Socio-economic Impact and Governance of AI/AGI (2026–2035)

Executive Summary

By mid-2026, Artificial Intelligence (AI) has moved beyond the “proof-of-concept” phase into a stage where enterprises demand a clear return on investment (ROI) and significant workflow redesign. The outlook for the next decade is not one of “robots taking all jobs” overnight, but rather a complex restructuring process, unfolding at vastly different speeds and scales across industries, age groups, and nations.

Four key drivers will shape the 2026–2035 period: (1) increased productivity, but with uneven distribution, as most enterprises have yet to realize clear financial value; (2) a tightening entry-level job market rather than mass layoffs; (3) energy and water becoming physical bottlenecks for AI ambitions; and (4) compute power concentrating in a few corporations and nations, transforming AI into a geopolitical battleground. Concurrently, three governance models – “risk-based” (EU), “innovation-first” (US), and “content control” (China) – are competing to define global rules.

This article synthesizes estimates from McKinsey, IMF, Goldman Sachs, IEA, Stanford University, and regulatory bodies, while also presenting two contrasting perspectives (optimistic and cautious) for readers to evaluate.

---

1. Jobs and Productivity: Big Promises, Fragmented Reality

Immense Economic Potential…

McKinsey estimates that generative AI could contribute $2.6 trillion–$4.4 trillion in economic value annually across approximately 63 application areas, primarily in customer service, marketing & sales, software engineering, and R&D (McKinsey, 2025). Goldman Sachs is even more optimistic: if widely adopted, generative AI could boost global GDP by approximately 7% (nearly $7 trillion) over 10 years and increase labor productivity by about 1.5 percentage points annually (Goldman Sachs, 2023). The IMF, more cautiously, forecasts that AI could boost growth under various scenarios but warns that benefits heavily depend on each economy’s capacity to absorb technology (IMF Working Paper, 2025).

…but the “Productivity Paradox” Persists

The gap between potential and reality remains wide. According to McKinsey’s State of AI 2025 survey, although nearly 9 out of 10 enterprises have deployed AI in at least one function, only about 5–6% of organizations report tangible financial benefits, with most reporting an impact on earnings before interest and taxes (EBIT) of less than 5%. The decisive factors are not acquiring the most powerful models, but rather workflow redesign, leadership commitment, and robust data governance. This is why the 2025–2026 period is being termed the “optimization phase”: businesses are using AI to cut costs and streamline operations rather than expand headcount.

The Shock Concentrates on Young People and Entry-Level Positions

The IMF estimates that AI will impact approximately 40% of global jobs – up to ~60% in advanced economies and ~26% in low-income countries (IMF, 2024–2025). Crucially: about half of the affected jobs are augmented (AI helps workers perform better), while the other half are displaced. Goldman Sachs notes a striking figure: the equivalent of 300 million full-time jobs are at risk of automation, but emphasizes that most are temporary displacements and could be offset by new job creation.

The clearest empirical evidence to date comes from research by the Stanford Digital Economy Lab (November 2025): employment for workers aged 22–25 in AI-impacted sectors has seen a relative reduction of approximately 16% since generative AI became widespread; specifically, entry-level jobs in software engineering and customer service decreased by nearly 20% between late 2022 and mid-2025 (Stanford / CNBC, 2025). Notably, companies are not cutting salaries but rather reducing hiring – creating a “skills trap”: the entry-level rung for gaining experience is shrinking, while the demand for senior personnel adept at collaborating with AI processes remains unmet.

---

2. Energy and Environment: AI’s Physical Bottlenecks

The biggest barrier to AGI ambitions may not be algorithms, but rather electricity and water.

Electricity. The International Energy Agency (IEA) estimates global data center electricity consumption at approximately 415 TWh in 2024, projected to rise from ~485 TWh (2025) to around 945 TWh by 2030 – more than Japan’s entire current electricity consumption, accounting for about 3% of global electricity demand. Electricity demand specifically from AI-optimized data centers is expected to more than quadruple by 2030; the IEA’s baseline scenario sees this figure reaching ~1,200 TWh by 2035. The US accounts for the largest share of this increase, followed by China (IEA, 2025).

Water. Data centers consume water directly for server cooling and indirectly through electricity generation. A widely cited study (Li & Ren’s group) estimates that global AI infrastructure could consume 4.2–6.6 billion m³ of water annually by 2027 – nearly half of the UK’s annual water withdrawals (Tom’s Hardware summary, 2025). This is a serious issue as most of this water is drawn from regions already facing water stress.

The Green Paradox. The shift towards renewable energy or biofuels to reduce data center carbon emissions, however, increases pressure on land use and water resources – a trade-off without a straightforward solution.

---

3. Power and Geopolitics: The Compute Race

AI amplifies the concentration of power at two levels. Enterprise level: a handful of major technology corporations (Big Tech) control frontier models and most compute infrastructure. National level: the US and China lead, with competition revolving around chips and computing power.

Export controls on chips are a central tool for the US. By late 2025–early 2026, the policy showed signs of softening: in December 2025, the US allowed the sale of certain chips to approved customers in China, and in January 2026, the Department of Commerce shifted to case-by-case license reviews with strict conditions (US testing, volume capped at 50% of domestic sales, 25% tariff) (Mayer Brown, 2026). Nevertheless, analysts suggest that the US maintains a dominant advantage in manufacturing compute in the short term due to China’s bottlenecks in advanced chip production.

In response to this concentration, a wave of “Sovereign AI” has emerged: India, UAE, South Africa, and many other countries are building their own infrastructure, data, and models to avoid “digital extractivism” – i.e., becoming data providers while value flows to a few technology hubs. The core tension: expensive and difficult-to-localize frontier models remain in the hands of US/Chinese Big Tech, causing “AI sovereignty” efforts to often stop at the application and data layers rather than the foundational model layer.

---

4. Governance Frameworks: Three Models, One Tug-of-War

European Union – Risk-Based Governance. The EU AI Act is the world’s first binding law, implementing a phased rollout: obligations for General-Purpose AI (GPAI) models become effective on August 2, 2025; obligations for high-risk AI systems apply from August 2, 2026; and full application (including AI embedded in regulated products) takes effect from August 2, 2027. The highest penalties can reach up to €35 million or 7% of global annual turnover for prohibited practices (artificialintelligenceact.eu).

United States – Innovation-First, Anti-“Barriers”. The US does not yet have unified federal legislation. On December 11, 2025, President Trump signed an executive order, “Ensuring a National Policy Framework for Artificial Intelligence,” establishing an AI Litigation Task Force to challenge state-level AI laws deemed “burdensome” and threatening to cut certain federal funding sources (The White House, 2025). The executive order carves out laws pertaining to child safety, compute infrastructure, and public procurement. Observers anticipate that states will resist in court, turning AI governance in the US into a federal-state tug-of-war.

China – Content Control, Domestic Data. China has adopted a strict content management approach: Regulations for Labeling AI-Generated Content (effective September 1, 2025) mandate clear labeling (text/audio/graphics) and hidden labels (metadata) for AI-generated content (China Law Translate, 2025). Concurrently, Beijing announced a Global AI Governance Action Plan (July 2025), proposing the establishment of an international AI cooperation organization (potentially in Shanghai) and asserting the United Nations as the “primary channel” – a move to compete for influence with the West in shaping the rules of the game.

International Efforts. The United Nations (via the Global Digital Compact) and multilateral forums are striving to create common ground, but fragmentation among the three models makes a unified global framework still distant. A notable shift: the focus of governance is moving from “AI ethics” and static content management towards accountability for agentic AI – systems that make autonomous decisions and take actions, raising new questions about who is responsible when agents cause harm.

---

5. Scenarios and Counterarguments

Aspect	Optimistic View	Cautious View
Jobs	AI augments, creates new jobs like all tech revolutions; wages unaffected	Entry-level rung shrinks, youth suffer; inequality grows (IMF)
Productivity	GDP could rise +7% in 10 years (Goldman)	Only ~5–6% of enterprises see real benefits; “productivity paradox” persists
Environment	AI also optimizes power grids, accelerates clean energy	Electricity x2 and billions of m³ water create real pressure on scarce regions
Power	Sovereign AI helps decentralize power	Frontier models still held by Big Tech; dependence difficult to unwind

Short-term Scenarios (2026–2027): EU AI Act enters high-risk phase; heavy investment in electricity and renewable energy for data centers; federal-state legal tensions in the US.

Mid-term Scenarios (2028–2030): Agentic AI becomes industrial standard; clear pressure from imbalances in mid-level and entry-level labor; public disclosure of carbon/water footprints becomes common.

Long-term Scenarios (2031–2035): Potential for early AGI threshold; nations clearly define sovereign AI models to balance dependence; global governance either converges around certain norms or fragments into “techno-blocs”.

---

Implications for Vietnam and Regional Readers

For a developing economy, opportunities lie in the application and sovereign data layers: building capacity to use AI to boost productivity in services, manufacturing, and agriculture – rather than pursuing expensive frontier models. The greatest risk for young workers is the shrinking entry-level rung, so investing in AI collaboration skills (evaluation, verification, process coordination) is more crucial than skills easily automated. Regarding infrastructure, the electricity-water challenge for data centers needs to be integrated into energy planning from the outset.

---

References

• McKinsey & Company (2025). Where AI Will Create Value — and Where It Won’t / State of AI 2025. Link
• International Monetary Fund (2024–2025). Gen-AI: Artificial Intelligence and the Future of Work; The Global Impact of AI — Mind the Gap (WP/25/76). Link
• Goldman Sachs (2023). Generative AI Could Raise Global GDP by 7%. Link
• International Energy Agency (2025). Energy and AI — Executive Summary. Link
• Li, Ren et al. / summary (2025). AI Water Footprint 2027 (4.2–6.6 billion m³). Link
• Stanford Digital Economy Lab / CNBC (2025). AI and entry-level jobs for young workers. Link
• EU AI Act — Implementation Timeline. Link
• The White House (2025). Ensuring a National Policy Framework for Artificial Intelligence (Executive Order, 12/11/2025). Link
• China Law Translate (2025). Measures for Labeling of AI-Generated Synthetic Content. Link
• Mayer Brown (2026). Administration Policies on Advanced AI Chips Codified. Link

Updated: 2026-06-13. This article synthesizes estimates from various sources; figures are projections and subject to change. All statistics are sourced for reader verification.