AI Industrial Policy
AI industrial policy is the contemporary application of strategic state intervention to the technology that is reshaping global economic structure. The leading practitioners — the United States, China, the European Union — deploy a comprehensive toolkit: direct subsidies for compute infrastructure, public funding of basic research, export controls on advanced chips, technology transfer restrictions, government procurement preferences, and regulatory frameworks designed to advantage domestic firms. Chang's framework illuminates what makes AI industrial policy distinctive and what makes it familiar. It is distinctive because the technology operates at unprecedented scale and speed. It is familiar because the structural pattern — wealthy nations deploying strategic intervention while preventing developing nations from doing the same — is the consistent feature of every previous round of industrial development. The question for the next decade is whether developing nations will be allowed the policy space to build domestic AI capability, or whether the rules being written now will lock them out of producing what they will be required to consume.
In the AI Story
The American AI industrial policy is conducted under multiple banners: the CHIPS Act for compute infrastructure, the Department of Defense for foundational research, the Department of Energy for high-performance computing, NIST for standards-setting that shapes global norms, and the Commerce Department for export controls on advanced chips. The total public investment running through these channels dwarfs the entire research budgets of most developing countries — and the gains, captured by private firms, are presented as triumphs of American entrepreneurship.
Chinese AI industrial policy is more explicitly state-directed but structurally similar in scope. The 2017 New Generation AI Development Plan committed enormous public resources to building domestic AI capability across research, infrastructure, application development, and talent cultivation. The plan's results have been mixed but its ambition is unmistakable: China intends to be a producer rather than a consumer of frontier AI capability, and is willing to deploy state power on the scale required to achieve that objective.
European AI industrial policy operates through different mechanisms — the AI Act's regulatory framework, the European Innovation Council's research funding, the InvestEU program's infrastructure financing — but pursues the same strategic objective: building European capability sufficient to avoid total dependence on American or Chinese systems. The EU's approach is more rule-based and less subsidy-heavy than the American or Chinese approaches, but it is industrial policy nonetheless.
The developing world faces a fundamentally different choice set. Without the resources to mount comparable interventions, and operating within rules that the leading nations have written to constrain the policy options available to others, developing nations face genuine difficulty in building domestic AI capability. The conventional prescription — adopt the leading tools, integrate them into the economy, compete on existing comparative advantage — is the contemporary equivalent of the Washington Consensus advice that produced two lost decades of development.
Origin
The phrase 'AI industrial policy' is recent, but the practice it describes has been operational since at least the 1960s when DARPA began funding the research that eventually produced the internet, neural network research, and the algorithmic foundations of modern AI. The contemporary visibility of AI industrial policy reflects not its novelty but its emergence from invisibility — what was previously presented as basic research and free-market innovation is increasingly recognized as the deliberate state-led building of strategic capability.
Chang's engagement with AI industrial policy applies his decades of work on developmental state theory to the most strategic technology of the contemporary era. The framework's predictive power has been substantial: the AI ecosystem has developed precisely along the lines that the developmental state literature would have predicted — concentration in nations that practice strategic intervention, marginalization of nations that do not.
Key Ideas
Comprehensive toolkit. Subsidies, research funding, infrastructure investment, export controls, procurement preferences, regulatory frameworks — deployed in coordination by the leading nations.
Strategic asymmetry. Wealthy nations practice AI industrial policy aggressively while constraining the policy space available to developing nations through trade agreements and conditionality.
Producer vs consumer divide. The fundamental developmental question is whether a country participates in the AI economy as producer or consumer — and the answer is determined by the presence or absence of effective industrial policy.
Recurring pattern. The contemporary AI ecosystem reproduces the structural pattern of every previous round of industrial development — strategic intervention in the lead nations, market dependency for the followers.
Appears in the Orange Pill Cycle
Further reading
- Mariana Mazzucato, The Entrepreneurial State: Debunking Public vs. Private Sector Myths (Anthem Press, 2013).
- Daniel Zhang et al., The AI Index 2024 Annual Report (Stanford HAI, 2024).
- Jeffrey Ding, Technology and the Rise of Great Powers (Princeton University Press, 2024).
- Anu Bradford, Digital Empires: The Global Battle to Regulate Technology (Oxford University Press, 2023).