By Adam Levine

This year is when artificial intelligence really began to be adopted by enterprises, and it came with sticker shock. Organizations are still trying to figure out how they can get the most bang for their buck, and they are eyeing Chinese models, which are very good while being much cheaper than the U.S. models that are driving up costs.

Leaving politics out of it, lower AI costs would be broadly good for the economy, especially for the AI infrastructure trade, but the major U.S. model makers like OpenAI, Anthropic, and Alphabet's Google would see their pricing come under pressure.

DeepSeek became the poster child for Chinese models last year, and it is still incredibly inexpensive at only 3% of the price per token of OpenAI's GPT 5.5. Since then, several companies, from Alibaba to Moonshot AI to Zhipu AI, have jockeyed for the lead in Chinese model benchmark rankings.

Currently, the most highly regarded Chinese model is GLM 5.2 from Zhipu. It's a great example, because it shows both the benefits and risks for U.S. enterprises. In Artificial Analysis' multi-benchmark rankings, it comes in fifth behind three models from Anthropic and one from OpenAI, and ahead of Google's Gemini models. If a U.S. organization used the version that Zhipu hosts in its own cloud in China, the cost per token would be just 15% of that of the OpenAI model.

But no U.S. enterprise is about to do that. In the first place, putting company secrets on a Chinese server is a non-starter from a security standpoint. There are also geopolitical concerns. Zhipu and other Chinese AI start-ups are heavily subsidized by local Chinese governments and state-owned enterprises, which is a big reason for their fast advancement and low prices, and they have ties to the Chinese military. Last year, Zhipu was added to the U.S. Department of Commerce Entity List, generally seen as a signal to U.S. companies to stay away.

But GLM is an "open-weight" model, which means that companies can host it on their own servers or private cloud without paying Zhipu. The major U.S. public clouds — Amazon Web Services, Microsoft Azure, and Google Cloud — all offer GLM at a massive discount to the top models from the U.S. companies.

There are differing interpretations of what the financial impact of widespread use of Chinese models would be. Like during the DeepSeek selloff of January 2025, many have seen the possibility of declining prices for AI as a sign that there is overinvestment in data centers, which now totals hundreds of billions of dollars a year.

But the history of drastically reduced prices shows that lower prices can increase demand to such an extent that more money flows to producers than otherwise. One of the first automation technologies was the power loom, which crashed the price of textiles. Much cheaper clothing led to an explosion of demand, as people went from having just one or two outfits to many. Textile mills multiplied, as did employment in the sector.

The equivalent of those textile mills today is the AI infrastructure indsutry — chipmakers, energy providers, cooling companies, and cloud providers. Cheaper AI would increase demand for them, not lessen it.

Lower AI costs would be broadly beneficial to the economy, except for the model makers, who would be forced to lower prices to compete. They could even find themselves out of business.

Google plays both sides of this coin. Google Cloud would be one of the beneficiaries, but Gemini would lose what little pricing power it has.

Write to Adam Levine at adam.levine@barrons.com

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.