AI Race Shifts Focus From Frontier to Broader Applications

For weeks this summer, the AI sector focused on Anthropic’s latest frontier models and Washington’s efforts to regulate access. Yet, while attention remained on the leading edge, developers continued building without waiting for permission from major labs like Anthropic or OpenAI.

Chinese open-weight models captured 41% of Hugging Face downloads this spring, outpacing U.S. competitors. On OpenRouter, the top six models are all open-source offerings from Chinese firms, including Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai. Anthropic’s Claude Opus 4.7 ranked seventh at the time of writing. Vercel data further shows open-weight models handling much of the high-volume AI infrastructure, while closed models serve as the premium, higher-cost layer. Open models processed nearly a third of AI requests on the platform in June.

These platforms represent only part of the AI ecosystem, excluding sessions hosted by major labs that likely drive most OpenAI and Anthropic usage. However, the growing market share of open-source models raises a critical question: How relevant are frontier models if most production AI runs on cheaper, customizable alternatives?

Some experts view the rise of open-source models as a signal that the most advanced models will eventually serve only specialized tasks. “In a few years, frontier models may be reserved for experimentation and high-value tasks, while most production workloads will rely on private or open-source models,” Hugging Face CEO Clem Delangue stated on a recent Equity podcast.

Loading the player…

Hugging Face is a platform and developer community renowned for hosting, sharing, and assisting companies in deploying open models. Delangue notes that customers increasingly emphasize the benefits of owning AI models rather than renting them, a trend accelerated by the high costs of scaling closed frontier models.

“If you are an AI or technology company, you should not outsource core capabilities to another entity via a black-box API that you cannot control, monitor, or own,” Delangue explained.

Delangue argues this shift is evident in Hugging Face’s activity. A new repository is created every seven seconds on the platform, which hosts nearly three million public models and one million public datasets. This contradicts the “one model to rule them all” narrative, suggesting instead that companies use diverse models customized for specific use cases. Half of the Fortune 500 firms use Hugging Face to deploy private and open-source models, he says.

The increasing popularity of open models coincides with frequent releases of capable models from Chinese AI labs.

Every few months, a Chinese AI company releases a powerful open-weight model that is cheaper to deploy and easier to customize than proprietary competitors, challenging the economics of U.S. firms’ billions in investment. Recently, Beijing-based Z.ai released GLM-5.2, an open-weight model excelling in agentic coding and competing with Anthropic’s latest models in security vulnerability detection.

Delangue is not the only executive advising enterprises to avoid reliance on a single model provider.

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

“While fair use rights for model providers to train on public data drive innovation, it is ironic that restrictive terms are then imposed on distillation, and the right to learn from customer data is reserved,” Nadella stated. “If learning flows in only one direction, economic value concentrates among infrastructure owners rather than knowledge creators. Therefore, distributing learning infrastructure to every firm is imperative so they can control their own learning loop.”

The rise of open models has also intensified debates about whether increasingly capable models should be widely available.

Anthropic CEO Dario Amodei argues that scaling powerful open model weights could be dangerous, as released models become difficult to control. Others contend that open models are more accessible to bad actors who might use them for disinformation, cyberattacks, or biological warfare.

Delangue views the tradeoff differently.

“The greatest risk in AI is the concentration of power,” Delangue said. “Making the world safer, in my view, involves leveling the playing field and increasing model transparency.”

Transparency allows defenders to “patch cybersecurity risks that open-source models can exploit,” he explained.

The Hugging Face executive argues that keeping powerful models closed does not eliminate advanced AI risks, partly because frontier model API guardrails are easy to bypass and weights can be stolen and disseminated openly. Restricting powerful models, Delangue argues, simply concentrates technology in the hands of a few companies while reducing transparency into system operations.

“Keeping models behind closed doors for a few players does not make them safe,” Delangue said. “It makes them more dangerous by creating asymmetries of power and capabilities.”