced AI voice synthesis technology used in media, entertainment, telecom, and publishing. Its ultra-realistic multilingual speech capabilities are deployed across high-impact enterprise use cases from news narration to customer experience enhancement. | PwC (internal) ~4,000 in legal division (piloting AI voice for consulting) Audacy ~10,000 employees NVIDIA ~26,000 employees Bertelsmann ~145,000 employees Deutsche Telekom ~207,000 employees (investor/partnership) ESPN ~8,000 employees (noted via parent company Disney) The Washington Post ~2,500 employees |
Ironclad | Ironclad streamlines contract lifecycle management through AI-powered workflows that automate legal document creation, negotiation, and insights. Its platform is adopted by major enterprises to reduce friction in legal operations and accelerate deal velocity. | L’Oréal ~85,000 employees Mastercard ~29,000 employees Staples ~70,000 employees Hormel ~20,000 employees |
Harvey | Harvey AI brings large language models into legal workflows, enabling document review, contract analysis, and regulatory compliance automation. Its deployments span global law firms and major enterprises integrating legal AI into their internal teams and service offerings. | Deutsche Telekom ~216,000 employees Bayer AG ~100,000 employees (corporate legal using Harvey) PwC ~328,000 employees Allen & Overy (now A&O Shearman) ~5,500+ employees |
Moveworks | Moveworks provides enterprise-ready conversational AI to automate IT, HR, and other internal support workflows. Integrated across large organizations, its AI assistant resolves employee requests in real time, reducing operational overhead and boosting productivity. | Western Digital ~65,000 employees LinkedIn ~21,000 employees Broadcom ~15,000 employees Palo Alto Networks ~10,000 employees Leidos ~47,000 employees CVS Health ~300,000 employees (Also deployed at Slack, DocuSign, Autodesk, etc.) |
Synthesia | Synthesia transforms enterprise training and communication with AI-generated video avatars and multilingual narration. | WPP (GroupM) ~109,000 employees (marketing giant uses AI video) Heineken ~85,000 employees (70k staff trained via Synthesia) BSH (Bosch-Siemens) ~62,000 employees (global appliance training) DuPont ~23,000 employees (AI video for workforce upskilling) |
DeepL | DeepL provides enterprise-grade AI translation tools known for high linguistic accuracy and data privacy. Trusted by banks, manufacturers, and media companies, DeepL helps global organizations localize content and streamline multilingual communication. | Deutsche Bahn ~322,000 employees (German Rail, uses DeepL enterprise) Beiersdorf AG ~20,000 employees (consumer goods, DeepL customer) Panasonic Connect (Panasonic) ~240,000 (Parent co.) KBC Bank ~41,000 employees Nikkei ~10,000 employees |
Liquid AI | Spun out of an MIT research project, Liquid AI is a frontier research and model development lab building Liquid Foundation Models (LFMs) for multimodal deployment with smaller memory and GPU requirements. Through industry collaborations, they are targeting research, infrastructure, and developer tooling layers. | AMD ~24,000 employees Amazon Web Services (AWS) ~1,600,000 employees Perplexity AI ~150 employees (strategic integration partner) Lambda ~50 employees (developer partner for deployment testing) |
Mercor | Mercor is an AI-powered talent marketplace connecting enterprise clients with global technical talent. Leveraging machine learning for role matching and productivity optimization, it supports large-scale hiring across banking, consulting, and tech industries. Using AI to streamline hiring pipelines. | JPMorgan Chase ~317,000 employees Goldman Sachs ~40,000 employees Dell Technologies ~108,000 employees (From Felicis Ventures public note: “[Mercor has] nearly all top AI labs and hyperscalers as customers”) |
Writer | Writer delivers enterprise-grade generative AI for content creation, governance, and brand alignment. Its platform is integrated across marketing, communications, and product teams to enforce tone, accelerate output, and ensure compliance at scale. | Intuit ~14,000 employees Uber ~33,000 employees Vanguard ~17,000 employees Salesforce ~79,000 employees Qualcomm ~50,000 employees |
H2O AI | H2O.ai builds enterprise-grade AutoML tools, enabling organizations to rapidly develop AI-driven applications. | AT&T ~160,000 employees Hitachi ~350,000 employees PwC ~300,000 employees Wells Fargo ~268,000 employees (investor & customer) |
Snorkel AI | Snorkel AI accelerates AI development by programmatically labeling training data, enabling rapid model development for structured and unstructured datasets. Trusted by Fortune 500 firms, Snorkel is widely used in financial services, tech, and enterprise ML pipelines. | U.S. Air Force ~685,000 personnel (uses Snorkel for data labeling in intelligence) Goldman Sachs ~45,000 employees (investor & user for AI data prep) BNY Mellon ~51,000 employees JPMorgan Chase ~293,000 employees (implied via industry report) Google ~156,000 employees |
Instabase | Instabase offers enterprise automation through AI-powered document processing and workflow orchestration. | Standard Chartered ~85,000 employees (strategic investor & customer for document AI) Sonic Automotive ~47,000 employees Direct Assurance (AXA Group) ~145,000 employees NatWest ~60,000 employees Uber ~33,000 employees US Patent and Trademark Office ~13,000 employees |
Ada Cx | Ada powers enterprise customer service with text and voice bots that automate common support interactions. | Meta (Facebook) ~71,000 employees Block (Square) ~11,000 employees Telus ~90,000 employees AirAsia ~22,000 employees |
Methodology
We began with a list of 35 GenAI startups which, according to sources like TheInformation, Pitchbook, and Crunchbase, had crossed the critical 1B valuation into unicorn territory. We then conducted LLM-powered ‘Deep Research’ using both Perplexity and ChatGPT side-by-side.
To get the best results from any Deep Research project, it’s important to be very clear and precise in your prompts. The best current approach we know of is getting the LLM to write those research prompts. Through some back and forth meta-prompting we generated the following Deep Research prompt for both systems to execute.
Deep Research Task: GenAI Unicorn Enterprise Client Identification Objective: Identify 3-5 major enterprise clients (10,000+ employees) for each GenAI unicorn startup from the provided list. Prioritize Fortune 500/publicly traded companies with verified deployment evidence. Methodology: Primary Source Extraction – Scrape: `[startup_domain]/customers`, `[startup_domain]/case-studies`, `[startup_domain]/press` – Capture: Client logos, named case studies, press release partnerships Secondary Source Triangulation – Cross-reference with: – Crunchbase partnership data – LinkedIn implementation announcements – TechCrunch/VentureBeat deployment reports – Validate headcount via: Apollo.io, SEC filings, Bloomberg terminal data Gap Analysis Protocol – If no clients listed on startup site: – Search pattern: `”[Startup Name]” + (“enterprise deployment” OR “enterprise case study”) + (“Fortune 500” OR “10,000 employees”) site:techcrunch.com OR site:venturebeat.com` – Prioritize sources: CEO interviews, earnings call transcripts, Gartner/IDC reports Output Requirements: – Structured markdown table with columns: | Startup | Verified Clients (Enterprise) | Client Headcount | Evidence Source | – Evidence citations: [URL] for each client verification – Confidence rating per entry (High/Medium/Low) based on source quality Constraints: – Exclude non-enterprise clients (<10k employees) – Flag unverifiable entries for manual review – Time allocation: 8min per startup max |
The Deep Research for all 35 startups took each LLM about 15-20 minutes. Coffee break time. We then had two lists of unicorns and their enterprise clients which we needed to reconcile against each other. Not surprisingly, each LLM found examples the other missed. We then had each LLM fact check the other.
We followed that with some additional manual fact checking to address hallucinations in the data set. But to be clear, for all its hyped ability, the “Deep Research” capabilities of these LLMs is still based on the underlying error-prone RAG (retrieval augmented generation) that often finds pieces of a story on the web but takes those pieces out of context. For example, a mention of a large enterprise and a unicorn might refer to a lawsuit and not a partnership. Yes, RAG can make that kind of mistake.
Finally, after removing some hardware unicorns and some purely software developer players, we whittled down to the following 25 unicorns and their notable enterprise clients (primarily those with > 10,000 employees, though we included a few interesting customer examples from smaller big organizations). There are certainly more enterprises engaged with these unicorns. We aimed for representativeness rather than completeness. We hope this table gives some insight into the rate of change and the rapid proliferation of GenAI across private and public sectors. Unicorns are sorted by valuation, as of June 21st, 2025.
You must be logged in to post a comment.