At the CIIE, two booths told the same story.

At Haier's booth, a professor from Beijing University of Posts and Telecommunications brought a millimeter-wave sensor that reads your heartbeat without touching you, achieving 97.3% accuracy on arrhythmia detection. But when it came to real-world use, the phrase he kept repeating was "the first scene" — the technology is already excellent, but it doesn't yet have a real place to be used.

At another forum, a Singapore-based service provider told a small story: his development team in the Philippines had been cut from more than eighty people to just over twenty, because with AI, "five people can now do the work of twenty." The other fifteen? In his words, "they've all turned into computing-power consumption."

Two booths, one question: as algorithms and computing power keep getting cheaper, where is the moat for Chinese companies? A moat, by the way, is the thing others can't learn and can't take away from you.

The answer may be neither the algorithm nor the computing power, but the work machines can't do — real scenes, compliance, local relationships, the craft that only experience brings — and how much of it you've built up. That's the "scenario density" in the title.

1. Haier's hospital is a table that brings people together

The point about Haier isn't that it built a great AI itself, but that it set a table.

The tabletop is a real hospital it owns — Sichuan Friendship Hospital, a private Grade 3A hospital known for oncology. With real patients, real wards, and real clinical workflows, algorithms finally have somewhere to be put to use.

Sitting around the table are people who normally have nothing to do with each other:

These people come from different fields — physics, neuroscience, cloud computing, imaging, clinical medicine. It was Haier, a company that started out making refrigerators, that gathered them around one table, because Haier holds the one thing they most lack: a real hospital.

Nobody here works for anybody else. Haier doesn't build the algorithms itself — the algorithms belong to these partners. But even the best algorithm, without Haier's hospital, would stay stuck in papers and demos. School-enterprise collaboration and cross-disciplinary co-creation, in plain terms, is trading "scenes" for "technology" — each side gets what it needs.

Someone from United Imaging Intelligence said something on point at the roundtable: in the future, doctors will build their own AI, and manufacturers will have to step back into the role of "platform provider." That's exactly the platform work Haier is doing now.

2. The same logic holds in Southeast Asia

If Haier proves that "algorithms need a place to be used," then Chinese companies going to Southeast Asia prove that "computing power needs a place too."

Chinese computing power is cheap — resold overseas, it can be five to ten times cheaper. Indonesia already has 204 data centers and is still expanding toward Jakarta, Batam, and Surabaya. On the surface, computing looks like a business that works anywhere.

But actually landing it is where the trouble begins. A slide at the forum listed five difficulties of going overseas: people with data-center experience are hard to hire, language and culture barriers, local graduates starting from zero, expatriate approvals taking two to three months, and good talent being poached by big companies — four and a half of those five are about "people."

Harder still are power and licenses. Singapore stopped approving new data centers back in 2019, loosened slightly in 2022 (releasing only about 280 MW across two tenders), and in October 2026 went further, writing green-power commitments into law and using licenses to cap data-center electricity use[6]. Across the water, Johor in Malaysia — the hotspot meant to absorb Singapore's overflow — has raised its approval bar in recent weeks, with S&P calling it a "reset mode"[7][8]. No matter how cheap the computing power, without electricity, licenses, and people, it's just a pile of metal.

3. An old question — how people and machines divide the work — has a new answer

Put the domestic and overseas threads together, and it's really an old question: how should people and machines divide the work?

Information-systems researchers have long had a phrase for this: "task-technology fit" — whether a technology works depends on whether it fits the task at hand[1]. But these two cases point to something more down-to-earth: AI doesn't simply "take over work." It absorbs the work that can be standardized and written down — writing code, running processes — and the work that's left, compliance, negotiation, local relationships, judgment born of experience, doesn't shrink. It gets more sought-after.

The economist Autor observed this polarization long ago: machines eliminate "routine, rule-following" jobs, but they grow the "interpersonal, experience-based" jobs[2]. Bainbridge, a pioneer of automation research, put it sharply — the "irony of automation" is that the more capable machines become, the less dispensable people are, because the hardest, most unexpected, most accountable moments all still require a human[3]. And Brynjolfsson warned of the "Turing Trap": if AI's default direction is to replace people rather than help them, the whole road leads the wrong way[4].

In plain words: AI has done all the work it can do, and what's left is exactly the work machines can't do — and it's worth more. Haier's "hospital scenes," and the "compliance, power, local relationships" of going overseas, are all this kind of machine-proof work.

4. Not who chokes whom — it's complementarity

Look at the Southeast Asian countries together, and it's really a win-win board: China has computing power and technology, Southeast Asia has demand, people, and markets. Each side has its own weaknesses, and together they fill each other's gaps.

Singapore and Malaysia are short on power and land, so data centers can't keep being built there — but Chinese companies hold cheap computing power and AI transformation know-how, so they can bring the technology over and help local companies do more with less electricity.

What Indonesia wants isn't a bunch of cold machines — it's "technology transfer." When it approves foreign workers, what it looks at is whether you can teach the technology to local people. That's exactly a Chinese company's opportunity: instead of just sending people over, bring the technology and the training together, and help Indonesia grow its own local talent.

What Thailand lacks is skilled people, but it has BOI investment incentives, third-party staffing agencies, and an end-to-end pipeline from campus to employment. Chinese companies provide the technology and the jobs; local people provide the manpower — together they build the team.

See? This isn't one side choking the other. Each side is short somewhere and strong somewhere. China lacks overseas scenes and local relationships; Southeast Asia lacks computing power, technology, and investment. Put the two sides together, and that's win-win cooperation.

5. The moat is built from the work machines can't do

Back to the opening question: as algorithms and computing power keep getting cheaper, where is the moat for Chinese companies?

Haier's answer is to build up real hospital scenes and let all kinds of experts land on its own turf. The answer for companies going overseas is to do the "machine-proof work" — compliance, power, local relationships — solidly. The two answers say the same thing: what's truly scarce and hard to replicate has never been the algorithm or the computing power, but the work machines can't do and only people can manage.

Haier started from −80°C refrigerators and walked all the way to the 37°C human body. What it holds isn't the best algorithm, but the hospital where algorithms most want to land. Chinese companies going overseas don't hold the cheapest computing power (that price gap will be erased sooner or later) — they hold the "human work" where computing power most wants to land.

Machines will only get better at working. In the end, the contest isn't about whose algorithm is more accurate or whose computing is cheaper — it's about who has built up more depth in the work machines can't do.


Source note: Figures in this article — "millimeter-wave arrhythmia detection at 97.3% (80,000 samples)", "blood-pressure error of 5/4 mmHg (2,500 patients)", "taste 95%/smell 98.6% (247,000 samples)", "Chinese computing power resold at 5–10× cheaper", "Philippine development team cut from 80 to just over 20 people, with AI enabling 5 people to do the work of 20", "Indonesia's 204 data centers, 7 hyperscale facilities, ~580 MW of installed capacity" — were all disclosed on-site at the 2026 CIIE forum and have not been independently verified. Information on Sichuan Friendship Hospital and home-appliance companies' expansion into healthcare comes from public sources. Singapore's data-center moratorium and electricity figures, and Malaysia's approval-bar adjustments, are cited from public news reports (see references [6][7][8]). Thailand's and Indonesia's foreign-worker quotas and the OSS system are cited from the Ministry of Commerce's Country/Region Guides for Outbound Investment Cooperation ([5]). Judgments such as "scenario density" and "the work machines can't do" are the author's.

References

  1. Goodhue D L, Thompson R L. Task-Technology Fit and Individual Performance[J]. MIS Quarterly, 1995, 19(2): 213-236.
  2. Autor D H. Why Are There Still So Many Jobs? The History and Future of Workplace Automation[J]. Journal of Economic Perspectives, 2015, 29(3): 3-30.
  3. Bainbridge L. Ironies of Automation[J]. Automatica, 1983, 19(6): 775-779.
  4. Brynjolfsson E. The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence[J]. Daedalus, 2022, 151(2): 272-287.
  5. Ministry of Commerce of the People's Republic of China. Country/Region Guides for Outbound Investment Cooperation · Thailand/Indonesia[EB/OL]. https://www.mofcom.gov.cn/dl/gbdqzn/
  6. The Paper. Asia's data-center boom hits a "ceiling": power, water, talent, and public opinion as constraints[EB/OL]. https://m.thepaper.cn/newsDetail_forward_34109613
  7. The Business Times (Singapore). S&P: Malaysia's data-center boom enters "reset mode" as resource and funding realities bite[EB/OL]. https://www.businesstimes.com.sg/zh-hans/international/asean/malaysia-data-centre-boom-enters-reset-mode-resource-funding-realities-bite-sp
  8. Taiwan External Trade Development Council (TAITRA). Malaysia focuses on high-value AI data centers, raising investment thresholds[EB/OL]. https://www.taitra.org.tw/News_Content.aspx?n=104&s=128954