The artificial intelligence industry spent several weeks this summer intently focused on Anthropic's cutting-edge frontier models and the ongoing debate in Washington regarding who would be granted access to them. Yet, while collective attention gravitated towards these advanced developments, developers worldwide continued their work, unconstrained by the permissions or timelines of industry giants like Anthropic and OpenAI.
A significant shift in the AI landscape became apparent this spring, with Chinese open-weight models accounting for 41% of downloads on Hugging Face, thereby surpassing U.S. models. On OpenRouter, the top six most frequently used models are all open offerings from Chinese firms, including Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai, with Anthropic’s Claude Opus 4.7 ranking seventh at the time of this report. Furthermore, data from Vercel illustrates that open-weight models are increasingly handling the high-volume infrastructure demands of AI applications, while closed models are positioned as the higher-cost, premium tier. In June, open models processed nearly a third of all AI requests on the Vercel platform.
It is important to note that these platforms capture only a segment of the broader AI ecosystem; specifically, they do not include sessions hosted by major labs, which likely represent the bulk of usage for companies like OpenAI and Anthropic. However, the substantial and expanding market share of open-source models prompts a critical question: how much relevance do frontier models retain if the majority of AI in production ultimately relies on more affordable, customizable alternatives?
Some interpret the growth of open-source models as an indication that the most intelligent models might eventually be reserved for highly specialized use cases. Clem Delangue, CEO of Hugging Face, articulated this perspective on a recent episode of "Equity," stating, “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, a prominent platform and developer community, is well-regarded for hosting, sharing, and assisting companies in deploying open models. Delangue observes that Hugging Face’s customers and community members are increasingly advocating for the advantages of owning their AI models rather than merely renting them. This trend has gained considerable momentum, particularly as organizations confront the substantial costs associated with scaling closed frontier models.
Delangue emphasized the strategic imperative for companies to maintain control over their core capabilities: “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.”
This evolving preference, Delangue argues, is clearly reflected in the activity occurring on Hugging Face. A new repository is created every seven seconds on the platform, which currently hosts almost three million public models and one million public datasets, according to Delangue. He suggests this paints a different picture than the narrative of "one model to rule them all," instead indicating that companies are utilizing a diverse array of models, many of which are tailored to their specific use cases. He further noted that half of all Fortune 500 firms are leveraging Hugging Face to deploy their own private and open-source models.
The increasing popularity of open models coincides with a consistent stream of increasingly capable releases from Chinese AI laboratories.
Every few months, another Chinese AI company introduces a powerful open-weight model that offers more cost-effective deployment and greater ease of customization compared to closed competitors. This effectively challenges the economic framework of proprietary AI, into which U.S. firms have invested billions. Most recently, Beijing-based AI company Z.ai launched GLM-5.2, an open-weight model that excels in agentic coding and demonstrates competitive performance against Anthropic’s latest models in identifying security vulnerabilities.
Delangue is not alone among executives in advising enterprises to avoid becoming overly reliant on a single model provider.
Microsoft CEO Satya Nadella recently cautioned against single provider lock-in, asserting that control over data should be a paramount consideration for enterprises employing AI.
Nadella articulated his concern, stating, “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.” He further elaborated on the economic implications: “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 emergence of open models has also intensified a broader discussion regarding whether increasingly capable AI models should be made widely available at all.
Anthropic CEO Dario Amodei has argued that scaling powerful open model weights could become dangerous because, once released, they become exceptionally difficult to control. Others have similarly contended that open models are more easily accessible to malicious actors who could potentially utilize them to disseminate disinformation or to engage in cyber or biological warfare.
Delangue, however, perceives this trade-off from a different vantage point.
“The biggest risk in AI is concentration of power,” Delangue asserted. He continued, “The way you make the world safer, in my opinion, is by leveling up the playing fields and creating transparency on these models.”
He explained that transparency allows defenders to more readily “patch the cybersecurity risks that they already know open source models can exploit.”
The Hugging Face executive contends that keeping powerful models closed does not eliminate the risks associated with advanced AI systems. He points out that it is relatively easy to bypass frontier model API guardrails and to steal model weights for open dissemination. Delangue argues that restricting access to powerful models primarily serves to consolidate the technology in the hands of a few companies, while simultaneously reducing transparency into how these systems operate.
“You don’t really make it safe by keeping it behind closed doors for just a few players,” Delangue stated. “You make it more dangerous because you create asymmetry of power and asymmetry of capabilities.”
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.