The Next Great Divergence: Why AI may deepen inequality between countries

UNDP warns that uneven AI readiness risks a “Next Great Divergence,” with AI reaching 1.2 billion users in three years and nearly 70% of users in developing countries, yet usage ranges from roughly 66% in some high‑income economies to about 5% in many low‑income countries. The report says scaled AI could lift annual GDP growth by ~2 percentage points and boost productivity up to 5% in sectors like finance and healthcare (ASEAN could gain nearly $1 trillion over a decade), but also expects major labor disruption (75% of firms foresee job losses), disproportionate risks for women and informal workers, and rising cybersecurity and governance gaps unless countries invest in connectivity, skills, compute and regulation.

Analysis

Market structure will widen concentration: hyperscalers and AI‑compute suppliers (NVDA, TSM, ASML, MSFT, AMZN, GOOGL) are primary winners as GPU/cloud capacity becomes a scarce, high‑margin input; data‑centre REITs (EQIX, DLR) and cybersecurity vendors (PANW, ZS) capture recurring revenue. Losers include labor‑intensive informal sectors in EMs (India, Indonesia) and incumbents without data/governance capacity; ASEAN may gain ~US$1T over a decade while 75% of firms expect job disruption, amplifying capital‑labour divergence.

Tail risks: sudden export controls on advanced nodes or large‑model IP (days–weeks) and a spike in gen‑AI misuse leading to >40% of AI‑related breaches by 2027 could trigger regulatory fines, market shocks and higher credit spreads for EM sovereigns. Near term (0–3 months) watch regulatory headlines; medium (3–12 months) is enterprise capex reallocation; long term (3–5+ years) is structural divergence in productivity and FX/reserve pressures.

Trade implications: favor long positions in GPU & cloud leaders and data‑centre REITs for 6–24 months and add cybersecurity for defensive carry; tactically reduce broad EM beta and rotate into Singapore/ASEAN tech/infra ETFs where governance/connectivity is stronger. Use option structures (6–12 month call spreads on NVDA/PANW) to cap premium and buy EM downside protection (puts on VWO or CDS) as a tail hedge.

Contrarian angles: consensus underestimates mobile‑first AI in low‑cost markets that can leapfrog via edge models—this favors local telecoms and regional cloud providers if data‑localisation rules proliferate. Hyperscaler multiples already price near‑term wins; consider partial hedges rather than naked longs. Historical analogue: internet era produced initial winner concentration then multi‑year catch‑up by regional players when infrastructure matured.

Key Decisions for Investors

• Establish a 1–2% portfolio long in NVDA via a 6–9 month call spread (buy 6‑month ATM call, sell 30% OTM) to capture sustained GPU demand while limiting premium outlay; review after quarterly data‑centre capex prints (next 3 quarters).
• Allocate 2–3% to data‑centre REITs EQIX + DLR (equal weight) for a 12–24 month hold to capture secular colocation demand; trim 50% if 10‑year UST yield rises >100bps from current levels or if same‑store NOI growth falls below 4% YoY.
• Reduce broad EM equity exposure (VWO/EEM) by 2–4% within 30 days and reallocate 1.5–2.5% to Singapore ETF (EWS) and Vietnam ETF (VNM) combined, targeting countries with stronger connectivity/regulation and expected AI dividend capture over 3–5 years.
• Buy 9–12 month call options on PANW or ZS (size 1–1.5% equivalent exposure) to hedge cyber‑risk monetization, and simultaneously purchase 6–12 month puts on VWO (0.5–1% portfolio cost) as protection against an EM divergence/shock caused by regulation or capital flight.