Since the launch of enterprise LLM/AI platforms like ChatGPT, and Claude, a few questions have remained frustratingly difficult to answer: Which industries and countries are actually on the forefront of adoption? Who is lagging behind? Are enterprises in the 1st, 3rd, or 5th inning of adoption? And which type of companies typically choose Claude over ChatGPT?
The Survey Problem
Existing research on enterprise LLM adoption has relied mostly on surveys, which are incredibly flawed. These surveys ask executives whether their companies are using AI or planning to adopt it. Yet these figures are nearly meaningless because of:
- Inherent bias – executives want to appear innovative and cool to investors and customers
- Definition ambiguity – does “using AI” mean one employee tried ChatGPT, a department running a pilot, or enterprise-wide deployment?
- Selection bias – AI-conscious companies are more likely to respond to AI surveys
The result: We have plenty of data on what executives say about AI, but little rigorous evidence on what’s actually happening in large enterprises. We have hand wavy projections on what % of companies are adopting AI, but no concrete numbers.
A Different Approach: Direct Observation at Scale
This study takes a drastically different approach. Rather than asking companies what they’re doing, we directly observe enterprise LLM deployment by analyzing digital signals that reveal when companies roll out enterprise LLM tools to their workforce.
Our data comes from tracking corporate domains associated with 76,084 companies with 500 or more employees globally. When a company deploys ChatGPT, Claude, Perplexity, or Mistral, we detect specific signals in their DNS configurations that allow us to identify:
- Which companies have deployed an enterprise LLM product to their workforce
- Which vendor they’ve chosen
- When deployment occurred
This approach captures actual deployment behavior rather than intentions, plans, or pilot programs. We only count a company as “adopted” when we observe evidence of a company-wide rollout – not when a single team runs a pilot or individual employees use consumer versions. In short, we are measuring what companies do, not what they say.
We define an enterprise LLM platform as 1 of these tools: ChatGPT, Claude, Perplexity or Mistral. To be clear, we do not count Google Gemini, Microsoft Copilot, or any Chinese LLM from Baidu or Alibaba. We also do not count vertical AI tools like Harvey AI or Decagon AI. So we admit upfront this is a limitation of our study.
This dataset represents, to our knowledge, the first comprehensive, observational study of enterprise LLM adoption based on actual deployment behavior rather than self-reported surveys.
Coverage and Scope
Our dataset includes:
- 76,084 companies with 500+ employees globally
- 18 months of deployment data (May 2024 – October 2025)
- 4 enterprise AI LLM vendors: ChatGPT, Claude, Perplexity and Mistral
Now that the methodology, and limitations are known, let’s dig into the findings…
1. Adoption among the Fortune 500 has tripled in the past year.
At this time last year (October 2024), only 22 Fortune 500 companies had deployed at least one enterprise AI LLM platform across their workforce.
As of October 2025, 67 Fortune 500 companies now have deployed an enterprise LLM product (13.4% of the Fortune 500) to their employees. A more than 3x increase from a year ago.
Looking more broadly, 6% of companies with >= 500 employees globally have deployed enterprise LLM tools across their workforce.
To provide some context to how early this is, roughly 20% of these same companies have deployed Atlassian (a mature SaaS product) to their workforce. So if we assume that enterprise LLM platforms will reach the popularity of Atlassian, at its peak, then we’re probably in the 3rd inning right now.
However, if we assume that enterprise LLM platforms will reach the popularity of Microsoft Teams someday, we’re somewhere between the 1st and 2nd inning.
2. Marketing/Advertising companies lead all industries in adoption, followed closed by Tech/Software
Marketing (15.28%) and Tech (13.29%) lead for obvious reasons – marketing firms create a lot of content, and tech companies have both the infrastructure and “try new things” culture baked in. No mystery there.
The real story is in the gaps.
Why is Financial Services (6.73%) adopting 2-3x faster than Healthcare (3.35%) despite similar regulatory scrutiny?
For starters, banks have clear, measurable use cases: customer service chatbots, document review, and compliance monitoring. A bank can A/B test AI on 10,000 support tickets and see exactly how much it saves. Healthcare, on the other hand, can’t do this – medical decisions are too high-stakes. There’s no ethical way to run the experiment.
Construction at 4.69% is surprisingly high – higher than retail (4.19%) and education (4.36%). AI is probably attacking the massive paperwork layer: permits, contracts, change orders, safety documentation.
Telecom companies at 3.42% seemed surprisingly low, as they all have massive customer service operations that should benefit from AI,. Maybe they already invested heavily in previous-generation chatbots and face switching costs. Or maybe telecom customer service is harder than it looks?
Government (1.32%) lagging makes perfect sense: 18-month procurement, risk-averse culture, no competitive pressure. The surprise would be if government weren’t last.
Which Professional Services Are Actually Using AI?
I wanted to delve into professional services, since it encompasses a lot of little sub-industries. Where there subcategories that adopted enterprise AI more than others?
Business consulting leads at 11.10%, slightly ahead of law firms at 8.98%. Consulting makes obvious sense – these firms sell research, and strategic recommendations, which is basically what LLMs do. McKinsey analysts can use AI to synthesize market research or draft client presentations, see immediate productivity gains, and the risk if something’s slightly off isn’t catastrophic (because they’re McKinsey for crying out loud).
Law firms at 8.98% is the surprise here. Legal work has genuine malpractice risk – an AI hallucination in a contract could be a lawsuit. Yet surprisingly, law firms are adopting at nearly the same rate as consultants. And remember, this is excluding existing popular law AI tools like Harvey AI.
This tells you the efficiency gains must be really compelling. When you’re billing $1,500/hour and AI can handle initial document review or legal research in a fraction of the time, apparently the math works even with careful human oversight. Big law firms seem to have figured out how to integrate AI safely for high-volume, lower-risk tasks while keeping humans in the loop for anything that matters.
Accounting at 5.90% seems surprisingly low to me, but it probably reflects genuinely murky use cases. Accounting work is about numbers, compliance, and structured data – not really where LLMs shine the most. The firms that adopted are probably using AI for client communications and advisory work, not actual audit or tax prep. Worth noting: almost every single big accounting firm adopter uses ChatGPT only. Zero Claude, zero Perplexity, zero Mistral.
HR services at 3.79% are way behind, and honestly it’s not entirely clear why. On paper, HR should be higher – these firms do tons of text-heavy work like job descriptions, employee handbooks, recruiting outreach, and training materials. A possible theory could they’re afraid of discrimination liability? Using AI for candidate screening or performance review language could create bias issues that are both legally risky and reputationally toxic.
Whatever the reason, HR is clearly more hesitant than other professional services, even those with comparable risk profiles.
3. Israel and US leads all regions in enterprise AI adoption, with the EU lagging
Enterprise AI adoption reveals a stark geographic reality: adoption rates range from 12.2% in Israel to 1.35% in the Middle East, among large companies worldwide.
The United States dominates with 9.67% of large companies deploying enterprise AI. This probably isn’t surprising to anyone.
But here’s what might be: Israel leads everyone at 12.2%. For a country with a tiny domestic market, this is remarkable. It being the leader reflects its status as a tech powerhouse with deep Silicon Valley connections and a culture of early adoption.
The “rich country club”—Switzerland (6.89%), Canada (6.39%), Australia/New Zealand (5.92%) clusters around 6%, roughly two-thirds the US rate. Switzerland’s 6.89% is particularly notable because it sits outside the EU regulatory framework.
Great Britain at 5.32% occupies an awkward middle ground. Post-Brexit Britain has regulatory independence but hasn’t translated that into US-level adoption (though it is ahead of most of the EU)
Europe’s Puzzle: Same Rules, Different Results
The European Union shows 4.21% adoption – less than half the US rate. This will probably prompt endless debate about whether EU regulations (the AI Act, GDPR) are “killing innovation.”
But it’s not all uniform, as adoption within the EU varies dramatically: Finland at 8.9% nearly matches the US, while Portugal at 1.2% is closer to emerging market levels. That’s a 7-fold gap within the same regulatory framework.
Finland’s high adoption probably reflects Nokia’s legacy. The country went through a massive tech transformation in the 1990s-2000s when Nokia dominated global mobile phones, building deep technical infrastructure and a culture of early technology adoption across Finnish companies. Even after Nokia’s decline, that institutional knowledge and comfort with new tech stuck around. Finland also has strong English proficiency and close ties to Silicon Valley, making it easier to adopt US-based AI tools quickly.
France (3.17%), Italy (2.75%), and Portugal (1.2%) clustering at the bottom is harder to explain cleanly. One theory: these are larger Southern European economies with more traditional corporate cultures and stronger labor protections that make workforce transformation politically sensitive. France in particular has powerful unions and strict employment laws – deploying AI tools that might displace white-collar workers is a more fraught decision than in Finland or Switzerland.
The EU’s aggregate 4.21% compared to 6.89% in Switzerland does suggest regulatory friction plays some role. GDPR and the AI Act probably create compliance costs that particularly burden smaller companies. The regulations aren’t killing innovation, but they’re definitely slowing it down.
4. Almost half of Claude customers are also ChatGPT customers
It’s ChatGPT vs Claude. Only one of them can win this market, right? Well.. maybe not.
To understand competitive dynamics beyond simple market share, we analyzed if a non-trivial # of companies deploy multiple enterprise AI products – and in what order.
The overlap data reveals a striking asymmetry: nearly half of Claude customers (48.66%) also pay for ChatGPT , while only 6.5% of ChatGPT customers pay for Claude.
The timing data exposes the actual market dynamic at play.
| Adoption Sequence | # of companies | Average days between adoption | Percentage of dual adopters |
| ChatGPT first, then Claude | 260 | 128 | 52.4 |
| Anthropic first, then ChatGPT | 102 | 45 | 20.6 |
| ChatGPT first, then Perplexity | 100 | 156 | 20.2 |
| Perplexity first, then ChatGPT | 28 | 15 | 5.6 |
| Claude first, then Perplexity | 4 | 52 | 0.8 |
| Perplexity first, then Claude | 2 | 31 | 0.4 |
The overlap data reveals two distinct adoption patterns.
Pattern 1: Simultaneous Dual Deployment
A significant portion of the dual-vendor companies deployed both platforms within 45 days of each other. These aren’t organizations trying one platform, finding it inadequate, and adding another – they’re probably making the decision to use both from the outset (because 45 days in between is way too short). Just like a lot of companies adopt a multi-cloud strategy, a good number of companies are adopting a multi-AI strategy.
Different departments likely identified different requirements during evaluation: engineering teams may have advocated for Claude’s technical capabilities for code-heavy workflows, while business teams pushed for ChatGPT’s ecosystem and brand recognition. Rather than forcing a winner-takes-all decision, these enterprises chose a hybrid approach where both platforms serve distinct organizational needs from day one.
Pattern 2: ChatGPT-First, Claude-Later
The second pattern involves companies that deployed ChatGPT first and added Claude months later. This sequential adoption suggests a different motivation: these organizations standardized on ChatGPT as their primary platform, but over time identified specific use cases (ie. coding) or teams where Claude performed better or offered necessary capabilities ChatGPT lacked.
This could be technical teams needing longer context windows, specialized reasoning tasks where Claude excels, or simply dissatisfaction with ChatGPT’s performance on particular workloads. The decision to add a second expensive enterprise contract indicates the pain point was significant enough to justify the additional cost and complexity.
The One-Way Street
What’s notable is that this multi-vendor behavior is essentially one-directional. While half of Claude customers also use ChatGPT, only 6.5% of ChatGPT customers have added Claude. Organizations that choose ChatGPT typically see no need for alternatives – 91% use it exclusively. But organizations that select Claude – despite it being the less popular choice frequently maintain ChatGPT alongside it.
This asymmetry suggests Claude attracts a specific buyer profile: technically sophisticated organizations with complex requirements who view multi-vendor AI infrastructure as strategic, whether planned from the start or added over time to address ChatGPT’s limitations.
5. Silicon Valley favors Claude. NYC and everyone else favor ChatGPT
When we aggregate all Bay Area tech hubs – San Francisco, San Jose, Mountain View, Palo Alto, and surrounding cities – a striking pattern emerges: Silicon Valley companies adopt Claude at nearly double the rate of New York (5.8% vs 3.42%)
The Claude-to-ChatGPT ratio tells an even clearer story. In Silicon Valley, for every four companies using ChatGPT Enterprise, roughly one also deploys Claude. In New York, that ratio drops to one in six. In London, it’s one in seven.
Why does Silicon Valley favor Claude? One is that more software companies often means engineering tasks – code review, debugging, technical documentation. Which is where Claude’s capabilities (esp. with Claude Code) particularly shine. A fintech company in New York might primarily use AI for customer service or compliance documents, where ChatGPT’s broader training and integration ecosystem matters more. A software company in Mountain View is more likely to use AI for helping engineers write and review code, where Claude’s technical performance creates measurable productivity differences.
Conclusion
This study represents the first large-scale observational analysis of enterprise AI adoption based on actual deployment behavior rather than survey responses. The findings reveal we’re still early – somewhere between the 1st and 3rd inning depending on your assumptions about AI’s ultimate penetration. Only 6% of large companies globally and 13.4% of the Fortune 500 have deployed enterprise AI tools to their workforce.
The adoption patterns tell a clear story about where AI is having immediate impact versus where it faces resistance. Marketing, tech, and professional services are moving fast because the use cases are obvious and the risks are manageable. Healthcare, manufacturing, and energy lag not because they can’t benefit from AI, but because they haven’t figured out how to deploy it safely and measure ROI clearly.
Geography matters more than expected. The 7-fold gap between Finland and Portugal within the same EU regulatory framework suggests culture and industrial structure drive adoption as much as regulation.
The competitive dynamics are more nuanced than simple market share numbers suggest. Claude’s asymmetric relationship with ChatGPT – where nearly half of Claude customers also use ChatGPT, but only 6.5% go the other direction – reveals that enterprises are building multi-vendor AI strategies rather than picking winners. This isn’t a “ChatGPT versus Claude” battle; it’s enterprises figuring out which tools work best for which jobs.
The biggest unanswered question: what happens to the 94% of large companies that haven’t adopted yet? Do they follow the early adopters once ROI becomes clearer, or do they represent industries and use cases where enterprise AI tools simply don’t make sense? The answer will determine whether we’re in the 1st inning or the 3rd right now.
