Ten times the efficiency — so why start losing money?

On WeChat Channels, economist Fu Peng made an uncomfortable point: AI has raised production efficiency tenfold, yet AI comic dramas have started losing money.

It stings because it violates something we were all taught: higher efficiency, lower cost, thicker margins. Reality runs the other way. The problem is not AI — it is who allocates the means of production.

1. First, the numbers: where did the efficiency dividend go

In the first half of 2026, China's AI micro-drama and comic-drama market was about RMB 22 billion, with the full year expected to pass RMB 40 billion. Supply is even more extreme: on Douyin alone, 221,900 new AI comic dramas appeared in six months — over 1,200 a day, or one every 20 seconds at peak. Nationwide, 430,000 micro-dramas launched in the first eight months, 13 times the whole of last year, more than 90% of them AI-made.

The hit rate? 0.47%. The rate of passing 100 million views? 0.48%. About nine in ten companies lose money; industry estimates put the share that actually clears break-even at roughly 5%.

Capacity grew thirtyfold. Profit did not grow — it disappeared. That is the paradox this article tries to explain.

2. Who received the money that was saved

Fu Peng's answer is two words: paid traffic. AI pushed production cost down to a few hundred yuan per minute, but as supply exploded, being seen became the expensive part. The money saved on production went, almost intact, to the traffic platforms — plus compute costs and GPU depreciation. Do the math and, in his words, “you'd be better off delivering food.”

One fact is easily missed: traffic is also a means of production. When everyone can produce content, what is scarce is no longer content but the gateway to attention. Whoever owns the gateway takes the efficiency dividend.

So AI comic dramas are not losing money because the technology is weak — they are losing money because the technology removed the barrier to entry. Remove the barrier, supply becomes infinite, price collapses. No industry escapes that rule.

3. The real question: who allocates the means of production

Zoom out one level. An economy's industrial order depends essentially on three things: who allocates land, capital, and access — licences and traffic. Different allocation rules produce completely different outcomes from the same efficiency gain.

The table below is how I read the two allocation logics. It judges neither as superior — it only describes the mechanism.

• China: state-led allocation plus platforms holding the traffic. Local governments use land, subsidies and industrial funds to pull capacity up fast; platforms use algorithms to concentrate attention at a few gateways. Capacity forms extremely quickly, but demand does not scale with it.
• Singapore: the state allocates rules, not goods. Government supplies rule of law, tax architecture, land planning, talent visas and external connectivity; firms decide what to invest in, how much to make and whom to sell to. The state does not decide which company makes comic dramas or how many.
• The difference in outcomes. The first can build world-leading capacity in a sector within twelve months — and bankrupt nine in ten participants in the same twelve months. The second rarely produces such explosions, but also rarely produces an entire industry losing money at once.

That is the difference in how governments think about allocating society's means of production: one allocates things, the other allocates rules. A government allocating things must judge, on the market's behalf, which direction is right; a government allocating rules only guarantees the game is played by the rules. Both demand great capability — but the cost of error differs.

4. Two allocation models, two costs

• The cost of the Chinese model: involution and overcapacity. Once policy pulls capacity up, it is hard to bring down — jobs, tax revenue and local KPIs are tied to it. So price wars run until the whole industry loses money, and nobody wants to exit first.
• The cost of the Singapore model: a size ceiling and high costs. A market of 4.7 million citizens cannot support mass manufacturing; labour, land and rent are expensive, so firms must go regional or global very early — or they do not survive.
• But the Singapore model carries a hidden benefit: it forces firms to become irreplaceable. You cannot win here by scaling into a price war; you win on expertise, reputation, position and familiarity with rules.

5. Three lessons for anyone running a business

• Efficiency is not a moat; scarcity is. Anything AI lets everyone do is worth little no matter how fast you do it. The question is not “can I go faster” but “can others not do this — and is that because I am here?”
• Identify who holds the gateway in your industry. In content it is the traffic platform; in education it is schools and permits; in accommodation it is location and compliance. Your margin depends on your relationship to the gateway, not on how hard you work.
• Between two markets, translate rather than transport. Explain Singapore's rules, schools as they really are and accommodation as it really looks in language Chinese parents understand. AI cannot do that, because AI is not on the ground.

In closing

When Fu Peng says new productive forces cannot change old production relations, I read the emphasis not as “technology is useless” but as: technology only makes the cake bigger; how it is divided is a separate system.

For SMEs in Singapore this is arguably good news: we cannot compete on capacity, but we can compete on the small thing nobody else can replace. That small thing is often the entire business.

Video source note: the video referenced is “Fu Peng: AI makes efficiency ten times higher — so why are AI comic dramas losing money?”, published by WeChat Channels account “Buhuijiuwen TALK” on 6 November 2026; its public remarks are used as a starting point for discussion. Data come from public industry reporting (DataEye, ifeng, Huxiu, Sina Finance and others). The views and further analysis are written independently by this publication and do not represent the video author's position.