China Leads the Open AI Model Ecosystem, Raising Concerns About Global Concentration

“Most of the open models operating around the world are made in China,” Krikorian said. He compared China’s approach to the strategy used by US companies with Android: Give the technology away for free while retaining influence over the ecosystem built around it.

The Risk of Concentration in Open AI

Krikorian said he is less concerned about the “China model” itself than about the risk of concentrating AI development in too few hands. At present, the strongest open AI models are concentrated in China, while US companies offer the strongest closed frontier models.

He urged US and European research institutes to compete in the same open AI market so that no single country can establish the global default.

“The unpleasant truth is that currently multiple ecosystems around promiscuous AI are primarily funded by Chinese capital,” Krikorian said. “It is both plurality and concentration.”

What Open-Source Software Can Teach AI Developers

The history of open-source software, including Linux, could provide lessons for building alternative coalitions around AI, Krikorian said. Open-source software infrastructure was funded by a coalition of organizations, including the Neutral Foundation, “each of which required a commodity layer to exist.”

He said a coalition focused on developing more open AI models would likely need to include “institutions with a mission, not a market,” rather than relying primarily on cutting-edge AI labs.

We need a public computing program that funds a fully open reference model. Apertus, the Swiss state-owned computing company, is one example. The Foundation must maintain the same neutral position it held toward the Internet’s open protocols, which extend to model, harness, and agent standards. Companies that benefit from the commodity model need to participate. And philanthropy needs to cover things that no other field can, especially evaluation and audit infrastructure.

Calls for More Transparency in Open-Weight AI

Krikorian also called on the Alternatives Coalition to make open-weight AI models more open-source. Without knowing how a model was trained or what data and evaluations were used to assess it, he argued, it is difficult to fully trust the model to make decisions.

Greater transparency, independent evaluation, and shared audit infrastructure could help prevent any single country or ecosystem from setting the default for global AI development.