Key Highlights

• Open AI models accounted for 58% of usage on one major platform last month, up from just 10% a year earlier.

• AT&T's use of open models rose from 20% in May to 40% today and may reach 60% in coming months, with the company reporting cost savings of up to 80%.

• Nvidia (NASDAQ:NVDA) announced its USD 12.9 billion Acquisition of Hugging Face, a library of open AI models, underscoring the shift's commercial significance.

• Many leading open models now come from Chinese companies, with some reaching 80% to 90% of the performance of top American closed models at a fraction of the cost.

A Major Deal Underscores the Shift

The shift's significance was underscored on Thursday when chipmaker Nvidia (NASDAQ:NVDA) announced it was buying Hugging Face, a library of open AI models, for USD 12.9 billion. Nvidia's chief executive Jensen Huang said the acquisition would help make AI "more open, more capable and more accessible to people and institutions around the world."

Open models come in two varieties: open-source models, which make their underlying code public, and open-weights models, which make public the numerical calculations, known as weights, that determine how a model reasons. Many of the most popular open models come from Chinese companies including Moonshot AI, DeepSeek and Alibaba. Rather than paying subscription fees to access closed models, developers can download a Chinese open model and build products on top of it, paying mainly for the servers needed to run it. Some Chinese open models now reach 80% to 90% of the performance of leading closed models from OpenAI and Anthropic while costing as little as a fifth as much, according to Jerry Tang, chief executive of Atlas Cloud, a startup providing access to various AI models. He compared the top closed models to expensive sports cars, powerful but not always necessary, arguing open models "offer a better bang for your buck."

Implications for Anthropic and OpenAI

The rise of open models could complicate matters for Anthropic and OpenAI, both of which are working toward eventual initial public offerings and need to charge premium prices to cover the billions they spend on research and computing power. After OpenAI released ChatGPT in 2022, many companies gravitated toward closed models, with open-source alternatives remaining mainly popular among individual developers. That began shifting last year after the Chinese startup DeepSeek released a model called DeepSeek-V3, which proved nearly as capable as leading closed models while requiring substantially less computing power. Open models from Chinese companies have continued closing the performance gap with closed offerings from Anthropic, OpenAI and Google (NASDAQ:GOOGL) since then.

Many open models can now run on a phone or laptop, according to David Stout, chief executive of WebAI, an AI startup, making the technology increasingly accessible even as leading closed models grow more complex and computationally demanding. Most American companies still rely on a combination of open and closed models, with closed systems generally remaining preferable for demanding tasks like advanced coding and image or video generation, while open models tend to excel at simpler, specialized tasks.

Security Concerns Persist Alongside Growing Adoption

Some American firms remain wary of using Chinese AI models specifically over regulatory and data privacy concerns. AT&T researches Chinese models but does not use them, working instead with alternatives from American companies including Google's Gemma model and Meta's (NASDAQ:META) Llama models. As Chinese models grow more popular, some American firms are releasing open models of their own. Meta made its leading AI system an open-weights model last month, offering it at a lower cost than other top American models, with chief executive Mark Zuckerberg arguing that restricting access to foreign open-source models is not an effective solution and that American open-source models should instead aim to be the best in the world.

Conclusion

The rapid shift toward open models reflects a straightforward calculation by corporate AI buyers that cost savings of up to 80% often outweigh the marginal performance advantages closed systems still hold for the most demanding tasks. Whether Anthropic and OpenAI can sustain their premium pricing as open alternatives continue closing the performance gap will likely become a significant Factor shaping both companies' prospects as they approach anticipated public offerings.