The Artificial Analysis AI Adoption Survey Report (H1 2025), released in the first half of 2025, shows that artificial intelligence is accelerating from the experimental stage toward scaled production deployment, with enterprise AI maturity significantly improving. The report is based on feedback from 1,036 AI users globally, covering multiple dimensions including technology, model selection, application scenarios, and infrastructure.
I. AI Application Maturity: 45% of Enterprises Have Deployed AI in Production
The survey shows that 45% of respondent organizations are already using AI in production environments, with another 23% in the prototyping stage and 27% still exploring. This reflects that AI technology has moved beyond proof-of-concept and entered the stage of supporting actual business operations.
In terms of build vs. procurement strategies, enterprises exhibit diverse choices:
32% prefer to build their own AI applications
27% choose to procure third-party solutions
25% adopt a hybrid model (build + procure)
II. Engineering and R&D Become the Primary AI Application Scenario
Among anticipated internal AI application scenarios:
Engineering and R&D (66%) dominate
Customer support and success (37%) and sales and marketing (33%) follow
IT and cybersecurity (29%) and finance and accounting (28%) are also important application areas
III. Language Model Market: Gemini and GPT Run Neck and Neck, DeepSeek Emerges as Top Open-Source Choice
Google Gemini and the OpenAI GPT/o series remain the mainstream market choices, with usage/consideration rates both near 80%.
DeepSeek has become the most popular open-weight model, with a usage/consideration rate of 53%.
xAI Grok's usage rate jumped from 15% in 2024 to 31%, a significant increase.
Notably, the average number of LLMs enterprises use or consider increased from 2.8 in 2024 to 4.7 in 2025, indicating a maturing market and organizations' greater willingness to experiment with multi-model strategies.
IV. Enterprises Show High Acceptance of Chinese Models but Rely on Infrastructure Outside China
55% of respondents indicated they are willing to use LLMs developed by Chinese AI labs, provided the models are deployed on infrastructure outside China. Only 27% are willing to directly use model services located within China.
V. Multimodal Models: OpenAI Leads in Speech, Image, and Video Generation
Speech Generation: OpenAI and ElevenLabs lead, with key selection factors being streaming quality, natural speech, and low latency.
Image Generation: OpenAI dominates, with users prioritizing prompt adherence most highly.
Video Generation: OpenAI and Google lead, with prompt adherence and realism being the most important metrics.
VI. Inference Services: Native APIs Preferred First, Chip Challengers Like Groq and Cerebras Rise
Vendors offering first-party model APIs, such as OpenAI, Google, and Anthropic, dominate the market.
Chip vendors like Groq and Cerebras have seen significant growth in the inference service market.
Amazon and Azure's market shares have declined.
VII. Training Hardware: NVIDIA Absolute Dominance, Followed by Google TPU and AMD
NVIDIA holds a 78% share of the training accelerator market.
Google (27%) and AMD (17%) trail far behind but remain significant competitors.
VIII. Major Challenges: Intelligence Level, Reliability, and Cost
The primary challenges enterprises face in using AI include:
Insufficient intelligence level (55%)
Reliability issues (50%)
Excessive costs (50%)
System integration difficulty (38%)
Conclusion and Outlook
In the first half of 2025, AI technology is moving from proof-of-concept to scaled application. Enterprises are placing greater emphasis on model diversity, cost control, and integration with real business scenarios. Open-source models (such as DeepSeek) and emerging players (such as xAI) are reshaping the market landscape, while competition in multimodal AI and inference infrastructure is also intensifying.
Artificial Analysis provides more detailed industry segmentation, regional data, and advanced insights. The full report is available through subscription to its AI Trends service.
Full Report: