THE 2028 CHINESE INTELLIGENCE CRISIS

Bob Chen’s biting burlesque of the Citrini piece imagines China emerging unscathed from the job-killing AI wave.

The 2028 Global Intelligence Crisis, a speculative macro memo published on 22 February 2026 by U.S. research firm Citrini Research on its Substack platform, has caused major tech and financial firms’ share prices to tumble, sparking heated debate among economists and strategists over the realism of its scenario.

Written as a “thought exercise” from the vantage point of June 2028, the piece, co‑authored with Alap Shah, Principal of Lotus Fund, imagined a dystopian aftermath in which rapid AI adoption triggers mass white‑collar layoffs, collapsing consumer demand, and inflicting deep losses on software, payments and delivery stocks.

None of this would have happened in The 2028 Global Intelligence Crisis: The China Edition, which is explicitly modelled on the Citrini piece. The author, Bob Chen, argues that China’s relatively slower AI penetration, its manufacturing‑heavy employment structure, and the entrenched offline, bespoke nature of its information economy would have surprisingly insulated the country from the kind of AI‑induced upheaval depicted in the original. Chen’s tongue-in-cheek observation: In the absence of standardised SaaS platforms and deep digitisation, there was little for AI to replace in the first place.

Bob Chen is an economist‑turned venture capitalist at BroadVision Fund (博华资本), a China‑based VC fund focused on the technology sector. He focuses on Chinese companies going global, backing enterprises that leverage China’s manufacturing and technological capabilities to serve global markets. Previously a macro‑economist at the Chinese Academy of Social Sciences, he now combines macro policy analysis with hands‑on investing to interpret China’s reforms, industrial upgrading, and global supply‑chain shifts.

—Yuxuan Jia

This article was published on Chen’s personal WeChat blog Jolly Maker (嬉笑创客) on 24 February. Chen has kindly authorised and reviewed the translation.

---

---

2028年AI末日推演 中国版

The 2028 Global Intelligence Crisis: The China Edition

Citrini’s 2028 Global Intelligence Crisis sparked a violent market sell-off. Meanwhile, on the other side of the Pacific, China ran the same kind of tabletop exercise.

In 2028, as U.S. unemployment printed 10.2% and equities dropped more than 30% from their 2026 highs, China looked like a different world.

Many people came to see this economy as a fortress that had “survived the AI bombardment”. Of course, that story only took hold in 2028. Back in 2026, as the U.S. bull market roared on, plenty of people were still shaking their heads at China’s relatively weak equity performance.

When U.S. firms used AI to drive large-scale layoffs, China’s AI penetration and roll-out looked glacial by comparison, and it drew plenty of criticism. Few expected that two years later, that “slow progress” would read like an iron wall against AI. U.S. profit growth was built on using AI to cut white-collar headcount, while the very first step of that sequence kept getting stuck in China’s mud.

China was a manufacturing-heavy economy with ~28% output share and over 120 million manufacturing workers (~16% of the total 740 million employed population). But “true” white-collar workers—those in market competition, private firms doing high-cognition work—were less than 4% of total employment (~30 million), and heavily concentrated in top-tier cities.

A much larger cohort of “pseudo white-collar” workers—those in government organs and state-owned enterprises—were not easily shaken by algorithms. In leader-driven systems, AI penetration was not particularly welcome. A lot of information, instructions, and paperwork still moved on paper, and tight confidentiality disciplines blocked AI from meaningfully reading even what had been digitised.

Many meetings still happened in old-style, almost antique conference rooms, with staff coming in every ten minutes to top up tea. Nothing was recorded, and nothing made it to the AI-readable digital domain. What eventually filtered out was often just a highly distilled A4 sheet, with just a few lines of text. Much offline information was private, face-to-face, and vivid. It couldn’t be analysed digitally, and it lived in human memory and judgement.

To reverse-engineer, from a few A4 pages, the vast decision-making information dissemination and the dense web of relationships behind them, was, for AI, close to a hopeless task. The leading position of state organs, state institutions, and state-owned enterprises became a solid fortress. And many private firms that served that system—winning orders, bidding for projects, and operating around it—were likewise hard to digitise in full.

What about the most fragile parts of the private sector? White-collar workers had already taken a shelling during the pandemic that began in 2020, with large numbers pushed into the gig economy. As the saying goes, shells rarely land in the same crater twice. Or put differently: if your position was already blown apart, whoever survived had already adapted. That included many foreign firms—especially U.S.-funded ones—which had been exiting for years.

White-collar incomes did take damage, but if white-collar workers were only a small slice of the economy? The system, in effect, performed an “immune response”, cutting out the cancerous parts with surgical precision.

SaaS collapses, but there is no SaaS here

Across the Pacific, the U.S. faced a SaaS unwind. “Software is eating the world” was once the American slogan; China, meanwhile, spent years complaining that digitalisation could not move under endless bespoke deployments and on-premise requirements. Yesterday’s curse became today’s kudos.

SaaS never truly took off in China. It left behind a long trail of broken investors and operators alike. By 2028, people suddenly realised: if SaaS couldn’t eat this world, AI wasn’t going to take a bite, either.

Paradoxically, an information economy without SaaS DNA can benefit from AI’s rise. SaaS works by packaging “best practices” into standardised tools. But what if your industries don’t converge on best practices? What if every individual and organisation insists on bespoke solutions? In China, that was normal: heavy staffing, offline meetings, and customised development working together, with leaders ultimately making the final call.

This soil was far more suited to AI-driven efficiency gains than to AI replacement.

People who rely heavily on Excel become atomised information and decision terminals. They collect information, process it, and feed it upwards through verbal reporting. Leaders’ relative sluggishness in adapting to digital workflows, ironically, prevented AI from establishing a seamless connection between decision-making and execution. The oral briefings, the eye contact in the room, the interpretation of casual gestures, and the seating arrangements at meetings—these became the true barriers.

The atomised work structure of these information squads allowed those Excel-heavy workers to use AI to raise personal productivity. And remarkably, no one was replaced in this process, because there was no target to replace—there was never that third-party external service provider in the first place. All the while, some customised vendors used AI to improve their delivery efficiency.

Of course, there was still an impact. Chinese suppliers adept at price wars used AI to cut costs and laid off plenty of developers. But developers were only ever a small fraction. Large numbers of implementation and deployment staff still had to fight their way through “mountains of documents and seas of meetings”, and those roles simply cannot be replaced by AI.

There was an unwritten pattern among China’s information suppliers: headcount can be large in the build phase, but once a company stabilises, big cuts often follow. Many developers had long adapted to a life where “35 and out” is a familiar path. In an industry already pushed to the floor, AI could not push the posture any lower. An economy already skilled at in-house customisation gained more momentum with AI’s efficiency boost. One curious beneficiary was gaming and short-video platforms, because developers suddenly had more time to scroll.

Intermediation dies, but the key remains offline

By 2028, LLMs were the default tool. In the U.S., people used AI without even thinking about it. In China, AI was impossible to miss. No matter the entry point, if an agent called into trading platforms or transaction-matching systems—shopping, flights, insurance, property—free usage inevitably comes with a tidal wave of advertising and traffic diversion.

Internet firms internally absorbed the enormous token bill, but there is no free lunch. What is free is often the most expensive, because the price is attention. Voice input became mainstream, yet a slightly unclear command could suddenly launch a shopping app. Users could not pretend they were not inside a specific platform’s walled garden.

The U.S. problem was that many internet platforms could not monetise transaction matching via commissions. China’s tech giants, having already price-warred commissions to the floor, were thoroughly fluent in the logic of “getting something for nothing.” All they did was move the ad monetisation from one interface to another.

When China’s AI agents tried to roam the sea of information, they ran into walls everywhere. Each platform built its own fortress and treated others as a threat. Huge amounts of information locked inside apps became private strongholds that AI could not breach. WeChat couldn’t crawl off Xiaohongshu, and Xiaohongshu couldn’t break into Ctrip. Even by 2028, when you ask an agent to gather high-quality Chinese content from WeChat blogs, it still cannot break WeChat’s anti-crawling mechanisms, and can only scrape the public internet for information that was outdated, untimely, and uneven. The vivid lifestyle tips on Xiaohongshu might as well not exist on the public web.

In 2027, the giants sat down and tried to build interoperability protocols to share each other’s information. But after an incident dubbed the “Wuzhen breakup dinner,” failed bargaining turned the whole idea into a mirage.

In the leaked photos of that dinner afterwards, the twenty-something internet tycoons could only be seen stone-faced, appearing cordial but estranged. Once information dries up, AI becomes a stream without a source.

And those who wanted to use AI agents to replace intermediation fell at the model-training stage. Those trying to use AI to match property transactions found that much of the training data consisted of low-priced, overly retouched listing images engineered for clicks. Especially those who tried to bypass Beike and train on 58.com’s open data ultimately ended up with predicted prices 50% below the market. Users tried it, then shook their heads. In the end, people still walked into offline brokerage shops, where a local behind the desk, who truly understood the physical condition of homes, waited happily for “fish”, with no sign of panic about being replaced.

When DoorDash across the Pacific was battered and collapsed by AI agents, China’s food-delivery giants were in their element. Every effort to promote AI invariably came with offering a cup of milk tea. Milk tea and AI became tightly bound, and food delivery was the only channel through which milk tea reliably reached individual users.

In a market where food delivery was deeply developed, everyone understood the real moats: not information, but logistics and traffic. Control enough couriers, and you can buy traffic cheaply enough to get people to order on your platform—that is the decisive factor. New platforms that could not organise sufficient delivery capacity got sneered at by users who were left waiting, and they certainly did not get to enjoy AI-linked milk-tea subsidies. Meituan even found a second wind through those subsidies, with its share price doubling off the bottom. Many new platforms trying to promote AI ended up funnelling a meaningful share of investment money straight into Meituan.

AI agents hit payment platforms in the United States hard, and card networks such as MasterCard suffered heavy blows. To save the 2% - 3% interchange rate, agents routed transactions through stablecoins as intermediaries. But China, with its foresight, had long since cut off this path at the root. In 2025, many disagreed with this move. Today, they are ashamed of themselves. Online payments, already low-fee, did not suffer too much impact in this process.

Private credit collapses, but households already deleveraged

While the U.S. scratched its head over private credit blowing up, China found the episode puzzling. The conclusion was that America’s financial innovation had gone too far, building flashy structures on fragile underlying assets—until one crack brought down the whole edifice.

Countries that regulate financial products strictly do not face the same worry. The personal deleveraging wave that began in 2022 meant borrowing did not even get the chance to erupt in the first place, let alone private credit. The biggest transmission channel of the broader AI shock to China was probably those who invested in U.S. equities, and even that had been quietly discouraged back when the Chinese government’s global taxation arrived in 2026.

Geopolitical shifts, and a massive token trade surplus emerges

Yes, to respond to a shock of this scale, the U.S. once again revived the push to bring manufacturing home. It had discovered that white-collar workers were becoming increasingly vulnerable under AI’s impact, while a manufacturing heavyweight like China stood firm. That triggered a new scramble for critical resources and industrial capacity. Geopolitics grew increasingly tense. The U.S., sensing economic strain, tried—while it still held military advantage—to secure more global resources and rebuild supply chains.

But a question followed: can an economy built on a virtual world and information systems really support extreme control over the physical world? Can firms that have forgotten how to turn formulas into products through dense manufacturing clusters actually stand up new industrial bases?

China, of course, has its own sweet troubles. Several Chinese frontier-model firms, able to train models and run inference at extremely low cost, export vast quantities of tokens, pushing China’s token export surplus to an extreme.

The U.S. has begun to accuse China of helping destroy the American economy by supplying “intellectual weapons”. Streams of cheap tokens flow to U.S. users, backed by China’s low-cost electricity.

China’s domestic system, meanwhile, stays intact. Facing the accusations, China shot back: wasn’t this exactly what the U.S. did in World War I? And wasn’t that precisely the opening that powered America’s rise?