3 Million Yuan to Snap Up PhDs; Post-95s Are Already “Aged”: AI Recruitment Is “Burying Alive” the Middle Tier
TechFlow Selected TechFlow Selected
3 Million Yuan to Snap Up PhDs; Post-95s Are Already “Aged”: AI Recruitment Is “Burying Alive” the Middle Tier
The prosperity of the talent market is illusory; the illusion of liquidity is real.
Navigating Web3 tides with focused insightsBy Ada, TechFlow
“A major Chinese internet company extended offers this year to 60 fresh PhD graduates with AI backgrounds—each at an annual salary exceeding RMB 3 million.” When Xiao Mafeng, founder of TTC—a headhunting firm that has served over 1,500 AI companies—cited this figure, his tone was utterly matter-of-fact, as if reporting today’s temperature.
That same month, Maimai data showed AI job postings surged 29-fold; Zhaopin reported a 200% spike in job seekers. Twenty-nine times more positions, a 200% influx—these numbers gleam like bull-market K-lines.
Yet hidden within these figures lies a secret: massive capital and attention are flooding into an extremely narrow funnel. A few dozen elite individuals at the tip have inflated the entire market’s salary expectations, while hundreds of thousands at the base absorb all the anxiety.
Meanwhile, those in the middle tier—professionals with five, ten, or even fifteen years of work experience—are being quietly hollowed out.
The talent market’s prosperity is illusory; its liquidity is a mirage.
One General Hard to Find, Ten Thousand Soldiers Fighting
According to a Liepin report, 47% of AI positions require master’s or doctoral degrees, and nearly half of employers accept only candidates from China’s elite “Project 985” or “Project 211” universities.
Headhunter Eva put it more bluntly: “When big tech firms hire, a ‘211’ degree is barely acceptable—they demand ‘985,’ and resumes without vertical project experience are virtually ignored.”
What does the tip of the pyramid look like?
On the day Lin Junyang—the lead researcher behind Alibaba’s Qwen—announced his departure, “People from every major tech company reached out to us, asking whether we could connect them with Lin,” recalled Xiao Mafeng.
There may be only dozens of individuals globally at this level. To locate them, headhunters no longer scan resumes. They comb GitHub for code commit histories, track paper authors on Google Scholar, and infiltrate podcast listener groups and AI startup communities. Eva even joined a Tsinghua University AI startup competition WeChat group—populated by 21- and 22-year-olds. “We start chatting early, so when they need jobs two or three years down the line, we’ve already staked our claim.”
Another headhunter, Steve—who began recruiting for AI roles in 2022—offered a telling observation: “I seriously doubt resumes will even exist in the future.”
He gave an example: In January this year, a company needed someone familiar with OpenClaw—a newly emerging domain where no one would list such expertise on their resume. Steve’s solution? Break down the requirement: Fundamentally, this is a multi-agent framework problem. Has anyone built similar frameworks? Are those frameworks open-sourced? Who are the contributors in those open-source communities?
Resumes are losing value; traditional hiring channels are failing.
Some are seizing opportunity from this crack.
Sam, co-founder of DINQ, launched his startup based on a similar insight: The most brilliant paper authors at OpenAI often lack elite academic pedigrees—some even dropped out of school. They’re young, title-less, and indistinguishable to non-technical observers. LinkedIn’s logic—prioritizing degrees and career history—fails completely for AI talent.
So Sam built DINQ: “LinkedIn for AI scientists and developers”—a platform that ignores resumes and focuses instead on verifiable impact: citations in top-tier conferences, GitHub contribution metrics, and whether collaborators include recognized technical leaders. When an HR professional types “Sora 2” into DINQ, the platform surfaces related-paper authors—not just candidates listing “Sora 2 experience”—surfacing talent beneath the surface.
Xiao Mafeng offered an alternative: build in public. Launch your product directly—that’s the strongest proof of capability.
Though 621 universities now offer undergraduate AI programs, McKinsey forecasts a 4-million shortfall in China’s AI talent by 2030. Yet “shortfall” is deceptive: What’s truly missing are experimental scientists who’ve trained models across 100,000 GPUs, and hybrid professionals who simultaneously grasp large-model capabilities *and* identify viable commercial applications. People who declare “I’m super interested in AI!” after listening to two podcasts? The market has never lacked those.
Ye Xiangyu, founder of Niuke, summarized it precisely: “One general hard to find” at the top; “ten thousand soldiers fighting” at the bottom. Maimai’s claim—“two AI positions per qualified candidate”—applies only to the tip. What about the base? No one tracks it, because those resumes never make it into the system.
Leveraged Pricing: Closer to the Model, Higher the Value
So where exactly is the money flowing?
Eva provided some figures. At major tech firms, Level P7 non-technical roles cap out around RMB 1 million annually. For AI technical roles at the same level, salaries range from RMB 1.5 million to RMB 2 million. Jumping between jobs widens the gap further: Technical hires commonly see 50% raises—some even double their pay—while non-technical roles typically rise only 10–20%, rarely exceeding 30%.
Steve used one word to explain this pricing logic: leverage.
Imagine the model as the sun. Those closest to its core wield greater leverage—and thus command higher compensation. A core researcher’s breakthrough in model capability can boost a major tech firm’s market cap by billions. The cost of running 100,000 GPUs vastly exceeds their salary. From this perspective, paying them RMB 100 million wouldn’t be excessive.
Those farther from the sun—product managers, operations staff, salespeople—experience less direct leverage, naturally capping their pay. Steve estimates that in application-layer roles, the salary gap between technical and non-technical roles exceeds two- to threefold.
Xiao Mafeng added a crucial variable. He argues this “hierarchy of prestige” reflects supply-and-demand dynamics operating on two levels. Macroscopically, only a handful of people globally have trained models across tens of thousands of GPUs—so their salaries soar into the stratosphere. Microscopically, however, it depends on the founding team’s DNA. If founders themselves are Tsinghua professors surrounded by lab-based technical talent, then commercialization experts become comparatively more valuable.
The scarcity of a few dozen individuals defines the industry’s entire salary narrative. Everyone else measures themselves against this yardstick—and finds only disparity.
A Purge of the “Old Guard”
“The AI era rejects the ‘old guard,’” Xiao Mafeng declared sharply.
Those who rode the last AI wave—through Megvii and SenseTime—are now in their mid-40s. Their experience has ironically become baggage.
Steve phrased it more diplomatically but pointed to the same conclusion: “We don’t believe old maps can guide us to new continents. People who’ve spent too long in one industry accumulate momentum and inertia. Their brain’s immediate responses reflect past training—but the world has changed, and the correct response may now be the exact opposite.”
Age anxiety permeates every layer. Some investment firms now scout 00s-born entrepreneurs; phrases like “95s are already outdated” are beginning to circulate.
It sounds absurd—but the hiring market sends a very real signal: When resources are scarce, the scale tips decisively toward youth.
“Today it’s about execution speed and rapid deployment. Companies are building special forces—not large formations,” said Steve. And special forces need fewer commanders.
But here lies a contradiction no one wants to confront head-on.
Truly deploying AI products—and converting technology into commercial value—relies precisely on industry experience, tacit knowledge, and hard-won lessons from past failures. Steve himself admits such tacit knowledge resides primarily in more seasoned professionals. They may not know which path leads forward—but they *do* know which paths are certain dead ends.
The industry needs youthful energy—and experienced judgment. Everyone nods along to both statements. But money flows only according to the first.
The Middle Layer Is Being Absorbed
All three headhunters independently noted one shift: Management layers are shrinking.
“Pure management roles are becoming increasingly difficult. So much is being disrupted—your entire operational system might be dismantled tomorrow,” said Steve.
Organizations are flattening radically—replacing hierarchical pyramids with agile squads where every member must fight. Relying on people to execute tasks is giving way to relying on agents. Formerly prized managerial skills—leading complex teams—now face fundamental challenges.
Boundaries among product managers, operations staff, and frontend/backend engineers are blurring. One person, empowered by AI, can now deliver an MVP single-handedly.
Chen Lei (a pseudonym), who served three years as Product Director at a mid-sized AI firm managing an eight-person team, described what happened this January: Her team was disbanded during a company restructuring—four members shifted to agent-product development, two were laid off, and her title changed from “Director” to “Senior Product Manager,” reporting to a technical lead five years her junior.
“I wasn’t fired—but I know this hurts more,” she said. “Everything you built over three years at this company vanishes overnight with an org-chart reshuffle. And you can’t even complain—because they’ll say, ‘You’re still here, aren’t you?’”
This is the cruelest part of the liquidity mirage. At the funnel’s apex, dozens of geniuses compete for sky-high bids. At its base, hundreds of thousands of newcomers can’t even clear the entry threshold. And in the middle? Professionals with five, ten, or fifteen years of experience are being hollowed out from within.
The career ladder has lost its middle rungs. Previously, you ascended floor-by-floor in an elevator; now it’s a parachute jump—either land directly atop the pyramid or free-fall.
Who’s Manufacturing This Mirage?
Who benefits from this liquidity mirage?
Recruitment platforms monetize “AI job postings up 29-fold” and “4-million talent shortfall” headlines—every repost pulls more anxious job seekers deeper into the funnel.
Enterprises use AI as a fig leaf. Forrester Research found that 55% of employers admitted regretting AI-driven layoffs—because the promised AI capabilities weren’t ready. Resume.org’s survey was even more direct: 59% of companies acknowledged rebranding layoffs as “AI-driven,” simply because it sounds better to stakeholders. Citing AI suggests strategic upgrading; citing poor performance implies managerial failure. AI has become the most convenient cover story.
Klarna cut 700 customer-service staff claiming AI replaced them—only to see service quality plummet, triggering customer backlash and prompting quiet rehiring. This isn’t isolated. Forrester predicts half of AI-related layoffs will eventually be reversed—but with lower salaries or outsourced overseas.
Steve captured current leadership mindset precisely: “Their first question is no longer ‘Who should we hire?’—but ‘Do we even need to hire anyone?’”
According to Forrester Research, only 16% of global employees possess high AI readiness. Companies invest little in training—leaving employees to self-educate. Gen Z shows the highest AI readiness (22%)—yet they’re the first pushed out of entry-level roles, precisely because those roles are easiest to automate. Mercer’s survey shows employee anxiety about AI-driven unemployment surging from 28% in 2024 to 40% in 2026.
AI is both the rationale for hiring—and the excuse for firing. Whoever holds the definition power controls the game.
The Funnel Won’t Widen
Return to those initial figures.
29-fold job growth, 200% surge in applicants, RMB 3-million salaries, 4-million talent shortfall. Every number is true—but together, they tell a radically different story: Jobs are growing, yet the funnel’s opening remains razor-thin; applicants flood in, yet the vast majority fail even basic screening; salaries skyrocket—but only for the few dozen at the pyramid’s peak; the shortfall widens—but what’s missing and what’s supplied are fundamentally misaligned.
And this funnel won’t widen. AI advances every six months; today’s hottest direction may become obsolete debris in half a year. You think you’re close to the sun—until a new model launches, and you’re suddenly flung to the periphery.
Steve delivered a line that serves equally well as the industry’s epitaph—or its entry ticket: “Measuring seniority by years of experience may no longer suffice. What matters is the density and depth of your interaction with AI. Someone entered the field four years ago but only uses AI superficially. Another joined last year but immersed themselves entirely. Tell me—who has deeper seniority?”
The three headhunters themselves are being reshaped by this industry. Eva is studying algorithm fundamentals; Steve is diving into agent frameworks; Xiao Mafeng just left a conference with 00s-born entrepreneurs, remarking, “Their cognitive framework has already advanced to the next level.” Even shovel-sellers must keep pace with gold-rush rhythms.
Chen Lei recently started a small project on GitHub: an AI agent framework that auto-generates legal documents. No one assigned it; no one paid her. She realized something: “Rather than waiting to be filtered by the funnel, I’d rather carve my own hole through it.”
This may be the sole near-optimistic note in this piece—though even that is qualified.
Most people are neither among the 60 PhDs earning RMB 3 million nor like Chen Lei—still capable and willing to carve their own path. They’re the silent majority in the funnel’s middle: not elite enough to command astronomical bids, not radical enough to dismantle and rebuild from scratch.
This funnel won’t widen.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
深潮TechFlow
