For weeks this summer, the AI industry has been focused on Anthropic’s latest frontier model and Washington’s battle to control who has access to it. But while everyone was focused on the frontier, developers kept building. And they weren’t waiting for permission from the Anthropics and OpenAIs of the world.
This spring, Chinese open weight models accounted for 41% of Hugging Face’s downloads, surpassing U.S. models. In OpenRouter, the top six most popular models are all open models from Chinese companies such as Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. As of this writing, Anthropic’s Claude Opus 4.7 follows in seventh place. Vercel’s data also shows that open-weight models absorb much of the high-volume infrastructure for AI apps, while closed models act as a high-cost premium layer. In June, nearly one-third of AI requests on the platform were processed in an open model.
These platforms capture only one slice of the AI ecosystem. In particular, we exclude sessions hosted by major labs, which likely account for the majority of OpenAI and Anthropic usage. However, the continued and significant growth in the share of open source models in the market poses difficult questions. Even if most of your production AI ends up running on cheaper, more customizable alternative models, how important are frontier models?
Some see the growth of open source models as a sign that the most intelligent models may end up being used only for the most specialized use cases. “Probably in a few years, the frontier models will be for experimentation and very high-value tasks, and most of the production workloads will actually be powered by private models within the enterprise or open source models,” Hugging Face CEO Clem DeLang said on a recent episode of Equity.
Hugging Face is a platform and developer community best known for deploying, hosting, sharing, and empowering businesses with open models. According to Delangue, Hugging Face’s customers and community members are increasingly touting the benefits of owning their own AI models rather than renting them, and this trend is gaining momentum as they pick up the bills associated with the costs of scaling a closed frontier model.
“If you’re an AI company or a technology company, you don’t want to outsource your core functionality to another company, a black box API that you don’t control, don’t have any visibility into, and don’t have any ownership over,” DeLang said.
That change, DeLang argues, is reflected in the activity happening at Hugging Face. Delangue said the platform creates a new repository every seven seconds and hosts approximately 3 million public models and 1 million public datasets. That presents a different picture than “one model to rule them all,” he says. In reality, many companies use a variety of models, many of which appear customized for specific use cases. He says half of Fortune 500 companies use Hugging Face to deploy their own private and open source models.
The rise in popularity of open models has coincided with a steady stream of increasingly capable releases from China’s AI research institutes.
Every few months, another Chinese AI company releases a powerful open-weight model that is cheaper to deploy and easier to customize than its privately held competitor, undermining the economics of the proprietary AI that U.S. companies have poured billions into. Most recently, Beijing-based AI company Z.ai released an openweight model called GLM-5.2 that excels in agent coding and competes with Anthropic’s latest model in identifying security vulnerabilities.
DeLang is not the only executive who says companies should avoid being tied to a single model provider.
Microsoft CEO Satya Nadella recently warned against single-provider lock-in, arguing that control of data should be a top concern for companies using AI.
“We need great innovation from model providers who have fair use rights to train models on publicly available data, but I find it ironic that the status quo has reversed to impose restrictive conditions on distillation and reserve the right to learn from customer usage and interaction data,” Nadella said. “When learning flows only in one direction, economic value is concentrated in the owners of the learning infrastructure rather than in the creators of the knowledge itself. Therefore, it is essential to distribute the learning infrastructure to all companies so that each company can control its own learning loop.”
The rise of open models has also intensified the debate about whether models with improved capabilities should be made widely available.
Anthropic CEO Dario Amodei argued that scaling the weights of a strong open model can be dangerous as it becomes difficult to control once released. Some argue that open models are more accessible because they can be exploited to spread disinformation or wage cyber or biological warfare.
DeLange sees the tradeoffs differently.
“The biggest risk in AI is concentration of power,” DeLang said. “In my opinion, the way to make the world safer is to level the playing field and make these models more transparent.”
Transparency means defenders can more easily “patch cybersecurity risks that we already know open source models can exploit,” he said.
Hugging Face executives argue that keeping powerful models private does not eliminate the risks associated with advanced AI systems. One reason for this is that it’s easy to sneak around the guardrails of the Frontier Model API and steal weights and disseminate them openly. DeLang argues that limiting powerful models only concentrates technology in the hands of a few companies and reduces transparency about how the system works.
“Keeping just a few players behind closed doors doesn’t actually make it safer,” DeLang said. “It creates an asymmetry of power and an asymmetry of capability, which increases the risk.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
