The real AI race may no longer be at the frontier

For several weeks this summer, the focus was on the artificial intelligence industry The latest frontier models from Anthropic and Washington struggles for control And who allowed him access to them. But while everyone was watching the border, developers kept building — and they weren’t waiting for permission from the Anthropics and OpenAIs of the world.

Represents Chinese models with open weight 41% of downloads On Hugging Face this spring, surpassing American supermodels. on OpenRouterThe top six most popular models are all open models from Chinese companies including Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. Anthropic’s Claude Opus 4.7 is ranked seventh at the time of this writing. and Data from Vercell It shows Open weight models It accommodates much of the high-volume infrastructure for AI applications, while closed models serve as a high-cost premium layer. Open forms handled nearly a third of AI requests on the platform in June.

These platforms capture only one slice of the AI ​​ecosystem; In particular, they ignore sessions hosted by major labs, which likely account for the bulk of OpenAI and Anthropic’s use. But the large and growing share of the market enjoyed by open source models raises a difficult question: How important will frontier models remain if most AI production ends up working on cheaper, customizable alternatives?

Some see the growth of open source models as a sign that smarter models may end up being used for only the most specialized use cases. “Maybe within a few years, the leading models will be for experimentation and some really high-value tasks, and most production workloads will actually be run either by proprietary models within companies or by open source models,” Clem Delange, CEO of Hugging Face, said in a recent report. Ring of stock.

Hugging Face is a developer platform and community known for hosting, sharing, and helping companies publish open models. Delangue says Hugging Face customers and community members are increasingly touting the benefits of owning their own AI models rather than renting them, a trend that gained traction in the cold light of day after the AI-related bill got passed. The cost of scaling closed-boundary models.

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

This shift is reflected in the activity happening at Hugging Face, Delango says. A new repository is created every seven seconds on the platform, which hosts nearly three million public models and one million public datasets, per Delangue. This suggests a different picture than “the one model that rules them all,” he says. In fact, it’s more like companies are using many different models, many of which are customized to their specific use case. He says half of the Fortune 500 companies use Hugging Face to deploy their own and open source models.

The growing popularity of open models coincides with a steady stream of increasingly capable versions 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, undermining the economics of proprietary AI in which American companies have invested billions. Recently, Beijing-based AI company Z.ai released an open-weight model called GLM-5.2 that excels at proxy encryption and competes with Anthropic’s latest models at identifying vulnerabilities.

Delango isn’t the only executive who argues that companies should avoid tying themselves to a single model provider.

Microsoft CEO Satya Nadella Recently warned Against restricting a single provider, arguing that data control should be a primary concern for organizations using AI.

“While there is a need for great innovation to come from model providers with fair use rights to train models on public data, I find it ironic that the status quo would then turn around and impose restrictive conditions on distillation, reserving the right to learn from customer usage and interaction data,” Nadella said. “If learning flows in only one direction, the economic value converges toward the owners of the learning infrastructure rather than the creators of the knowledge itself. Therefore, it is essential that we distribute the learning infrastructure to each company so that it can control its own learning loop.”

The emergence of open models has also intensified the debate about whether models with increased capabilities should be widely available at all.

Anthropic CEO Dario Amodei Argue Extending the weights of powerful open models can become dangerous because once they are released, they become difficult to control. Others They argued that open models are more accessible to bad actors who could use them to spread disinformation or enact cyber or biological warfare.

Delangue sees the trade-off differently.

“The biggest danger in AI is the concentration of power,” DeLange said. “The way to make the world safer, in my opinion, is by leveling the playing field and creating transparency in these models.”

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

The CEO of Hugging Face argues that keeping powerful models locked down doesn’t eliminate the risks associated with advanced AI systems, in part because it’s easy to bypass and overcome typical frontier API guardrails. Stealing weights And publish it publicly. DeLange argues that restricting powerful models simply concentrates technology in the hands of a few companies while reducing transparency in how systems work.

“You’re not really making it safe by keeping it behind closed doors for a few players,” DeLange said. “You make it more dangerous because you create an asymmetry of power and an asymmetry of capabilities.”