A year ago about 9% of US businesses paid Anthropic. In September 2026 it was 43.8%, ahead of OpenAI's 40.6%. Anthropic also leads in enterprise API spend and in European company-card data, while OpenAI still dominates seats and consumers, and cheap open-weight models are taking a growing share of tokens. Here is the business AI market in 2026, built from four independent data sources, and what it means for companies in Croatia and the DACH region.
The short answer: Anthropic leads where companies pay, OpenAI where people are
The enterprise AI market of 2026 looks very different from the consumer market. In consumer chatbots, ChatGPT still has more than half of all web visits and Claude about a tenth. Among businesses that pay for AI, the order is reversed: more companies pay Anthropic than OpenAI, they spend more per company on it, and Anthropic is estimated to take the largest share of enterprise API dollars. OpenAI remains the largest vendor by business seats and total users, and Google is a strong third through its cloud.
Ramp AI Index: Anthropic overtook OpenAI in April
Ramp, a US corporate-card and spend-management company with more than 70,000 business customers, publishes a monthly AI Index based on what those businesses actually pay for. Its data show one of the fastest shifts in enterprise software: about 9% of businesses paid Anthropic in May 2025, 24.4% in February 2026 and 34.4% in April 2026, when Anthropic passed OpenAI (32.3%) for the first time. In February, Ramp found that Anthropic won about 70% of head-to-head decisions among businesses buying AI for the first time.
Anthropic's lead grew to about four points in the summer (42.4% vs 39.5% in June) and has stalled since. In September 2026 Anthropic added only 0.25 points to 43.8%, while OpenAI gained 1.12 points to 40.6%, faster than overall AI adoption grew. Overall, 55.9% of businesses on Ramp now pay for at least one AI product, up from 47.6% in February.
Two caveats apply. OpenAI points out that its largest enterprise contracts are not paid by card and therefore do not show up in Ramp's data, and Ramp's customers are US companies that use its cards and bill-pay, not the whole economy. Spend is also extremely concentrated: according to Ramp, about 80% of spending on OpenAI and Anthropic comes from the top 1% of firms, which spent around $9,568 per employee per month on AI in September.
Menlo Ventures: 40% of enterprise API spend goes to Anthropic
The most-cited estimate of who earns the API dollars comes from Menlo Ventures, which surveyed 495 US enterprise AI decision-makers in November 2025. It estimates that Anthropic took 40% of enterprise LLM API spend, OpenAI 27% and Google 21%, with Meta, Mistral, Cohere and others sharing the remaining 12%. In 2023 the picture was the opposite: OpenAI 50%, Anthropic 12%, Google 7%. In coding, the largest single use case, Menlo puts Anthropic at 54% against 21% for OpenAI.
Menlo is an Anthropic investor and says so in its report, and a survey is an estimate, not an invoice count. But its direction matches everything else in this article. It also estimated that enterprise spend on generative AI reached $37 billion in 2025, up from $11.5 billion in 2024, and that open-source models fell from 19% to 11% of enterprise usage.
Six measures, one pattern
Put the sources side by side and a consistent pattern appears. Anthropic leads on every measure of business spend: paying companies in the US, API dollars, coding and share of AI spend in Europe. OpenAI leads on breadth: more European companies have at least one OpenAI subscription (73.2% vs 68.1% in Moss's data), and in consumer traffic it is still five to six times larger than Claude. The likely explanation is simple: many companies keep a ChatGPT subscription for general office use, and put their serious, high-volume workloads such as coding agents and document automation on Claude.
Customer counts tell a similar story from a different angle. OpenAI reported about 9 million paying business users in February 2026. Anthropic does not report seats; it said in April 2026 that more than 1,000 business customers each spend over $1 million a year, double the number from two months earlier, and its run-rate revenue reached about $65 billion in July, mostly from businesses. More on the revenue side in our analysis of AI lab revenue and valuations.
Europe: Anthropic takes the bigger share of the budget
For European companies the best public data come from Moss, a Berlin-based spend-management platform, which analysed about €302 million of anonymised company-card transactions from January 2024 to May 2026. Among companies paying OpenAI or Anthropic, 73.2% paid OpenAI and 68.1% paid Anthropic, but Anthropic captured 52.8% of the spend against 38.5% for OpenAI. The average company spent €745 a month on Anthropic and €453 on OpenAI.
The gap grows with company size. Small companies spend about the same on both (€450 vs €486 a month), but large companies spend on average €1,604 a month on Anthropic and €602 on OpenAI. Moss also found that AI tools grew from 5.1% to 15.4% of all SaaS spend and that AI spend rose 309% year on year. Medians are much lower than averages (Anthropic: €156 to €298 by size), so a few heavy users pull the averages up.
Tokens are getting cheaper, and open models win volume
Spend share and token share are not the same thing. On OpenRouter, a popular developer platform that routes requests to hundreds of models, the combined token share of Google, OpenAI and Anthropic fell from about 70% in June 2025 to about 30% in June 2026, according to a Bloomberg analysis of OpenRouter data. DeepSeek alone had about 16% of tokens, and Chinese open-weight models held six of the top ten spots. Claude stays near the top of the leaderboard but at a far higher price per token, so its share of dollars is much higher than its share of tokens.
Prices are falling fast. Ramp measures an effective price of $0.68 per million tokens paid by businesses in early September 2026, down 41% from a peak of $1.15 in March. Frontier models (Claude Opus and Fable, OpenAI's top models) still made up 45% to 53% of token volume over the summer. In September Ramp saw token volumes rise while token spend stayed flat. Open-weight adoption among US businesses remains modest: 6.4% of AI-spending companies paid for open models through routing platforms in August.
Google, Mistral and Manus
Google is the strongest number three. Menlo puts Gemini at 21% of enterprise API spend, up from 7% in 2023, much of it bought through Google Cloud and Vertex AI, which card data barely capture. That is why Ramp saw only 4.7% of businesses paying Google directly in February 2026. Gemini also comes bundled with Google Workspace, so many companies use it without a separate line item.
Mistral plays a different game. Ramp reports that less than 1% of US businesses pay Mistral. Its business is European: analysts estimated annual recurring revenue of about $400 million in January 2026, around 60% from Europe, and the €3 billion Series D in September gives it the means to keep building EU-hosted, open-weight models. For workloads that must stay in the EU, it is often on the short list; see our guide to sovereign AI and data residency.
Manus sells autonomous agents rather than model access, so it does not appear in API market-share data. After Meta's takeover was unwound it raised more than $500 million as an independent company in October 2026. For buyers it is an application-layer choice that sits on top of the models discussed here.
| Vendor | Ramp: US businesses paying (Sep 2026) | Menlo: enterprise API spend share (late 2025) | Where it is strongest |
|---|---|---|---|
| Anthropic | 43.8% | 40% | Coding, agents, document workflows, high-spend enterprises |
| OpenAI | 40.6% | 27% | Seats, general office use, consumers, large invoiced contracts |
| 4.7% (Feb 2026) | 21% | Cloud (Vertex AI), Workspace bundle, multimodal | |
| Mistral | <1% | in "others" (12%) | Europe, data residency, open weights |
| Open-weight via routers | 6.4% of AI buyers (Aug 2026) | 11% of usage (open source) | High-volume, cost-sensitive workloads |
What this means for companies in Croatia and the DACH region
- Expect to use two vendors. The data show that companies commonly pay both OpenAI and Anthropic: one for broad office use, one for heavy workloads. Plan licences, data policies and budgets for that reality instead of forcing a single standard. Our AI vendor selection scorecard helps decide which workload goes where.
- Budget by workload, not by seat. European companies spend 60% more per month on Anthropic than on OpenAI because API and coding workloads scale with use, not with headcount. Put limits, monitoring and model routing in place before usage grows.
- Use the price war. Effective token prices fell 41% in six months. Renegotiate commitments, use prompt caching and batch processing, and test cheaper models for simple tasks.
- Look beyond US data. Ramp and Menlo describe the US market. In Europe, data residency, the EU AI Act and language quality matter more, which keeps Mistral and EU-hosted deployments of Claude and GPT through AWS, Google Cloud and Azure relevant.
- Keep your architecture portable. Market leadership changed within a year. A thin abstraction layer lets you switch models when quality, price or terms change.
The bottom line
In 2026 the business AI market got a new leader. By the share of paying businesses, by API spend and by European company-card spend, Anthropic is now ahead of OpenAI, after starting far behind in 2025. OpenAI is narrowing the gap again and remains the biggest by seats and consumers, Google is a solid third through its cloud, and open-weight models are winning on volume while prices fall. For companies in Croatia and the DACH region, the practical lesson is to run more than one model, budget by workload and use the falling prices to their advantage.
Frequently asked questions
Which AI company has the largest enterprise market share in 2026?
It depends on the measure. By paying business customers on Ramp, Anthropic leads with 43.8% of US businesses in September 2026 against 40.6% for OpenAI. By enterprise LLM API spend, Menlo Ventures estimated Anthropic at 40%, OpenAI at 27% and Google at 21% at the end of 2025. By seats and consumer reach, OpenAI is still far ahead, with about 9 million paying business users reported in February 2026.
What is the Ramp AI Index?
The Ramp AI Index is a monthly measure of AI adoption by US businesses, published by Ramp's Economics Lab. It counts the share of businesses on Ramp's card and bill-pay platform (more than 70,000 companies) that pay for AI products such as subscriptions or API tokens. In September 2026, 55.9% of these businesses paid for at least one AI tool.
When did Anthropic overtake OpenAI in business adoption?
In the Ramp AI Index for April 2026, published in May, Anthropic was paid by 34.4% of businesses and OpenAI by 32.3%, the first time Anthropic was ahead. A year earlier about 9% of businesses paid Anthropic. Its lead widened to about four points in the summer; in September 2026 OpenAI narrowed it to 3.2 points.
What is Mistral AI's market share?
Small outside Europe. Ramp reports that less than 1% of US businesses pay Mistral. Mistral's revenue is concentrated in Europe: analysts estimated annual recurring revenue of about $400 million in January 2026, around 60% of it from European customers, and the company raised a €3 billion Series D in September 2026.
Why do AI market share figures disagree?
Because they measure different things. Ramp counts businesses that pay a vendor, Menlo Ventures estimates dollars of API spend from a survey, Moss measures European company-card spend, OpenRouter counts tokens routed through one platform, and Similarweb counts consumer website visits. Large enterprise contracts paid by invoice, cloud marketplace purchases through AWS, Google or Microsoft, and free use are missing from some or all of them.
Sources: Ramp AI Index (Ramp Economics Lab, March, May, June, July, September and October 2026 updates); Menlo Ventures, "2025: The State of Generative AI in the Enterprise" (December 2025) and 2025 Mid-Year LLM Market Update; Moss, European SaaS & AI Spending Report 2026; Bloomberg / OfficeChai analysis of OpenRouter data (June 2026); Anthropic (6 April 2026); Axios (13 May 2026); Sacra.
Choosing between Claude, ChatGPT, Gemini and Mistral?
We help companies in Croatia and the DACH region compare AI vendors on cost, risk and fit, set up multi-model architectures with sensible budgets, and roll them out with clear policies. As a Claude Certified Architect based in Zagreb, we start with your processes, not with market hype.
Talk to an AI consultant