How Asia Is Rewriting the Rules of Technology Power – E-International Relations

Artificial intelligence has become a primary axis of geopolitical competition. Yet most comparative analyses of the technology race stop at input metrics: GPU counts, investment figures, and patent filings. These are proxies, not explanations. What matters for international relations scholars, policymakers, and corporate strategists is a harder question: why are different political economies converging on different adoption strategies, and what do those strategies reveal about deeper assumptions regarding state capacity, risk tolerance, and the relationship between technological capability and national power? The divergence between Asia and the European Union is not primarily a story about investment gaps or regulatory philosophy in the abstract. It is a story about the purpose AI is being asked to serve. In Asia, AI is framed as a coordination problem that the state must solve. In Europe, it is framed primarily as a liability problem that the state must manage. That framing difference has structural consequences that compound over time, and the empirical evidence from 2024 to 2026 makes those consequences increasingly legible.

The Global Adoption Landscape: What the Data Now Shows

Before examining how different states are responding, it is worth establishing what the evidence shows about the current state of AI diffusion. The Stanford AI Index 2026 reports that generative AI reached 53 percent population adoption globally within three years — faster than the personal computer or the internet. Organizational adoption, as measured by McKinsey’s 2025 State of AI survey, is at 88 percent of organizations using AI in at least one business function, up from 78 percent the prior year. Global corporate AI investment more than doubled in 2025 to $581.7 billion, with private investment alone reaching $344.7 billion. These headline figures are striking, but they mask a structural divergence that is the real story: population-level adoption varies from 61 percent in Singapore to 28.3 percent in the United States. In comparison, the EU27 enterprise average sits at 20 percent (Eurostat 2025) — and that aggregate conceals a chasm between large firms and small ones that is analytically central to the European problem.

The productivity case for closing that adoption gap is increasingly supported by empirical evidence. OECD experimental studies, reviewed in a July 2025 research paper, find that individuals in customer support, software development, and consulting have seen average productivity gains ranging from 5 percent to over 25 percent with the integration of generative AI. Macroeconomic evidence is beginning to confirm aggregate effects: research published in April 2026 found that the rise in frequent AI users across occupations — from roughly 12 percent in mid-2024 to 26 percent by late 2025 — corresponds to approximately 1.4 to 2.8 percent higher real output, or about one to two percentage points of annualized growth. These are not speculative projections; they are measured outcomes. The economies that close the adoption gap earliest will realize those gains first, and in a compounding fashion. Figure 1 maps the current state of enterprise adoption across the jurisdictions examined in this article.

Three Models of State-Led AI Adoption

The five Asian economies examined here — China, South Korea, Japan, India, and Singapore — do not share a single model. They share a common premise: AI adoption will not occur at the required scale or speed without deliberate state intervention to accelerate demand. The form that intervention takes differs significantly and maps onto structural features of each political economy.

The first model is mandate-led diffusion. China’s “AI Plus” guideline, issued by the State Council in August 2025, sets a target of 90 percent penetration of intelligent terminals and AI agents across six designated sectors by 2030. What is analytically significant here is not the ambition of the number but the logic behind it. Beijing is treating AI diffusion the way it previously treated electrification or broadband rollout: as a coordination problem in which market mechanisms alone will underprice adoption because individual actors cannot capture the full social return. The solution is to legislate adoption rates, not merely subsidise inputs. The Trivium China analysis notes this mirrors the structural logic of the 2015 “Internet Plus” initiative. The results are evident: China’s AI user base more than doubled in the first half of 2025, reaching 515 million users, with AI adoption growing at 36.5 percent in six months following the launch of DeepSeek-R1 in January 2025 (CNNIC, cited in AI News). Enterprise-level adoption stands at approximately 58 percent according to IBM’s 2025 Global AI Adoption Index. China also leads globally in AI patent applications, with 38.58 percent of the worldwide total as of April 2025. The RAND Corporation’s June 2025 analysis concludes that China’s industrial policy will likely accelerate its progress particularly through subsidised compute and application deployment, with Chinese models now closing the performance gap with top US models to just 2.7 percent on the Arena Elo benchmark (Stanford HAI 2026).

The second model is statute-plus-capital, best represented by South Korea. The Framework Act on the Development of Artificial Intelligence, in force since January 2026, made Korea the first country to consolidate governance, industrial policy, and risk management into a single statute. Korea paired this with a tripling of the national AI budget to 10.1 trillion won (approximately $6.94 billion) in a single fiscal year, alongside the deployment of 260,000 Blackwell GPUs across Samsung, SK Hynix, Hyundai, Naver, and government infrastructure. The Stanford AI Index 2026 notes that South Korea now leads the world in AI patents per capita—a measure of innovation density that reflects the depth of industrial integration. The distinctive feature of this model is simultaneity: Korea is not sequencing regulation before deployment or capital before rules. It is running both tracks in parallel, using the statute as a coordination mechanism and the capital commitment as a credibility signal to private actors.

The third model is subsidy-as-infrastructure, exemplified by India. The IndiaAI Mission, approved in March 2024, treats compute the way earlier development states treated electricity: as a public good whose underproduction by markets creates a structural ceiling on downstream activity. The program has deployed over 38,000 GPUs — including 1,050 Google Trillium TPUs — against an initial target of 10,000. Startups and academics can access H100-class compute at approximately 65 rupees ($0.72) per hour, the cheapest subsidised rate in the world. Enterprise adoption has responded: IBM’s 2025 index places India at 57 percent enterprise AI adoption. Minister Ashwini Vaishnaw reported in February 2026 that committed AI-related investment in India stands at $90 billion, with projections that it will exceed $400 billion across the AI stack within two years. India is not betting on regulatory sophistication; it is betting that removing the compute bottleneck will generate compounding returns — and the early adoption data suggest this bet is paying off.

Japan and Singapore: The Catch-Up and the Precision Case

Japan represents a fourth model — capital-intensive catch-up — that is less a coherent strategic choice than a response to a documented failure to adopt. The Ministry of Internal Affairs and Communications’ 2025 White Paper revealed generative AI usage among the Japanese public at 26.7 percent in 2024, compared with 68.8 percent in the United States and 81.2 percent in China. More telling structurally is the OECD’s G7 AI adoption analysis, which places AI use in core business functions at just 1.9 percent among Japanese firms in 2024, compared to 6.1 percent in the United States. This is the deepest measure of productive integration, and it reveals a chasm that headline investment figures obscure. Tokyo’s response has been to deploy capital: NTT is committing $59 billion through 2027, and SoftBank has tied itself to OpenAI’s Stargate project with over $40 billion in commitments. Japan’s trajectory tests whether raw capital deployment can substitute for the demand-side coordination mechanisms that China and Korea are using — or whether structural adoption barriers (risk-averse corporate culture, lifetime employment norms) are not capital-soluble.

Singapore is the precision case — a fifth model, governance by measurement. The National AI Strategy 2.0, updated in May 2026, is notable less for its ambition than for its granularity. The Stanford AI Index 2026 places Singapore at 61 percent population-level generative AI adoption, the highest globally, alongside the UAE. Non-SME enterprise adoption reached 62.5 percent and SME adoption 14.5 percent in 2024 (IMDA), with the National AI Impact Programme targeting 10,000 enterprises over three years. What makes Singapore analytically distinctive is that its small scale enables a feedback loop that larger economies cannot replicate: the government can measure what it deploys, observe uptake in near real time, and adjust program design accordingly. Singapore functions as a living laboratory, where its measurement infrastructure produces policy insights that larger states cannot derive from aggregate statistics.

The EU’s Structural Misjudgment: Evidence from the Adoption Data

Set against these five approaches, the European Union’s position reflects a structurally different theory of what AI governance is for — and the evidence from 2024 to 2025 makes this difference starkly measurable. Eurostat’s December 2025 release records EU27 enterprise AI adoption at 20 percent, up from 13.5 percent in 2024 — a 6.5-percentage-point increase representing genuine momentum. Denmark leads at 42 percent, Finland at 37.8 percent, Sweden at 35 percent. But the aggregate conceals a structural problem that headline growth obscures: adoption is sharply stratified by firm size. As Table 2 shows, large enterprises (250 or more employees) reached 55 percent AI adoption in 2025, while small enterprises (10 to 49 employees) reached only 17 percent. The 38-percentage-point gap between large and small enterprises did not narrow in 2025; it widened by more than 7 percentage points (Figure 2).

This size stratification is the empirical heart of the EU’s adoption problem and directly relates to the AI Act. Advisory market estimates for high-risk AI compliance under Annex III place initial costs for SME providers between €200,000 and €500,000, phased across 2025 to 2027 (Figure 5). Even with the Act’s explicit SME provisions — proportional fees, sandbox access, simplified documentation — the enforcement obligations for high-risk deployers take effect August 2026, and 12 of 27 member states had missed the deadline to appoint competent national authorities. The European Commission’s own Digital Omnibus proposal, introduced in November 2025, acknowledges the problem by targeting a 25 percent overall compliance burden reduction and 35 percent for SMEs by 2029. Over 60 percent of EU startups now prioritise low-risk AI applications specifically to avoid compliance exposure. The governance architecture is shaping innovation trajectories before it is even fully enforced.

The investment picture compounds the concern. Europe attracted only $20.9 billion in private AI investment in 2025, compared with $285.9 billion in the United States — a 13.7-to-one ratio (Figure 3). China’s $12.4 billion in private investment substantially understates its effective total once government guidance funds are included; Stanford HAI estimates $912 billion in guidance funds deployed across industries between 2000 and 2023. The Commission’s own June 2025 State of the Digital Decade report concedes that at the current pace the bloc will not reach its 2030 targets until around 2040. An analysis by Public First estimates the EU is on track to unlock approximately 1.3 trillion euros of projected digital value by 2030, leaving up to 1.5 trillion euros unrealized.

AI Adoption as a Collective Action Problem

The comparative evidence supports reading AI adoption as a collective action problem of the type identified in the economic geography literature on industrial policy. The core challenge is not that individual firms lack access to information about AI — the technology is visible and its applications documented. The challenge is that adoption creates positive externalities (through data generation, workforce upskilling, and process knowledge that spills across firm boundaries) that individual firms cannot fully capture. Market-rate adoption will therefore systematically undershoot the socially optimal level. State intervention is not a distortion; it is a correction for a structural market failure.

What differs across the cases is the mechanism chosen to address this problem. China uses administrative mandates. Korea combines a statutory framework with direct capital deployment in parallel. India uses price subsidies on scarce inputs to lower the adoption threshold. Singapore uses precision measurement and targeted programmatic support. Japan is attempting to use private capital concentration as a substitute for demand-side coordination. The EU uses a compliance architecture—not a coordination mechanism at all, but a risk-allocation mechanism. Compliance architecture tells firms what they cannot do; it does not solve the adoption coordination problem that is actually binding. Critically, the OECD’s experimental evidence finds that productivity gains from AI require meaningful human-AI collaboration and task-fit. Building that capability requires adoption first. You cannot build organizational AI capability through compliance frameworks alone.

This framing connects to longer-running debates in comparative political economy. The literature since Rodrik (2004) on self-discovery and since Hausmann and Hidalgo (2011) on economic complexity suggests that the state’s most valuable contribution to industrial transformation is building shared capabilities — skills, infrastructure, standards — that make firm-level investment viable. The Asian cases surveyed here are each, in different ways, investing in those shared capabilities. The EU, whatever its regulatory sophistication, has not yet constructed the demand-side equivalent.

Implications for Firms and Policymakers

The practical stakes are considerable, and the productivity evidence gives them empirical weight. Studies summarized in the Stanford AI Index 2026 report productivity gains of approximately 14-15 percent in customer support, 26 percent in software development, and 50 percent in marketing output from AI integration (Figure 4). These are task-level gains; their aggregation across the economy depends on the breadth of adoption. If, by 2030, China reaches 90 percent AI penetration and the EU arrives at 30 to 40 percent, the productivity differential will not be recoverable through regulation. Early adopters accumulate process knowledge, generate better training data, and develop organizational capabilities that later adopters cannot simply purchase.

For multinational corporations operating across these jurisdictions, the asymmetry creates both risks and structural arbitrage opportunities. Firms developing AI capabilities in Asia’s more adoption-accelerating regulatory environments will enter the EU market with capabilities that EU-based competitors have not had an equivalent opportunity to build. The AI Act’s requirements apply to systems deployed in the EU, regardless of where they were developed—meaning the compliance burden falls on the European operations of firms that built their capabilities elsewhere. This is an ironic inversion of protective intent: the regulation taxes adoption by incumbents without preventing capability accumulation by foreign entrants.

The implication for EU policymakers is not to discard the AI Act. The risk-stratification logic addresses genuine concerns about fundamental rights and safety in high-stakes applications. The implication is about sequencing and the distribution of compliance costs. What is structurally missing is the demand-side complement: a European equivalent of India’s subsidised compute for SMEs, Korea’s adoption-focused budget commitment, or China’s sector-specific penetration mandates. The EU’s chip ambitions — targeting 20 percent of global production value by 2030 — are serious on the supply side. Building that supply without building the enterprise adoption base is analogous to constructing a motorway network without connecting it to the towns it is meant to serve.

Conclusion

The AI race is not primarily a competition over who has the most sophisticated regulations or the largest nominal investment commitments. It is a competition over who solves the adoption coordination problem first, and the 2024 to 2025 data make the contestants visible. China’s enterprise adoption at 58 percent, India’s at 57 percent, Singapore’s population-level adoption at 61 percent, and South Korea’s leadership in patent density are not coincidences. They are the measurable outputs of deliberate demand-side policy. The EU’s 20 percent enterprise average, its 38-percentage-point large-small firm gap, and the $20.9 billion in private investment, compared with the US’s $285.9 billion, are equally measurable outputs of a different policy orientation. Both orientations respond to real challenges. But they are not responding to the same challenge with equal urgency. The liability problem is real; the coordination problem is more binding. Brussels is not so much mistaking the rulebook for the game as it is prioritizing the wrong chapter. That is a more tractable problem than structural underinvestment — but only if it is diagnosed correctly, and only if the window to correct the sequencing has not already closed.

Figures

Tables

Table 1. AI adoption models compared

Source: compiled by the authors from cited sources. IBM = IBM Global AI Adoption Index 2025; CNNIC = China Internet Network Information Center; MIC = Ministry of Internal Affairs and Communications (Japan); Eurostat ICT Enterprise Survey 2025; Stanford HAI AI Index 2026.

Table 2. EU enterprise AI adoption by size class (2024 and 2025)

Source: Eurostat ICT Enterprise Survey 2025 (isoc_eb_ai).

Further Reading on E-International Relations