The Water Is Rising, and Nobody Will Tell You Why

01 / THE PRIVATE MEETING

The layoff arrives one room at a time.

In mid-May, a group leader called Lin Yue into a meeting room at Ctrip and got straight to the point: he was on the list.

Lin had expected it — rumors of layoffs had been circulating inside China's internet companies since March. He'd joined the company a year earlier, fresh out of university, landing what looked like a lucky break in the company's most profitable division. When the moment actually arrived, though, composure didn't come easily. His hands started shaking. He needed a long moment before he could walk through the door. Later, he said simply that he never wanted to go through it again.

Nobody at Ctrip announced a layoff. Nobody at Meituan, Alibaba, or Tencent has either. What's happening instead is quieter and harder to point to: contracts left unrenewed, headcounts “optimized,” teams folded into new divisions that happen to need far fewer people. It doesn't show up as a press release. It shows up as thousands of individual meetings like Lin's, each one framed as a private, almost personal decision, none of them officially connected to the others.

They are connected.

Late last year, AI wrote roughly one-fifth to one-third of the code shipped inside China's major internet companies. This year, at some firms, that figure has climbed to 90 percent. That single number is doing more to reshape who keeps their job in Chinese tech than any strategy memo has.

The scale, once you start counting, is hard to explain away as routine belt-tightening. By one widely circulated tally of public disclosures, Alibaba's total headcount fell from roughly 194,000 to 128,000 over the course of a year — a drop of more than 66,000 people, around 34 percent of the company. JD's first-quarter net profit for 2026 fell 53 percent year over year even as its founder publicly pledged not to fire a single frontline worker displaced by robots.

03 / COMPANY SCRIPTS

The logic rhymes. The mechanics do not.

One analysis of the sector frames the last five years of tech layoffs in China and the U.S. as three overlapping waves: de-bubbling in 2022–23, when companies simply cut pandemic-era overhiring; normalization in 2024–25, when quarterly "optimization" became routine; and, starting in 2026, structural AI replacement — cutting the roles AI can now do, and funneling the savings straight into AI infrastructure. Each wave, the analysis argues, has been less reversible than the one before it.

The mechanics differ by company, but the logic rhymes. Alibaba didn't just cut headcount — it restructured around a new division called Token Hub, run directly by the CEO, built around the idea that the flow of tokens matters more than the cost of people. The name alone tells you where the company thinks value now lives. ByteDance has reportedly taken a more surgical approach: non-AI departments face a roughly 20 percent staff review every six months, while core AI teams face closer to 5 percent — a company that is, internally, fairly explicit about which parts of itself it considers expendable. Tencent's Docs team, according to industry commentary this spring, shut down its Beijing office entirely, consolidating the group in Shenzhen and offering affected staff internal transfers dressed up as opportunity. And at Tencent's cloud and enterprise group, 36Kr reported, engineers were once handed a personal monthly token allowance worth roughly $2,000 to spend on AI coding tools — with usage tracked as a performance metric. Employees who didn't spend enough got asked why by their managers; some quietly lent their unused quota to colleagues rather than explain a shortfall.

Not every company has picked the same script. On May 27, JD's founder made headlines with an internal speech promising that not one frontline worker replaced by a robot would be let go, backed by a national network of over 80 "robobase" retraining centers meant to turn displaced blue-collar staff into robot technicians. His stated ambition: for JD to still be China's largest employer in twenty years. On the very same day, Alibaba eliminated its traditional 13th-month salary, folding it into a renamed year-end bonus and pushing the payout from December to the following April or May — a change that, among other things, makes it cheaper to lay someone off before that bonus ever has to be paid. Read side by side, the two announcements say more about the range of choices companies are actually making than either one says alone.

04 / THE ATMOSPHERE

Nobody knows where the water is going.

None of this fully explains itself through cost-cutting alone. A former Meituan employee described a subtler and, in some ways, more corrosive pressure: after meeting an AI industry figure, the company's founder told his management team that the water was already rising and everyone had better learn to swim, fast. What followed wasn't a plan. It was an atmosphere — every team was expected to find something, anything, to bolt onto AI, whether or not it was ready, just to prove effort had been made. The employee's complaint wasn't really about AI. It was that nobody above him seemed to know exactly where the water was rising toward, only that standing still wasn't an option.

Further up the ladder, the anxiety doesn't disappear — it just gets more competitive. An Alibaba engineer building an internal AI coding tool described watching colleagues at every level chase the same validation simultaneously: one engineer trying to automate a single task, another trying to automate an entire workflow, a more senior architect already sketching a company-wide automation platform that would make both of their efforts redundant before they shipped. His own conclusion was bleak — that almost everyone chasing safety through AI projects would end up, in his words, running alongside the winners rather than being one of them, and that the industry might have no room left for most of them within a year or two.

Younger, cheaper employees have absorbed the sharpest edge of this. In a 36Kr investigation, one campus hire, Li Chuan, joined Baidu in 2025 doing frontend work at a moment when AI still functioned mostly as a smarter search box during interviews. By this April, that had changed completely: when Zhipu released its GLM-5.1 model, cheap and, in Li's own assessment, a viable substitute for tools like Claude Code, he understood immediately that his job was no longer secure. He was right — by May, he was on a list. HR staff quoted in the same report describe a consistent pattern across companies: frontend and QA roles are usually first, because managers now see less "value" in work AI can visibly replicate, and junior staff cost less to let go than veterans who actually understand what the business needs from the tools in the first place. One line from the same investigation, echoed by several of the engineers interviewed, captures the shift in identity better than any statistic: without AI, they said, they simply couldn't work anymore — and if a coding assistant went down, they'd rather wait for it to come back than open the code themselves.

05 / THE VISIBLE 80%

The missing work is the work nobody documented.

The trouble is that the logic driving all of this rests on an assumption that doesn't hold up well under scrutiny. Companies making these cuts routinely see only the visible 80 percent of a role — the parts that show up in tickets and documentation — while missing the invisible 20 percent: the judgment calls, the vendor relationships, the tacit knowledge nobody wrote down because it never seemed necessary to. When an AI agent hits an edge case it can't handle, the senior employee who once knew the answer is often already gone. AI systems hallucinate with confidence and without anyone clearly accountable for the result. And automating a process frequently creates new problems that then require more automation to patch — a loop with no obvious floor. Every technological leap, the argument goes, doesn't eliminate scarcity so much as relocate it: once execution becomes cheap, judgment, integration, and the kind of institutional memory that lives in long-tenured people become the things companies can least afford to have cut.

06 / TEMPORARY BUFFER

Messy systems buy time. They do not buy safety.

Zoom out to a global comparison and the picture gets less tidy, and more interesting. Silicon Valley's AI-driven layoffs have hit with a force China hasn't yet fully experienced — one widely shared piece of commentary among Chinese tech workers points out that Tencent's headcount actually grew slightly last year, and Huawei's R&D staff has kept expanding past 114,000, even as Oracle, Google, and Meta cut deep. The explanation offered isn't that Chinese engineers are somehow safer. It's structural: American firms run on a SaaS stack with penetration reportedly above 70 percent, meaning AI can slot directly into standardized, cleanly digitized workflows and eliminate entire functions overnight. Chinese firms, by contrast, are said to spend roughly 10 percent of IT budgets on software versus hardware and infrastructure — the underlying systems simply aren't clean or modular enough yet for AI to switch off a department in one move. Chinese engineers are also, on average, a fraction of the cost of their Silicon Valley counterparts, which buys a little more time before the math tips. None of this is protection. It's a delay, and the commentary itself concludes the buffer is temporary — a byproduct of unfinished digitization and cheap labor, not evidence of durable safety.