China Tech – Black and White

Hype tells one story. The stock market tells the opposite.

• While the latest Chinese model releases have put the technology media and the luminaries in a tailspin with the degree of the supposed challenge, the Chinese stock market is telling precisely the opposite story.
• Over the last few weeks, China has released two models that, according to the accompanying press releases, challenge the frontier labs and are far more efficient when it comes to running inference.
• As usual, the devil is in the details, as these models are only more efficient when running on foreign hardware and in the case of Kimi K3, the model is so large that compute savings appear to be dwarfed by the memory footprint that is required.
• However, details aside, it is clear that China offers a challenge to the USA when it comes to the quality of the models that it can produce, meaning that AI in China should be booming and the companies that are in line to benefit should be rapidly increasing their valuations.
• However, the opposite is true in the vast majority of cases where hot IPOs have crashed, and the big technology companies remain moribund value traps taken hostage by the domestic economy.
• Against this backdrop, why any Chinese company would want to go public (other than to achieve an exit) is something of a mystery.
• The signs are everywhere and include Alibaba and Baidu, which remain 40% to 50% below their peak despite investing aggressively in AI, SenseTime more than 50% below; Z.AI, which has almost halved in the last month although still well above its list price; MiniMax, which is down 50% in the last 6 months; and so on.
• The exceptions are the silicon chip companies like Cambricon up 235% YoY, Iluvatar up 200% since it listed and Hygon IT up 143% YoY.
• The problem with this is that I don’t think that it is going to be the semiconductor companies that are going to make the real returns from AI because it remains far more efficient to run one’s model in a data centre in Thailand or Malaysia on non-Chinese hardware.
• This means that providers of AI services will be more competitive using overseas data centres than they will be using domestic data centres without a sustainably large subsidisation program from the Chinese state.
• Hence, in a competitive situation, no one is going to want to use data centres in China to run inference on their AI models to provide their services.
• Take this in the context of domestic consumption that refuses to pick up, and a state that is cutting spending as opposed to trying to boost the economy, and the outlook is bleak indeed.
• China completely dominates the market for open source software, but how the model creators will benefit from foreign start-ups downloading and running their models on their own hardware is unknown.
• The other problem is that investing in China is much riskier now than it was 5 years ago, as the spectre of some sort of government action permanently haunts Chinese companies.
• The net result is that the majority of capital does not want to invest in China despite the indubitable fact that almost all of the sector that is supposed to challenge the might of the USA is trading on bargain-basement valuations.
• This is an unusual state of affairs and makes one wonder whether the Chinese sector is capable of mounting a challenge when it comes to the widespread roll out, adoption and monetisation of AI services in companies and with consumers.
• I do not doubt China’s technological prowess in terms of producing AI technology, but the stock market is telling us that it may not be China that earns a return from the fruits of its labours.
• I have been waiting for Chinese technology to mount a recovery for 3 years now and I am beginning to wonder if it ever will.