Sovereign AI Will Extend the Global Capital Expenditure Cycle While Driving Further Divergence Between US, Chinese, and Regional Technology Ecosystems

Overview

The report examines "sovereign AI" from the perspectives of national security, industrial policy, and capital markets. Its core conclusion is that AI is becoming strategic infrastructure, and countries will rely more heavily on domestic capabilities and trusted partners, creating a more fragmented and capital-intensive global ecosystem. This trend broadly supports a larger and more persistent AI capital expenditure cycle, but expanding technology controls will also alter market access, costs, and the competitive landscape.

Core views

The report defines sovereign AI as a country's ability to maintain meaningful control over computing power, data, talent, energy, models, supply chains, and technology stacks, while reducing dependence on technology providers, jurisdictions, and infrastructure deemed strategically vulnerable. This is not equivalent to pursuing absolute self-sufficiency across the entire technology stack; Morgan Stanley believes a more realistic model is to develop a degree of domestic capability while selectively relying on trusted partners. As AI becomes a source of national economic and geopolitical power, countries are prioritizing security and resilience over pure economic efficiency, even at the cost of higher expenses, duplicated construction, or reduced market access. This shift has two major policy consequences. First, countries' development of local and regional capabilities will make the global AI system more fragmented and capital-intensive. The report expects US AI infrastructure construction to reach $860bn in 2026 and believes the federal government will continue supporting development through permitting reform, expanded use of Defense Production Act tools, subsidies, loans, and incentives similar to the CHIPS Act. Sovereignty requirements will also increase demand for data centers, reserve capacity, power generation, transmission and distribution, semiconductors, networking, cloud services, and on-premises deployment platforms. Infrastructure will therefore no longer be freely substitutable worldwide, and some regions will require parallel technology stacks. On this basis, Morgan Stanley believes sovereign AI is more likely to expand and extend the global AI capital expenditure cycle than disrupt it. Second, US-China "de-risking" may expand from advanced chips and semiconductor equipment to AI training, hosting, distribution, and access infrastructure. Digital models can spread through downloads, APIs, third-country hosting, or offshore cloud facilities, making enforcement more complex than restrictions on physical products. Chip and equipment controls constrain the ability to develop future systems, while model, cloud, and distribution controls restrict access to existing capabilities; the report expects policymakers to pursue both types of measures simultaneously. Potential tools include restrictions on offshore data centers and cloud capacity, customer and end-user verification, reporting obligations for advanced computing power, restrictions on hosting or distributing designated foreign models, limitations on model marketplaces and app stores, and procurement and cybersecurity standards. The US policy base case is not complete closure. Morgan Stanley expects policy to adopt a gradual framework of "open below the frontier, controlled frontier capabilities": preserving a degree of open-weight models to support diffusion, development, and access, while more strictly protecting the intellectual property of leading US large-model developers and combating large-scale automated model extraction. Specific approaches could include placing certain Chinese AI developers on restricted lists, prohibiting US companies from using their models, and using the Commerce Department's Entity List, reviews of the information and communications technology and services supply chain, Treasury sanctions, federal procurement exclusions, and other administrative tools to restrict related transactions. This approach could avoid comprehensive escalation while protecting leading US laboratories and continuing to allow open models from non-designated companies to enter the market. Administrative measures are reversible, so congressional legislation offers another path to making restrictions more durable. The report focuses on whether the AI OVERWATCH Act and MATCH Act can be incorporated into the FY27 National Defense Authorization Act: the former focuses on exports of advanced AI chips to foreign adversaries, while the latter tightens controls further upstream on semiconductor manufacturing equipment and components and promotes coordination between the United States and allies such as the Netherlands. Neither measure directly prohibits Americans from downloading Chinese models, but they would restrict China's access to the chips and equipment needed to train future frontier systems, complementing restrictions on models, applications, procurement, and hosting. The report believes some provisions could be enacted through the FY27 legislation expected to pass before the end of 2026, but stresses that the legislative process remains uncertain and that negotiations between the House and Senate will be critical. The overall direction of policy remains toward tighter restrictions. In April 2026, US officials alleged that Chinese entities were conducting industrial-scale extraction of model capabilities; on July 22, US Treasury Secretary Scott Bessent stated that sanctions and Entity List measures could be used if Chinese companies exploited stolen US technology. Morgan Stanley accordingly raised its assessed probability of restrictions on the use of Chinese AI models, while still viewing a comprehensive ban as uncertain. The US focus is expected to include advanced chips, semiconductor equipment, cloud and computing-power access, procurement, model distribution, and allied coordination. China, meanwhile, is more likely to pursue sovereign objectives through regulatory approvals, local model integration, data and cybersecurity rules, localized procurement, platform and app-store access, and administrative restrictions on foreign providers. Domestic AI development in the United States continues to enjoy bipartisan strategic support. The report believes AI policy is likely to continue advancing regardless of the outcome of the 2026 midterm elections, because restricting domestic data-center construction would weaken US competitiveness relative to China. A nationwide ban or moratorium on data-center construction is therefore unlikely in the near term; election outcomes are more likely to affect the pace and geographic distribution of deployment than its overall direction. Energy availability, permitting, labor, and public opposition are becoming constraints. Targeted regulation may emerge around security, data governance, and infrastructure siting, but comprehensive and unified regulation remains unlikely in the near term. China's sovereign AI strategy emphasizes localization, independent capabilities, public procurement, and application diffusion, while building ecosystems overseas that are compatible with Chinese technology. The report argues that China does not necessarily need to lead at the technological frontier to create a commercially meaningful AI market. Its competitive focus is shifting from training to inference, from technology itself to applications, and from potential to actual profitability, while using deployment speed, cost efficiency, and system-level integration to drive AI adoption in industrial and consumer settings. Divergence between the US and Chinese ecosystems could therefore make the global addressable market for Chinese AI solution providers larger than the market currently recognizes, particularly in markets such as the Global South where standards are not yet fully established. Other regions are taking different approaches. Europe aims to increase strategic autonomy in computing power, cloud services, and semiconductors. Its long-term data-center outlook remains strong, but insufficient financing capacity and scale pose material challenges. ASEAN generally does not pursue self-sufficiency across the entire technology stack, instead combining domestic infrastructure control with cooperation among trusted partners; the report identifies Singtel and Indosat as important participants in ASEAN sovereign AI. South Korea is addressing weak links through proactive government policy and public-private cooperation. SK Group, Samsung SDS, and KT have announced major AI data-center investment plans, although access to land, power, and water may constrain the pace of construction. At the market level, the report views the principal beneficiaries as semiconductors, data centers, networking, power, cloud services, sovereign deployment platforms, and infrastructure software, cybersecurity, and IT service providers capable of meeting compliance, security, and governance requirements. If low-cost Chinese models are restricted, US model providers could gain market share and pricing power, although higher inference costs could also slow enterprise adoption. In robotics and physical AI, restrictions may protect non-Chinese robot manufacturers but would raise input costs because supply chains remain dependent on China, while also increasing the risk of countermeasures. Morgan Stanley includes "AI Sovereignty and Semiconductor Localization" in its global thematic map and measures companies' business exposure to this trend across more than 3,800 covered stocks. The report states that the theme as a whole is among the fastest-growing thematic groups and commands a valuation premium. However, technology restrictions will create winners and losers across different parts of the technology stack, so thematic growth does not mean all related companies will benefit equally. Finally, the report compares AI historically with the previous five waves of general-purpose technological innovation. It concludes that AI could increase output per worker once organizational structures adjust in parallel. Although past innovation waves displaced jobs, they did not evolve into large-scale permanent technological unemployment. However, Morgan Stanley's labor-market tracker shows that occupations highly exposed to AI task substitution currently have a residual unemployment rate 0.5 percentage points above the level explainable by the economic cycle. The report also warns that current AI infrastructure investment resembles the railway and telecommunications construction booms and may be accompanied by investment exuberance, financial excess, and volatility. Income and wealth inequality are already at a 125-year high, and AI diffusion could widen the gap further.

Analysis framework

The report first defines sovereign AI in terms of control over computing power, data, energy, talent, models, and supply chains, and then compares the policy objectives and implementation paths of the United States, China, Europe, ASEAN, and South Korea. It subsequently divides policy tools into chip and equipment restrictions, cloud and computing-power access, model hosting and distribution, procurement, and cybersecurity rules, while constructing a "middle path" as the US policy base case. The market analysis then traces transmission along the technology stack to semiconductors, data centers, networking, power, cloud services, software, and robotics, and identifies relevant companies by mapping thematic exposure across more than 3,800 covered stocks. The macroeconomic section combines historical innovation waves with labor-market tracking to assess implications for productivity, employment, investment volatility, and income distribution.

Methodology notes

• Industry/Sector Analysis FrameworkUpstream, Midstream, and Downstream Industry-Chain Transmission

  Transmission from sovereignty policies to the AI technology stack and infrastructure demand

  The report traces government requirements for control and security through chips, semiconductor equipment, data centers, networking, power, cloud services, software, robotics, and other segments to assess incremental demand and identify beneficiaries and adversely affected areas across the value chain.

• Quantitative/Factor/Portfolio Theory

  Global thematic map and company exposure mapping

  Morgan Stanley measures the business relevance of the "AI Sovereignty and Semiconductor Localization" trend across more than 3,800 globally covered stocks to identify companies with significant thematic exposure and compare the growth and valuation characteristics of the thematic group.

• Event-Driven Strategy and Behavioral FinanceEvent-driven analysis

  Analysis of policy tools, administrative pathways, and legislative timing

  Based on government statements, existing executive authority, and legislative windows such as the FY27 National Defense Authorization Act, the report analyzes how Entity List designations, sanctions, procurement exclusions, chip controls, and model restrictions could be implemented, while explicitly noting uncertainty arising from congressional negotiations.

• Cycle and Business Conditions Framework

  Analogy with five historical waves of innovation

  Using historical general-purpose technology construction booms such as railways and telecommunications as reference points, the report explains that AI could simultaneously produce productivity gains, changes in job structures, a capital expenditure boom, financial excess, and volatility, without assuming that history will repeat exactly.

• Macroeconomic framework

  AI diffusion and residual unemployment tracking in the labor market

  The report uses a labor-market tracking tool to separate unemployment changes explainable by the economic cycle from additional unemployment in occupations highly exposed to AI task substitution, currently identifying a residual unemployment rate of 0.5 percentage points.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Singtel

  The report identifies it as a sovereign AI leader in ASEAN and highlights its GPU-as-a-service offering, collaboration with NVIDIA to build a sovereign AI center of excellence, and replication of its operating model across ASEAN.

  Strengths

  GPUaaS capabilities, a sovereign AI center of excellence, and potential for replication across ASEAN.

  Comparison

  Named alongside Indosat as one of the report's sovereign AI leaders in ASEAN.

• Indosat

  The report views it as an important participant in developing Indonesia's sovereign AI capabilities.

  Strengths

  Direct involvement in developing Indonesia's domestic sovereign AI capabilities.

  Comparison

  Named alongside Singtel as one of the report's sovereign AI leaders in ASEAN.

• SK Group

  The report states that it has announced major AI data-center investment plans and is one of the principal participants in South Korea's public-private sovereign AI ecosystem.

  Strengths

  Major AI data-center investment initiatives.

  Comparison

  Participating alongside Samsung SDS and KT in the development of major AI data centers in South Korea.

  Risks

  Access to land, power, and water may constrain AI data-center construction.

• Samsung SDS

  The report states that it has announced major AI data-center investment plans and is one of the principal participants in South Korea's sovereign AI strategy.

  Strengths

  Major AI data-center investment initiatives.

  Comparison

  Participating alongside SK Group and KT in the development of major AI data centers in South Korea.

  Risks

  Access to land, power, and water may constrain AI data-center construction.

• KT

  The report states that it has announced major AI data-center investment plans and is one of the principal participants in South Korea's sovereign AI strategy.

  Strengths

  Major AI data-center investment initiatives.

  Comparison

  Participating alongside SK Group and Samsung SDS in the development of major AI data centers in South Korea.

  Risks

  Access to land, power, and water may constrain AI data-center construction.

Key data

• Scale of US AI Infrastructure Construction$860bnThe report's stated scale of US AI infrastructure construction in 2026.
• Global Thematic Mapping CoverageMore than 3,800 stocksThe range of Morgan Stanley's globally covered stocks used to measure exposure to the AI sovereignty and semiconductor localization theme.
• Residual Unemployment Rate in Occupations Highly Exposed to AI Task Substitution0.5 percentage points higherRelative to the unemployment rate explainable by changes in the economic cycle.
• Level of Income and Wealth Inequality125-year highThe report notes that AI diffusion could widen the gap further from its current elevated level.
• Timing of the FY27 National Defense Authorization ActExpected to pass before the end of 2026The report believes some provisions of the AI OVERWATCH Act and MATCH Act could be incorporated, but the legislative outcome remains uncertain.

Impact & implications

The report believes the global AI market will shift from a single, freely substitutable system toward a structure in which US, Chinese, and regional ecosystems coexist. Additional domestic capabilities, redundant facilities, and trusted supply chains will expand total infrastructure demand, but market access, compliance capabilities, energy conditions, and supply-chain positioning will determine the actual benefits for individual companies. US model providers and compliance-oriented infrastructure companies may gain opportunities, while companies dependent on low-cost models or Chinese inputs may face higher costs and pressure from countermeasures.

Risks

• Further expansion of technology restrictions could intensify global market fragmentation and create clear winners and losers across different parts of the AI technology stack.
• Restricting low-cost Chinese models could increase inference costs, thereby slowing the pace of enterprise AI adoption.
• Robotics and physical AI supply chains remain dependent on China, and restrictions could raise input costs for non-Chinese manufacturers and trigger countermeasures.
• Europe's sovereign AI development faces material financing and scale challenges.
• Energy availability, permitting, labor, and public opposition may constrain US AI infrastructure deployment.
• AI data-center development in South Korea may be constrained by shortages of land, power, and water.
• Provisions related to the FY27 National Defense Authorization Act remain subject to legislative uncertainty and negotiations between the two chambers of Congress.
• An AI investment boom resembling the railway and telecommunications construction booms could be accompanied by financial excess, cyclical volatility, and capital misallocation.
• AI task substitution could increase unemployment pressure in some occupations and further widen income and wealth inequality.

What to watch

• Monitor whether the FY27 National Defense Authorization Act incorporates provisions from the AI OVERWATCH Act and MATCH Act, as well as the outcome of negotiations between the House and Senate.
• Monitor whether the United States expands restrictions from chips and equipment to the use of Chinese models, offshore computing power, cloud hosting, procurement, model marketplaces, and app stores.
• Monitor whether the United States ultimately implements a tiered policy of "open below the frontier, controlled frontier capabilities" rather than a comprehensive model ban.
• Monitor how energy, permitting, labor, and public opposition affect the pace and geographic distribution of US data-center deployment.
• Monitor whether China's local model integration, domestic procurement, and application diffusion can translate into actual profitability and overseas ecosystem expansion.
• Monitor whether Europe can address the financing and scale requirements for developing sovereign computing power, cloud services, and semiconductors.
• Monitor the residual unemployment rate in occupations highly exposed to AI task substitution, the pace of enterprise AI adoption, and changes in income distribution.