Corporate America is increasingly turning to open-source AI models as a cost-effective alternative to expensive proprietary solutions. AT&T, for example, saw its AI use of open models jump from 20% in May to 40% currently, with projections to reach 60% soon, saving up to 80% on AI costs. Similarly, companies like Airbnb and Deloitte are adopting these models, which allow for free download and modification, making them easier to customize.
The shift is driven by the soaring costs associated with closed AI models from providers like OpenAI and Anthropic, which have moved to usage-based pricing. Some companies, such as Uber, have even reported exhausting their entire 2026 AI budget by April, leading to spending caps like Uber's $1,500 per month per employee for AI coding tools. A KPMG survey indicated that nearly half of global business leaders scaled back AI use because costs outweighed benefits.
The popularity of open models is evident, with data from OpenRouter showing that they accounted for 58% of U.S. AI use last month, a significant increase from 10% a year ago. These models, often from Chinese developers like Moonshot AI and Alibaba, are estimated to be 80% to 90% as powerful as leading closed models but cost as little as 20% of the price. Even chipmaker Nvidia recently acquired Hugging Face, a prominent library of open AI models, for $12.9 billion, underscoring the growing importance of this technology.
While most U.S. companies still use a hybrid approach, combining open and closed models, the trend towards open-source is seen as a way to avoid vendor lock-in and manage costs. Although closed models excel in complex tasks like coding and image generation, open models are proving effective for specialized, simpler tasks. Companies like AT&T are customizing open-weights models, such as Google's Gemma and Meta's Llama, for specific applications like call transcription and customer service, while still being cautious about using Chinese models due to data privacy concerns.
