主权人工智能的经济学：平衡亚太地区的自主性、创新和增长（英）

RESEARCH FORTHE AIADOPTIONINITIATIVE ABOUT OXFORDECONOMICS Oxford Economics was founded in 1981 as a commercial venture with Oxford University’s businesscollege to provide economic forecasting and modelling to UK companies and financial institutionsexpanding abroad. Since then, we have become one of the world’s foremost independent global advisoryfirms, providing reports, forecasts and analytical tools on more than 200 countries, 100 industries sectors,and 8,000 cities and regions. Our best-in-class global economic and industry models and analytical toolsgive us an unparalleled ability to forecast external market trends and assess their economic, social andbusiness impact. Headquartered in Oxford, England, with regional centres in New York, London, Frankfurt, and Singapore,Oxford Economics has offices across the globe in Belfast, Boston, Cape Town, Chicago, Dubai, Dublin,Hong Kong, Los Angeles, Mexico City, Milan, Paris, Philadelphia, Stockholm, Sydney, Tokyo, and Toronto.We employ700 staff, including more than450 professional economists, industry experts, and businesseditors—one of the largest teams of macroeconomists and thought leadership specialists. Our globalteamis highly skilled in a full range of research techniques and thought leadership capabilities from econometricmodelling, scenario framing, and economic impact analysis to market surveys, case studies, expert panels,and web analytics. Oxford Economics is a key adviser to corporate, financial and government decision-makers and thoughtleaders. Our worldwide client base now comprises over3,000 international organisations, including leadingmultinational companies and financial institutions; key government bodies and trade associations; and topuniversities, consultancies, and think tanks. TABLE OF CONTENTS SECTION 1.QUANTIFYING THE ECONOMIC IMPACT OF SOVEREIGN AI MEASURES.........3SECTION 2.DOMESTIC INFRASTRUCTURE REQUIREMENTS......................................................5SECTION 3.INVESTMENT COSTS.......................................................................................................9SECTION 4.HOW SOVEREIGNTY RULES INFLUENCE AI ADOPTION......................................16SECTION 5.LINKING ADOPTION TO PRODUCTIVITY AND GDP............................................20 SECTION 1.QUANTIFYING THEECONOMIC IMPACT OFSOVEREIGN AI MEASURES Sovereign AI measures aim to protect sensitive information and ensure reliable access to digital services.However, these policies also influence how organisations deploy AI, the scale of investment required indomestic infrastructure, and the speed at which AI can drive productivity and economic growth.This appendixprovides a structured framework for assessing the impactsof sovereignty-related restriction levelsas shownbelow: The appendixis organised intofourmain sections: •AIadoption andinfrastructuredemand-Weestimate current and future AI adoption using harmoniseddatasets and diffusion modelling. These adoption trajectories are then linked to infrastructurerequirements, quantifying how sovereignty rules affect the extent of onshoring and the scale of domesticcapacity needed. •Investmentrequirements-We assess the costs of meeting these infrastructure and capability needsunder different sovereignty scenarios. This includes capital and operational expenditure for computeinfrastructure,AI tooling, software andapplication investments,andthe resources required to train a skilleddeveloper base for AI tooling and applications.•Estimating howsovereigntyrulesinfluence AIadoption,productivity andGDP-We analyse howrestrictiveness,measured usingcomposite indicesofdigital openness and foreign‑ownershiplimitationstogetherwith delaysassociated withinfrastructure development, affectsAI adoption rates.•Linking adoptionto productivity and GDP-Finally, we link changes in adoption toproductivity andeconomic growth using Oxford Economics’ AI productivity framework. This approach converts task-levelautomation potential into sectoral productivity gains and long-term impacts on total factor productivity(TFP) and GDP. SECTION 2.DOMESTICINFRASTRUCTUREREQUIREMENTS The first stageof our analysis estimatestheAI-related infrastructure each country requires to meet its domesticAI demandunder different sovereignty rules. In particular,itaims to quantify how sovereignty rules affect theextent of onshoring of AI-related infrastructure. We translate projected adoption into infrastructure demand by allocating the global stock of AI-relatedcompute infrastructure in 2024 across countries based on two factors: AI adoption rates and economic size(proxied by real GDP). This provides country-specific estimates of current infrastructure needs. Futureinfrastructure demand is then projected in line withmodelledchanges in adoption. CURRENTAIADOPTION We estimate currentAIdemand using firm-level adoption rates.Following the OECD, adoption is defined asthe use of AI by firms to improve the production of goods and services, as thi