Models ModelScope said it hosted (March claim): 170,000 (Source: Rest of World, citing ModelScope's own March statement (Sept 30, 2026))
A model hub is not just a download page. It is infrastructure, and infrastructure has a jurisdiction. A Rest of World report on China's open-model platforms shows what that means in practice.
Chinese regulators blocked the U.S.-based hub Hugging Face in 2023. Rest of World reports that they gave no specific reason. The block opened a market for domestic alternatives. Rest of World names two: Alibaba's ModelScope and OSChina's MoArk.
What Rest of World found
Alibaba started ModelScope in 2022, before the block. In March it gave figures of 170,000 models and 250 million users. OSChina started MoArk in 2023. Its chief executive, Xu Yong, said MoArk hosted more than 20,000 commonly used models. Hugging Face hosts more than 3 million open models, Rest of World reports.
One caution on the numbers. The two Chinese figures are the platforms' own statements. They may not be counted the same way as Hugging Face's total. Even so, Rest of World says the Chinese hubs "still fall far behind."
Both Chinese platforms offer model testing and customization. Some use is free. Extra computing power and advanced features cost money.
How an open-model hub works
Closed models from OpenAI and Anthropic are reached only through the companies that run them. Open models are different. Rest of World notes they can be downloaded, fine-tuned and run on local hardware.
A hub is where that download happens. It is the shared shelf that developers pull models from and publish models to. Think of it as a package registry for AI.
MoArk adds a second layer. Xu says a team of its engineers adapts models to run on different Chinese chips. Rest of World ties this to a computing shortage, since the U.S. restricts sales of top AI chips to China. Xu called chip compatibility for mainstream open models the platform's mission.
So a hub can decide more than where files are stored. It can shape which hardware a model runs on.
Why the source of a model is a dependency
The lesson here is that where your AI comes from is now a risk decision, not only a technical one. Two pressures show up in the report.
The first is access. Rebecca Arcesati of the Mercator Institute for China Studies said there is a real concern in China that access to U.S.-based platforms "could be disrupted at any point." The second runs the other way. Rest of World reports that the Trump administration has reportedly discussed bans on Chinese models.
Nvidia announced in September that it is acquiring Hugging Face for $12.9 billion. In its filing about the purchase, Nvidia flagged a possible ban on Chinese models appearing on the platform.
These are stated concerns and reported discussions. The report does not say any such ban exists. But it shows that both sides treat model access as something that could be cut.
The operator and the developer
The case for a domestic hub comes from the people who run one. "Not everyone is able to use a VPN all the time," said Xu, whose company operates MoArk. He argues that China needs a self-reliant AI ecosystem for Chinese-speaking users. That is a platform operator speaking, with a stake in the outcome.
Developers offer a more mixed picture. Rest of World reports that they see quicker downloads as the main draw of domestic platforms. Chen Yunfei, an independent developer in Kunming, still favors Hugging Face over a VPN because it offers more variety. "Domestic platforms need to convince us why we should use them," he said.
Rest of World also reports that Beijing tolerates VPN workarounds. Total isolation would starve its AI industry of global connections. Chinese AI companies also publish their own models on Hugging Face for use worldwide.
Questions for your team
This story is about China, but the questions apply to any company that builds on open models.
Ask which hubs your teams pull models from today. Ask whether anyone records the source, version and licence of each model in production. Ask what would break if a hub became unavailable, or restricted, in a market where you operate.
Ask also which chips your models are tuned for. If a supplier or a jurisdiction changes, is there a second route?
One caveat applies. The report is about developer platforms in China. It gives no growth figures for the Chinese hubs. It does not measure how many companies elsewhere use Chinese models, or how many developers switch to domestic hubs.
The takeaway is simple. Model hubs used to look like neutral shelves. The Rest of World report shows governments treating them as strategic ground, and your supply chain sits on that ground too.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: Rest of World.
