Hong Kong graduate vacancies fell 60 percent. Median starting pay rose. Both are true.
One city, one cohort of graduates, two labour markets moving in opposite directions. Economists call this shape K-shaped: one arm rising, one falling, and the middle disappearing.
Start with Hong Kong. The Census and Statistics Department reported a seasonally adjusted overall unemployment rate of 3.9% for July to September 2026 (provisional). On that number alone, the labour market looks stable. But within the same release, unemployment among 25 to 29 year olds runs materially above the overall rate. The problem is not that there are no jobs, but that the market no longer absorbs untrained people by default.
The vacancy structure explains why. JIJIS data show 6,811 full-time vacancies for Hong Kong university graduates in Q1 2026, down 14.1 percent year on year, with recruiting employers down by a fifth, the third consecutive annual decline. Over a longer horizon, the Labour and Welfare Bureau disclosed at LegCo that bachelor-level vacancies fell from about 80,000 in 2022 to about 31,000 in 2025.
Yet starting pay has risen. The UGC puts average annual salary for bachelor graduates at HK$336,000, up 2.1 percent, and JIJIS puts the median monthly salary for graduate vacancies at HK$20,000. Six in ten vacancies gone and pay up is not a contradiction; it is precision pricing.
Within a single quarter, some sectors more than doubled their vacancies while others fell by over 70 percent. Direction is settled by data, not by instinct.
| Sector | Change |
|---|---|
| Accounting and audit | +194% |
| Finance and insuranceLargest volume | 4,920 |
| Logistics, hotels, tourism, retail | Counter-cyclical |
| AI and robotics application | Continued |
| Sector | Change |
|---|---|
| Entry-level administration | near −90% |
| IT programming | about −80% |
| Surveying, electronics, mechanical | over −70% |
| Construction, traditional programming | over −60% |
Reading note: the 194 percent rise in accounting and audit includes both base effects and short-term demand from buoyant equity markets, and should not be read as a long-term trend. Determining whether a sector is genuinely expanding requires two or more years of consecutive data. Sources are in the register below.
The Labour and Welfare Bureau’s September 2026 Interim Update of Manpower Projection makes this concrete: as enterprises expand AI use for routine administrative work, Hong Kong’s clerical support workforce is projected to face a surplus of 25,000 to 30,000 people by 2028, with 20 to 30 percent of positions facing restructuring.
The same report narrows the projected 2028 overall manpower shortfall from 180,000 to about 130,000, attributing the improvement principally to slower manpower demand growth arising from economic transformation and the wider application of AI.
The answer is not that everyone becomes an engineer. New roles cluster in a narrow but deep fissure: people who understand an industry and can direct AI to deliver finished work. Hong Kong has only about four thousand such hybrid professionals, while demand grows by multiples.
| Emerging role | What the work involves | Direction |
|---|---|---|
| AI application specialist | Decompose business processes into steps AI can execute, verify output quality, handle exceptions. No model training required. | Rising |
| Workflow designer | Design the human-AI division of labour: which step goes to AI, which stays human, how handover and review work. | Rising |
| Data governance | Ensure data entering AI systems is accurate, compliant and traceable; manage annotation teams and quality standards. | Rising |
| AI product operations | Measure whether AI tools create real business value, and continuously adjust usage and adoption strategy. | Rising |
| AI risk and compliance | Handle data privacy, model bias and vendor risk; critical in regulated sectors such as finance and healthcare. | Rising |
| Human-AI team lead | Lead teams combining people and AI agents; responsible for objectives, quality control and capability building. | Rising |
| Care, education, counselling | Physical presence, emotional connection and long-term trust; lowest automation risk and the largest absolute global growth. | Growing |
| Basic copywriting, junior programming, clerical | Highly repetitive and reducible to explicit instructions, precisely the task sets AI absorbs first. | Declining |
Global data points the same way. The WEF reports that by percentage the fastest-growing roles are Big Data Specialists, FinTech Engineers, AI and Machine Learning Specialists, and Software Developers. By absolute numbers, the largest growth is in farmworkers, delivery drivers, construction workers, care workers and teachers.
One detail is easily overlooked: the most common workforce response to AI is internal reskilling, planned by 77 percent of employers, rather than outright redundancy, planned by 41 percent. Nearly half expect to redeploy AI-exposed staff elsewhere in the business.
Four levels, shallow to deep. The first can begin today; the fourth requires sustained investment. Each corresponds to a capability the market is actively validating.
AI replaces tasks, not jobs. The first capability is not prompt writing but clear decomposition: what steps make up this work, which have defined inputs and outputs, which require judgement and human sensitivity. Those who can decompose clearly know what to hand over and what to protect.
Employers do not screen on attendance; they screen on whether you can show a case where AI solved a real business problem, and explain how you set standards and verified quality. One directed deliverable, with a clear account of your judgement, outperforms ten certificates.
The former includes cross-domain integration, client relationships, complex negotiation and creative strategy. The latter includes data analysis, market research and product operations. Avoid the third category: purely standardised, fully decomposable entry tasks.
For non-Cantonese speakers, language should not be framed only as a barrier. Hong Kong simultaneously needs written Traditional Chinese, spoken Cantonese, English for international liaison, and understanding of Mainland business. The question is never whether your Chinese is good enough, but whether you can position the combination as a specific function.
Each data point carries its original source and reference period. Where a figure is revised or updated, we correct it in step, without embellishment.
Note: some figures are published by different bodies on different statistical bases. This page adopts the official figures of the Census and Statistics Department and JIJIS as authoritative; media-cited figures are background only.
Join free to use the AI career coach for a structured self-assessment, enrol in free AI courses co-organised with AIX Society, and enter licensed deep placement when ready.