The real AI race may no longer be at the frontier

For several weeks this summer, the AI industry was fixated on Anthropic’s latest frontier models and Washington’s fight to control who was granted access to them. But while everyone was watching the frontier, developers kept building — and they weren’t waiting around for permission from the Anthropics and OpenAIs of the world.

Chinese open-weight models accounted for 41% of downloads on Hugging Face this spring, surpassing U.S. models. On OpenRouter, the top six most popular models are all open models from Chinese firms, including Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. Anthropic’s Claude Opus 4.7 trails in seventh place, at the time of this writing. And data from Vercel shows that open-weight models are absorbing much of the volume-heavy infrastructure of AI apps, while closed models operate as the higher-cost, premium layer. Open models handled nearly a third of AI requests on the platform in June.

Those platforms only capture one slice of the AI ecosystem; in particular, they leave out sessions hosted by major labs, which likely account for the bulk of OpenAI and Anthropic’s usage. But open source models’ large and growing share of the market raises a difficult question: How much do frontier models still matter if most production AI ends up running on cheaper, customizable alternatives?

Some see the growth of open source models as a sign that the most intelligent models may end up being used for only the most specialized use cases. “Maybe in a few years, the frontier models will be for experimenting and [for] some really high-value tasks, and most of the production workloads will actually be powered either by private models within companies or by open source models,” Hugging Face CEO Clem Delangue said on a recent episode of Equity.

Hugging Face is a platform and developer community best known for hosting, sharing, and helping companies deploy open models. Delangue says Hugging Face’s customers and community members are increasingly touting the benefits of owning their own AI models rather than renting them, a trend that’s picked up steam in the cold light of day after getting the bill associated with the cost of scaling closed frontier models.

“If you’re an AI company or a technology company, you don’t want to outsource your core capabilities to another company, to a black box API that you don’t control, don’t have any visibility on, and don’t really have any sort of ownership,” Delangue said. 

That shift, Delangue argues, is reflected in the activity happening on Hugging Face. A new repository is created every seven seconds on the platform, which hosts almost three million public models and one million public datasets, per Delangue. That points to a different picture than the “one model to rule them all,” he says. In reality, it looks more like companies using many different models, many of which are customized for their specific use case. Half of all Fortune 500 firms are using Hugging Face to deploy their own private models and open source models, he says. 

The growing popularity of open models coincides with a steady stream of increasingly capable releases from Chinese AI labs. 

Every few months, another Chinese AI company releases a powerful open-weight model that is cheaper to deploy and easier to customize than closed competitors, undercutting the economics of proprietary AI that U.S. firms have poured billions into. Most recently, Beijing-based AI company Z.ai released an open-weight model called GLM-5.2 that excels at agentic coding and competes with Anthropic’s latest models on identifying security vulnerabilities. 

Delangue isn’t the only executive arguing that enterprises should avoid tying themselves to a single model provider.

Microsoft CEO Satya Nadella recently warned against single provider lock-in, arguing that control of data should be a primary concern for enterprises using AI. 

“While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation, and to reserve the right to learn from customer usage and interaction data,” Nadella said. “If learning flows in only one direction, economic value converges toward the owners of the learning infrastructure rather than the creators of the knowledge itself. Therefore, it’s imperative that we distribute the learning infrastructure to every firm so that they can control their own learning loop.”

The rise of open models has also intensified a debate over whether increasingly capable models should be broadly available at all. 

Anthropic CEO Dario Amodei has argued that scaling powerful open model weights could become dangerous because once they are released, they become difficult to control. Others have argued that open models are easier to access by bad actors who could use them to spread disinformation or enact cyber or biological warfare. 

Delangue sees the tradeoff differently.

“The biggest risk in AI is concentration of power,” Delangue said. “The way you make the world safer, in my opinion, is by leveling up the playing fields and creating transparency on these models.”

Transparency means defenders can more easily “patch the cybersecurity risks that they already know open source models can exploit,” he said.

The Hugging Face executive argues that keeping powerful models closed doesn’t eliminate the risks associated with advanced AI systems, in part because it’s easy to get past frontier model API guardrails and to steal the weights and disseminate them openly. Restricting powerful models, Delangue argues, simply concentrates the technology in the hands of a few companies while reducing transparency into how systems work. 

“You don’t really make it safe by keeping it behind closed doors for just a few players,” Delangue said. “You make it more dangerous because you create asymmetry of power and asymmetry of capabilities.”