Blog

Chinese AI may be good at agentic tasks, but…

In 2024, OpenAI talked about five levels of AI.

  1. Chatbot/Conversational AI
  2. Reasoners
  3. Agents
  4. Innovators
  5. Organizations

We are at level 3.

Chinese AI could be good or superb in the second and third level.

Chinese people are good at solving problems (think about exams) and doing tasks (think about how efficient Chinese workers can be at repetitive work ).

However, if Chinese AI needs to at level 4 and 5, and if it’s truly “Chinese”, then Chinese people need to evolve.

True innovation is scarce in China. Chinese society currently doesn’t reward true innovation. It’s like “solving news problems with new ways”. That is not the natural motivation. Chinese AI may stuck at level 3 if we don’t do innovation right.

It’s another problem at level 5. Most Chinese people don’t know how to organize themselves.

So it’s hard for me to see how Chinese AI can do this right either.

They may struggle for a while.

Or they might become a different animal.

The memory shock to investing

The explosive rise in memory prices has profound impact.

One is in investing and valuation framework – you now have several memory names (like SK Hynix, Micron, SanDisk) that is trading at single-digit or 10x current year P/E with expected high earning growth into the future.

And these companies are large enough to absorb tons of capital.

The indirect impact? Other companies’ valuation now looks absurd!

The previously dirty cheap companies lose investors. For 10x P/E, you can buy memory names with much stronger growth!

If the only reason to long is undervalued, then it becomes harder to justify the buy.

Additionally, those companies with reasonable moat, single digit growth, 20-30x p/e also lose – 2-3x PEG looks absurd now.

This hits many consumer or healthcare names.

No delta, no acceleration in growth = no interest.

Focus on OPEX

While the last few years AI investments are correlated with companies’ capex, I think the next stage is to focus on opex.

As customers of LLMs and agents, how many tokens are consumed and what is the cost of that? That will show up in operating expenses.

This will be supportive evidence of strong revenue growth for companies like Anthropic etc., just like looking at Meta’s capex comments for Nvidia’s revenue growth.

The 5-year cycle in China investing

Every five years, China will have a new Politburo Standing Committee.

There are 7 members now, which are considered the most politically powerful people in China.

Not all members are new, some can stay for 2-3 term.

But the fight for becoming a new standing committee member can be quite intense.

There can be ripple effects across other areas in China.

Stock market can be quite sensitive to unknowns and turbulence.

To avoid uncertainties, it’s wise to stay away from it.

To compensate for staying on the sideline for about one year, the previous year can be quite a good year.

That seems to the case for the last 2 terms – 2017 and 2022 were the year of new committee selection. The “fight” could start in 2016 and 2021. Thus the “good” years were 2015 and 2020.

The next term is 2027. The “fight” could start in 2026 and 2025 was the good year. That seems to be the case so far.

 

How high can ads ARPU go?

Bilibili, a younger Chinese user focused video website generates advertising revenue at about 7.5 rmb per DAU per month, slightly above $1 ARPU per month.

That is less than half of Kuaishou, which is above 16.5 rmb per DAU per month, or about $2.4 ARPU per month.

Kuaishou’s DAU size is near 4x Bilibili, so ads revenue is nearly 8x larger.

But these are nothing compare with Meta, which commands about $30-$40 ARPU per month in North America.

So Kuaishou is only like 1/10.

Douyin is the highest in China.

Using $50bn pure ads in China and 800mn DAU gives about $5 ARPU per month.

On top of that there are e-commerce, local services, payments, live-streaming etc. so overall Douyin app ARPU is a lot higher than pure ads.

Just comparing ads revenue: Douyin DAU 3-3.5x Meta NA but ads ARPU is 15% of Meta, which will give you about 50% of Meta NA ads rev of ~$100bn.

PE multiple

PE multiple is not only a reflection of earnings quality, earnings growth/cagr, etc.

It’s also an encouragement or discouragement for value creation.

If $1 of profit is worth 10x in HK and 50x in A-share, companies could be more encouraged to create more value for A-share shareholders.

The same rationale also applies to upstream or downstream players – PE multiple can influence whether revenue or profit should sit more or less in supplier or customer etc.

Japan before Meiji vs Qing Dynasty

Japan before Meiji vs Qing Dynasty have many similarities:

1/ both are agrarian based economy that is not industrialized as western peers at the time

2/ both faced challenges and threats from the west. Qing has the Opium Wars from the 1840s; Japan encountered Perry’s Black Ships in 1853–1854.

3/ both had a structured military class – Qing with 八旗 and Japan with samurai. Both had become partly hereditary status groups living on state stipends. Many were no longer effective soldiers.

The Meiji Restoration 明治维新 and 百日维新 Wuxu Reform / The Hundred Days Reform are similar and usually compared together.

The results are vastly different though.

Why?

Shogunate is politically easy to target. Reformers can support the Emperor Meji. Meji and Tokugawa are separate.

However, Guangxu, the Qing Emperor had “invisible enemies” inside – Cixi or other Manchu noble.

The political situation difference had another profound impact – the support of armies.

Meji had key army support from Satsuma 萨摩, Chōshū 长洲 which are clearly anti-shogunate.

Guangxu didn’t have army. Guangxu had Kang Youwei who had ideas but no real power.

Military forces at that time were still loyal to the Qing court on the surface. Whether it’s 袁世凯 or 湘军/曾国藩 or 八旗, no one is clearly anti-Cixi. People could be dissatisfied but Cixi was still the powerful core of Qing court.

Yuan was likely the most possible choice for Guangxu back then, but Yuan chose not the take the risk / help a coup etc.

Yuan Shikai 袁世凯 did not have a clean, rightful banner to be openly anti-Cixi.

Excerpt from 万古江河

明祚二百余年，其实不能与汉唐泱泱大国之风相提并论。其中缘故，可能由于开国之初，太祖立“皇明祖训”，严嘱子孙不得轻易改变他订立的典章制度。固然后世皇帝破坏“祖训”之处不少，这一保守的基调是一锤定音，以致明代政治，难有改革。

…

明太祖废宰相，皇权定于⼀尊，于是⽂官与皇帝之间，无复汉代与宋代可见的制衡。绝对的皇权，保护了保守主义，以致质疑当时制度的思想，都难逃政治权力的压制。皇帝自己不能行使权力时，绝对的权力落在近侍、宦官手中。于是有明一代，自仁宣以后，宦寺鱼肉文官与儒生的事迹，不绝于史书。明代宦官专权，擅作威福，其灾害不在贪污败坏，更在于斫丧了文化与思想的元气。

明代中叶以后，知识分子及社会大众都对上述令人窒息的压抑之气，兴起各方面的反弹。那些史事，将有专节论述，此处不赘。反弹的力量与压制的绝对权力之间，有各种冲突。中国的人才，以及社会的文化活力，都在这一长达百年的斗争中，消耗殆尽！后果是明末的中国，不再有余力面对正在开展的世界新局。能察觉这一变局的人已经太少，更论思考如何适应这一即将叩关的世界形势！

闭关锁国 and 落后 backwardness is a result of A) first emperor’s 皇明祖训 which forbids revolution, B) a lack of a balancing of power, C) pushbacks and crackdowns destroy resources and productivity to move forward.