AI Risks Sparking a New Era of Divergence as Development Gaps Between Countries Widen
Bangkok, 10 Dec 2025 - A new UNDP report warns that unmanaged AI could widen inequality between countries by stretching gaps in economic performance, people's capabilities, and governance. The issue isn't whether AI creates value. It's who can capture it, and who gets left with the downsides.
The report - The Next Great Divergence: Why AI May Widen Inequality Between Countries - argues that countries are starting this transition from very different baselines. Without strong policy action, the long "era of convergence" could stall or reverse.
Asia-Pacific: The epicenter of both upside and risk
Home to over 55% of the world's population, Asia-Pacific now hosts more than half of global AI users. China accounts for nearly 70% of global AI patents, and six economies in the region saw 3,100+ newly funded AI companies.
The upside is real: AI could lift annual GDP growth by about 2 percentage points and raise sector productivity by up to 5% in areas like health and finance. ASEAN economies alone could gain nearly $1 trillion in additional GDP over the next decade.
The risk is uneven benefits. Millions of jobs - especially those held by women and young people - face higher exposure to automation if ethics, safety, and inclusion are an afterthought.
"AI is racing ahead, and many countries are still at the starting line." - Kanni Wignaraja, UN Assistant Secretary-General and UNDP Regional Director for Asia and the Pacific
Thailand's split readiness: strong digital footing, weaker human development
Thailand ranks 52nd out of 170 countries on AI preparedness (AIPI), third in ASEAN after Singapore and Malaysia. That's a solid digital foundation.
But it ranks 76th of 193 in UNDP's Human Development Index. The takeaway: the tech stack is improving faster than people's capabilities to benefit from it - or to be protected from disruption.
"Thailand must continue investing in people through inclusive education, digital skills, and systems that protect communities from emerging risks, so that AI supports a fairer and more sustainable future for everyone." - Niamh Collier-Smith, UNDP Resident Representative in Thailand
The core fault line: capability
Digital readiness is uneven. Singapore, South Korea, and China are investing heavily in compute, infrastructure, and talent. Other countries are still working on basic access and literacy.
Shortfalls in infrastructure, skills, compute, and governance reduce the upside while amplifying risks like job displacement, data exclusion, and rising energy and water use from AI-heavy systems. Only around 5% of people in low-income countries use AI tools. In many contexts, fewer than 1 in 20 people can do basic spreadsheet work.
Across Asia-Pacific, 1 in 4 firms expects job losses. Yet just 1 in 4 people in urban areas - and fewer than 1 in 5 in rural areas - can perform basic spreadsheet tasks. That gap blocks AI productivity gains at the team level.
"The central fault line in the AI era is capability. Countries that invest in skills, computing power and sound governance systems will benefit, others risk being left far behind." - Philip Schellekens, UNDP Chief Economist for Asia and the Pacific
Where AI already improves services
We're seeing practical wins in public services:
- Bangkok's Traffy Fondue has processed ~600,000 citizen reports for faster municipal response.
- Singapore's Moments of Life cut new-parent paperwork from ~120 minutes to ~15 minutes.
- Beijing uses digital twins for urban planning and flood management.
These examples show what's possible when data, compute, and delivery teams align.
Women and youth: higher exposure, fewer protections
Jobs held by women are nearly twice as exposed to automation. Youth employment is already declining in high-exposure roles, especially ages 22-25.
Early-career disruption can compound over time if reskilling and job-matching systems lag behind deployment.
Governance is the bottleneck
Only a limited number of countries have comprehensive AI regulations. By 2027, more than 40% of global AI-related data breaches could stem from misuse of generative AI. Governance hasn't caught up.
Practical, widely adopted tools exist. See the NIST AI Risk Management Framework and the OECD AI Principles to baseline policy and engineering controls.
What IT and development leaders should do now
- Prioritize capability-building over feature-chasing. Budget for digital basics: data literacy, spreadsheet proficiency, prompt quality, and evaluation techniques across non-technical teams.
- Compute strategy with equity in mind. Mix cloud credits, regional compute partnerships, and efficient model choices (distilled, small, or task-specific models) to reduce cost and energy use.
- Production-readiness first. Stand up MLOps: data versioning, lineage, bias checks, red-teaming, model cards, and rollback plans. Track cost-per-inference and latency against service goals.
- Human-in-the-loop by default. For hiring, credit scoring, healthcare, and public services, require pre-deployment impact assessments, human override, and audit logs.
- Target inclusivity where exposure is highest. Fund reskilling for women and youth in roles with high automation exposure. Pair training with job placement and wage subsidies.
- Data governance with guardrails. Classify data, enforce consent and minimization, and block sensitive data from model prompts. Use retrieval with policy filters rather than broad fine-tuning on regulated data.
- Measure what matters. Track adoption (active weekly users), task-level uplift (time saved, error rates), cost-to-serve, and distributional effects across gender, age, and location.
- Plan for energy and water impacts. Choose efficient architectures, batch workloads, and greener regions. Publish energy estimates for transparency.
- Public-sector playbooks. Replicate proven service patterns (issue reporting, benefit enrollment, permitting) with clear SLAs and appeal processes.
- Security by construction. Implement prompt injection defenses, content filtering, PII redaction, and incident response drills for model misuse.
Signals to watch
- Share of firms reporting AI-linked job losses vs. role reconfiguration.
- Basic data-skill coverage: spreadsheet and data-entry proficiency across urban and rural areas.
- AI usage penetration in low-income segments and SMEs.
- Compute access: cloud credit distribution, regional GPU availability, and cost trends.
- Governance maturity: adoption of standardized risk assessments, audits, and public reporting.
Bottom line
The opportunity is large, but so is the spread of starting points. If countries invest in skills, compute access, and practical governance, AI can lift growth and service quality without widening inequality. If not, we risk losing years of convergence gains.
For the full analysis, see: The Next Great Divergence: Why AI May Widen Inequality Between Countries.
Build capability fast
If you're leading teams through this shift, prioritize training that upgrades day-to-day execution - data handling, prompt quality, evaluation, and deployment basics. These are the compounding skills that move outcomes.
Your membership also unlocks:
