Changing of the guard: Anthropic new AI market leader among US companies
Large US companies and start-ups are increasingly relying on AI models from Anthropic. Open source models, on the other hand, are losing popularity.
Companies more than doubled their spending on large language models (LLMs) in the first half of 2025 compared to the previous year – from 3.5 billion US dollars at the end of 2024 to 8.4 billion dollars at the end of the year. This is according to a recent report by Menlo Ventures; the figures relate to larger US companies and financially successful AI start-ups. AI products for end users are explicitly not included.
Anthropic is now the market leader
In the USA at least, the market leadership has also shifted: Anthropic is now the frontrunner in the enterprise use of LLMs with a market share of 32 percent. The loser is OpenAI, which had a share of 50 percent two years ago and has now fallen back to 25 percent. Google follows in third place with 20 percent. Meta still has 9 percent and DeepSeek only 1 percent.
Menlo Ventures dates Anthropic's rise to June 2024, when the provider released the Claude 3.5 Sonnet model. The follow-up update Claude 3.7 Sonnet in February 2025 further accelerated the increasing use by companies. The area of code generation stands out in particular, as Anthropic has clearly established itself here with a market share of 42%, ahead of OpenAI (21%). The market researchers therefore describe code generation as the first "AI killer app".
Users are now increasingly turning to the practical use of AI: while the training of models dominated until recently, this has now replaced AI inference –, i.e. the application of the AI model to new data –. In the survey, 74% of start-ups and 49% of companies stated that the majority of their AI workload consists of inference.
Videos by heise
No breakthrough for open source
In contrast, the preference for closed models – has not changed and the use of open source models has even decreased slightly according to the survey: at the end of 2024, they still had a share of 19%, but now only 13%. Meta's Llama model remains the most popular here. Menlo Ventures emphasizes the advantages of open models – Customization by users, cheaper and local use –, but the differences in performance weigh more heavily.
It is therefore not surprising that the performance of the models is the central argument for 61% for switching providers. Price, on the other hand, plays a subordinate role (36%), while security and compliance are only the trigger for a change for 22%. However, only 11% changed providers in the last twelve months, with two thirds switching to newer models. 23 percent stayed with the same model.
Menlo Ventures emphasizes that the market remains extremely dynamic – it changes on a weekly basis. It is therefore impossible to predict the future of AI. Details of the survey can be found here.
(fo)
