The logic was relatively straightforward: frontier AI models were extraordinarily expensive to train and operate, but the companies capable of building the most powerful systems could charge a premium for access to them. Today, that core assumption is being put to the test. Chinese companies are producing increasingly capable open-weight models at dramatically lower costs, while U.S. technology companies are stepping up their own efforts to develop alternatives.
The shift could make AI cheaper and more widely accessible, potentially accelerating adoption. But it could also undermine one of the most important pillars supporting the valuations of America's leading private AI companies just as they approach the public markets. Anthropic has already confidentially filed for an IPO, while OpenAI has also filed confidentially for a U.S. listing. reported that OpenAI could target a valuation of up to $1 trillion, while Anthropic was valued at $965 billion in its latest private funding round.
More recent investor expectations have pushed Anthropic's potential IPO valuation even higher. For investors, therefore, the open-model revolution is not simply a technology story. It is a question of whether the economic moat around frontier AI can survive falling model prices and increasingly credible alternatives.
The biggest strategic change is that the AI race is no longer being defined solely by who can build the most powerful closed model. Increasingly, the competition is also about who can distribute capable AI most cheaply and efficiently. Chinese companies have made significant progress on this front.
Moonshot's Kimi K3, for example, is a 2.8 trillion-parameter open-weight model that The Wall Street Journal reported was approaching the performance of leading U.S. systems. Z.ai's GLM-5.2 has similarly gained attention for coding and agentic capabilities while being offered at a fraction of the cost of leading U.S. models. This matters because open-weight models change the economics of AI.
Rather than paying a provider every time a model processes information, businesses can download the model weights, deploy them on their own infrastructure and customize them for specific applications. The result can be lower inference costs, greater control over sensitive data and less dependence on a single AI provider. That is increasingly attractive to businesses confronting rapidly rising AI bills.
Extract — continue reading at the source.
