On June 13, 2026, a Beijing company named Z.ai released a new open source LLM called GLM 5.2. Within days it was ranked among the best models, and many people called it a new “DeepSeek moment”.

I don’t agree. I think it is the new solar panel, and it is the new electric car.

With solar panels, the West itself created the opening: Western subsidies made panels worth producing, and only then did Beijing decide that solar power was strategic. 15 years ago the Party poured money into solar factories, and cheap Chinese panels flooded the world. With electric cars, the CCP planned everything from the start: subsidies first, then hundreds of funded carmakers (most since collapsed), then a flood of survivors into foreign markets. One could be challenged to find an inexpensive alternative to BYD these days. Now the Party wants to win the AI race, and Z.ai shows how the system works. Artificial intelligence is the third run of the same plan, and as before, nobody asks whether the money will ever come back. That was never how success was measured at a closed session of the Central Committee of the Chinese Communist Party.

Great large language model, lousy business model

This time, however, the economic model is different. Both solar panels and cars earn money for their makers even if they are sold at a loss. An open source model earns nothing: when a data center runs GLM 5.2, the money stays with the data center, and Z.ai gets customers’ love, not customers’ money.

The model is large ~750 billion parameters, and it needs substantial computing resources for pre-training and post-training. Because the American blacklist puts NVIDIA out of reach, Z.ai is using exclusively Huawei GPUs that are much slower and less efficient. We do not have credible information on the actual cost to train GLM 5.2, but obviously no yuan was spared by Z.ai’s investors.

Needless to say, almost none of the investors has business goals in mind. In China there are no private investors in the Western sense. Every large pool of capital there is the government itself, a state-owned enterprise, a state-run university or scientific institute, or a Party-supervised conglomerate like Alibaba. When money moves on a large scale in China, it moves because the Party allows it to move. Z.ai is no exception. On paper it is an ordinary private company, controlled by its founders and listed in Hong Kong, yet almost all the money that built it came from these Party-controlled vehicles. The solar panel factories and the electric car makers were financed the same way: the companies looked private, but the money answered to the Party.

The list of investors removes any doubt. 5 of the largest technology companies in China: Alibaba, Tencent, Xiaomi, Meituan and Ant Group, put money into Z.ai together, although each of them builds its own competing model. Imagine Google, Meta, Amazon and Elon Musk quietly funding OpenAI together. When one company supports a rival, it is an oddity, right? Here a whole industry supports the same rival at the same moment -it is coordination, and in China only one force can coordinate a whole industry: the Communist Party.

The Party does not need to own Z.ai, because it controls the money that owns Z.ai, and it has declared such models a national priority. It also supplies the customers: most of Z.ai's income comes from state-owned enterprises and government bodies, and the same local funds that invested in the company steer those bodies to buy from it. Money, customers, orders — the same sequence China used for computer chips, batteries, electric cars and solar panels.

None of this depends on the company ever making a profit. The same kind of money once paid for the famous empty ghost cities of Ordos, which stood for years without earning a single yuan. The people who spend it answer to the Central Committee of the CCP, not to customers or investors, and the fact that nobody expects the money back is not a flaw in the system. It is the system.

Where will China dominate?

The story of the overwhelming Chinese success in the solar panel industry still sends a chill down the spine of Western politicians. In the car segment we see attempts by Western countries to fight the flood of cheap Chinese EVs, but they need to balance the fate of the car industry with the fate of global trade, and this balance is intricate and tilts towards fewer barriers and more cooperation.

The world of the AI economy is brand new, and Western nations are yet to decide whether Chinese open source AI shall be barred from entering Western markets as their abundance might danger viability of the largest segment of venture investment and regulatory control that nations will be imposing on AI models and AI-driven systems.

I won’t be surprised if US labs are able to prove that Chinese open source models, including GLM 5.2, were trained using responses from American AI models. That constitutes a breach of the user agreement for US-made models, so using Chinese open source models will be treated as “consuming fruit of the poisonous tree” – using the results of illegal activity – and made illegal.

Still, most of the world will remain open for Chinese AI: the developing countries of Africa and Asia. In many of them, telephone networks and internet services already run on Huawei equipment, so a Chinese model is the next step on rails that China has already laid. The leading American models were never priced for these markets, and people there know that access to an American model depends on the goodwill of Washington and can be withdrawn at any time. A model running on a computer in their own country cannot be withdrawn.

The plan is narrow and deliberate: accept the losses at home, give the technology away everywhere, and build a position in the one region whose door is open. In time there may be no non-Chinese model on that shelf at all, just as today there is no non-Chinese solar panel.

Western commentators keep asking when Z.ai will finally make a profit. The question misses the point. A company whose spenders answer to state targets cannot go bankrupt, and a company that cannot go bankrupt does not compete on price; it sets the price.

Solar panels were the accident, electric cars were the rehearsal, and artificial intelligence is the performance now on stage. The West has watched this play twice, and both times it called it a free market. There is no reason to believe the third time will be different.