Why the Real AI Race Has Shifted Beyond the Frontier

For much of the summer, the AI conversation was dominated by Anthropic’s latest frontier models and the political tug-of-war in Washington over who could access them. But while the spotlight stayed fixed on cutting-edge capabilities, developers were quietly building at scale, and they weren’t waiting for permission from the industry’s biggest names.

Chinese open-weight models now account for 41% of all downloads on Hugging Face this spring, overtaking U. S.-based alternatives. On OpenRouter, the top six most popular models all come from Chinese firms such as Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. Anthropic’s Claude Opus 4.7 currently sits in seventh place. Data from Vercel reveals that open-weight models are absorbing a growing share of the heavy-lifting infrastructure behind AI applications, while closed models function as a premium, higher-cost tier. In June, open models handled nearly one-third of all AI requests on the platform.

These platforms capture only one slice of the broader AI ecosystem, leaving out the sessions hosted by major labs that likely represent the bulk of usage for companies like OpenAI and Anthropic. Still, the large and growing market share of open-source models raises a fundamental question: How much do frontier models really matter if most production AI ends up running on cheaper, customizable alternatives?

Some observers see the rise of open-source models as a signal that the most intelligent AI systems may ultimately serve only the most specialized tasks. “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,” said Hugging Face CEO Clem Delangue during 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 notes that customers and community members are increasingly touting the advantages of owning their own AI models rather than renting them, a shift that has accelerated as the true cost of scaling closed frontier models becomes clear.

“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 change is reflected in the activity on Hugging Face itself. A new repository is created every seven seconds, and the platform now hosts nearly three million public models and one million public datasets. This suggests a far more fragmented reality than the “one model to rule them all” narrative. In practice, companies are using many different models, many of which are customized for their specific needs. Half of all Fortune 500 firms now use Hugging Face to deploy their own private or open-source models.

The growing popularity of open models coincides with a steady stream of increasingly capable releases from Chinese AI labs. Every few months, another Chinese company unveils a powerful open-weight model that is cheaper to deploy and easier to customize than its closed competitors, undermining the economics of proprietary AI that U. S. firms have invested billions in. Most recently, Beijing-based Z.ai released GLM-5.2, an open-weight model that excels at agentic coding and competes with Anthropic’s latest offerings on identifying security vulnerabilities.

Delangue is not alone in arguing that enterprises should avoid locking themselves into a single model provider. Microsoft CEO Satya Nadella recently warned against single-provider lock-in, emphasizing that data control should be a top priority for businesses 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 about whether increasingly capable AI systems should be broadly available at all. Anthropic CEO Dario Amodei has argued that releasing powerful open-weight models could become dangerous because once they are out, they become difficult to control. Others worry that bad actors could easily access open models to spread disinformation or conduct cyber or biological warfare.

Delangue sees the tradeoff differently. “The biggest risk in AI is concentration of power,” he 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, he argues, allows defenders to more easily patch the cybersecurity risks that open-source models already expose. Keeping powerful models closed, he contends, doesn’t eliminate the risks, especially since it is relatively easy to bypass guardrails on frontier model APIs or steal and disseminate weights openly.

“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.”