Very interesting news have arrived from China this month that might impact the future of AI and AI companies in the West.

OpenAI beats everyone… until they don’t

If you are OpenAI, raising money, you need to explain why another company can’t do what you are doing, and get you out of business in 6 months.

In 2022, the explanation was: We are the Intel of LLMs. We are so good that no competitor can catch up with us. For the first 2 years after OpenAI released ChatGPT, igniting the ongoing AI boom, they were dominating the market. Between 2022-2024, truly nobody had a model as good as OpenAI had. This was OpenAI’s moat, their competitive advantage.

But unlike in lithography, in the LLM world, the know-hows are few. For you and me, LLMs can be too complex to understand. But a capable AI scientist could have always easily built a model that would be on par with the frontier. And OpenAI even used to publish details of their GPT LLM, making this job easier. We got a market proof of this in March 2024, when Anthropic released Claude 3 Opus. For the first time, a competitor lab caught up with the market leader.

Silicon Valley models beat everyone by a mile… until they don’t

This shook the market and the investors. They needed a new explanation of why investing in OpenAI - or Anthropic - is a good idea. And so it became: We are so good that very few competitors can catch up with us. Other teams are lagging years behind us. It was reasonable logic. Between 2024-2025, there were two companies, OpenAI and Anthropic, that had driven the competition. Google, Meta, and other well-established leaders were suddenly trailing by a huge margin.

But remember: LLMs are not that hard. And the AI boom made it very attractive to invest in solving the hard problem of AI. While engineers in the US (population ~340 million people) were making newer and better models, so did engineers in China (population ~1400 million people). Racism is not a good predictor of frontier AI capabilities, so it shouldn’t come as a surprise that when you have 4x people working on the same problem in China, they occasionally catch up with the West.

In January 2025, we got market proof of this - the so-called “DeepSeek Moment”. Chinese company DeepSeek had released a DeepSeek-R1 model. It was only marginally weaker (several months behind) the latest Western models at the time, and it was 10x cheaper to train. The markets exploded, making it one of the largest dips on chipmakers’ stock in one day¹.

Silicon Valley models beat everyone somewhat… right?

What was the new explanation for investors after the DeepSeek Moment? There wasn’t any. Because since January 2025, the gap between the frontier labs in the West and in China was measured in months. And what is worse (for the Western AI labs), this gap has only been shrinking during the last 1.5 years, and it brings us to today.

Opus 4.8 and Fable 5, the two latest most powerful agentic coding AI models, have been only 2- and 1-month-old, respectively, in July 2026, when two Chinese AI companies released GLM-5.2 and Kimi K3 models. GLM-5.2 reached the performance of Opus 4.8 / GPT 5.5. Kimi K3 reached the performance of Fable 5 / GPT-5.6.

China was a year behind the West in 2024, 6 months in 2025, and 1 month now. At this pace, Chinese companies will beat OpenAI and Anthropic by the end of the year, and release a model that beats any Western model. This might deal a huge blow to Western companies’ valuation.

To add insult to the injury, unlike their US competitors, the Chinese models are open weights. This means that you or anyone in the world (with a very powerful GPU rack) can download them and run them locally. Nobody has any control over anybody using this model, and nobody gets the cut. This is much more devastating for AI labs revenue. I can imagine Western CEOs saying “We can’t use Chinese models on Chinese servers, they may steal our secrets”, and sticking to Fable 11 even though Kimi K7 is better by a full generation. But if the same CEO can run the same model on their own server, the country of the model’s origin won’t matter anymore, yet their R&D budget would.

Tomorrow

Moonshot AI is scheduled to release the weights for Kimi K3, the would-be world-leading open weight model, tomorrow.

This alone has already started the fires and divided Silicon Valley. Chipmakers and infrastructure providers make money when people use models. The competition fueled by open weight models makes models cheaper and accessible to more people, so it benefits them. The leading AI companies make money only when people use their models, and thus any open weight model is a blow to them. Which is likely why the former have signed an open letter to protect the usage of open weight models, while the latter (initially) have not (link).

I’m going to make a few predictions here, what happens next:

• Is the fear of the Western AI frontier labs justified? Absolutely. As we discussed, their valuation comes from the belief that they have a secret sauce that nobody else has. K3 shows that this is just blatantly wrong.
• Is it realistic that Western AI frontier labs will lose 80% of their revenue to Kimi K3? Erh, I don’t think so. The business world is a complex matter. Better business solutions can beat better technical solutions, and US companies have a home-turf advantage.
• Can a multi-billion-dollar market of Western companies consuming Chinese models appear? Absolutely.
• Is it realistic that Western AI frontier labs will enter an insane crunch period to compete with Chinese companies? Yes, and that would be awesome because competition benefits the consumer.
• Is it realistic that Western AI frontier labs will try to use political pressure to kick the better competitor off the market? Yes, this happened before with TikTok that was no worse than Facebook/YouTube/etc., yet it was banned. But I don’t know if it will work. This time may be different, because US businesses (using AI) can push back politically against such developments much stronger than US teenagers (using TikTok).

Let’s see which future comes true. The disrupting world of AI frontier labs is now getting its own medicine and is being disrupted itself. I will be watching the market reaction closely.