What national AI plans get wrong and how to fix them

Almost every nation on Earth has an artificial intelligence (AI) plan in motion. Many plans attempt to replicate a generalized “AI stack,” which includes compute, chips, foundation models, and a generic regulatory framework. However, this approach misses the point. AI deployment should reflect a country’s existing strengths: its industries, workforce, institutions, and the role in global value chains.

Our discussions with policymakers, industry leaders, and researchers at the India AI Impact Summit and other events have underscored that trying to build fully sovereign AI stacks risks duplication and incompatible standards, reinforcing the need for countries to specialize in domains aligned with their strengths and to collaborate on evaluation, data, and deployment frameworks, standards, and governance.

The India AI Impact Summit put interoperability and cross-border cooperation at the center of its “impact” agenda. AI is not just large generative models, but rather “a sprawling field of interlocking techniques, tools, and capabilities” across a wide variety of domains and applications, as we’ve previously discussed. Similarly, each country is different, with an equally wide variety of endowments—unique mixtures of skills, production capabilities, service specializations, and institutional strengths. A one-size-fits-all approach to AI development and deployment does not fit this diverse global landscape.

Countries need to match AI capabilities to what they already do well, aligning AI domains and functions with their existing industrial capabilities (in goods and services), workforce skills, and institutional strengths, and then use these capabilities to diversify into adjacent industries, new goods and services, and higher-value AI applications. In short, AI cannot be deployed for its own sake but should enhance the real economy—to create jobs, boost productivity, and foster long-term competitiveness and growth. This growth requires a focus on “cognitive infrastructure“: the intelligence layer that connects data, human expertise, people, and systems across an economy, enabling more adaptive services and sustained competitiveness. Grounded in data, global patterns of specialization, and lessons from national AI plans, countries need to identify this cognitive infrastructure and harness it for local conditions.

Figure 1 shows the global distribution of AI activity, broken down by domain—from agritech and autonomous vehicles to cybersecurity and health care. This visualization reveals that AI innovation is not only geographically distributed, but also sectorally differentiated. The figure shows many AI hubs, each with its own sectoral focus.

Globally, the data in Figure 1 reveals a wide range of vertical AI funding priorities. Between 2014 and 2025, the largest AI private investment clusters include AI infrastructure, models, research, and governance (41.55% of total funding), data management and processing (9.16%), medical, health care, and pharmaceutical AI (6.48%), Internet of Things (4.24%), and cloud computing (2.99%). Sectors such as environmental tech (0.03%), workplace safety (0.02%), and transportation (less than 0.01%) receive relatively less funding, highlighting either underinvestment or underreporting in public-sector-aligned domains. These gaps can help governments identify market failures where policy can play a catalytic role.

Table A shows how countries are specializing across existing strengths. Among the major hubs shown in Figure 1, the U.S. spans a wide range of domains. Europe has deep clusters in enterprise software, mobility, and medical AI. India leans toward education and financial technology (fintech), while East Asia dominates in manufacturing and vision systems. The United States shows a broad and dense spread across clusters, while countries like France, Germany, Japan, and South Korea demonstrate concentration in domains tied to their traditional industrial strengths (e.g., manufacturing, mobility, robotics). China and India display heavy regional clustering in domains such as fintech, automation, and education technology (edtech).

These variations are not random; they reflect the economic DNA, or underlying assets and aspirations, of each region, including talent, skills, data resources, firms, and priorities for building institutional capacity, such as export growth, productivity, and resilience. In turn, this shapes cognitive infrastructure that should guide how national AI plans are framed. Countries tend to attract and build AI where they already have complementary capabilities, demand, or the regulatory or institutional capacity to deploy it. For instance, France’s strong presence in AI for logistics and smart cities could be strategically scaled with supporting infrastructure investments, while India’s AI activity in health and edtech suggests a targeted pathway for service export expansion.

Figure 2 complements this by showing verticals of AI funding flows across AI sectors and countries. This heatmap shows how different countries specialize in different AI application domains, based on cumulative investment amounts from 2014 to 2025. Larger circles indicate higher concentration of funding; smaller circles indicate lower levels. Figure 1 also makes clear that most countries, in fact, specialize in different types of AI that enhance their existing economies. For example, Estonia over-indexes on digital governance and health, while Brazil shows activity in agritech and logistics. Such variations should be seen as assets—not obstacles—and foundations for strategic planning. The clearest evidence here is that AI is not monolithic—each country has a distinct AI fingerprint. Examples include:

• The U.S. has widespread investments across nearly every cluster, reflecting both scale and breadth.
• India stands out in fintech, wellness, and edtech clusters.
• Japan concentrates heavily on robotics, manufacturing, and mobility-linked clusters.
• Germany and France specialize in AI for industrial automation, medtech, and legal tech.
• Smaller countries like Estonia, Kenya, and Singapore display deep specialization in only a few clusters, aligning with their strategic national priorities.

The key insight for policymakers is to build AI capacity to reinforce strengths, rather than in a vacuum. This should guide policymakers toward strategies that reveal a nation’s comparative AI advantage and therefore where the greatest opportunities lie for AI as an enhancement.

As AI is a lever for economic transformation, countries can embed AI into their existing strengths and plan realistic paths for diversification. Policymakers should treat AI not as a single stack to replicate, but as a mosaic of capabilities that intersect with goods, services, skills, and institutional capacity.

The sections below outline a practical roadmap for how countries can use AI to deepen what they already do well, and the capabilities that can support realistic, adjacent diversification over time.

AI for what you already do

To start, countries ask a simple question: What do we already do well—and how can AI amplify those strengths? At this first stage, AI is not about reinvention, but enhancement.

Norway, for example, can deploy robotics, advanced sensing, and predictive systems to make offshore energy extraction safer, more efficient, and less environmentally risky. Germany’s longstanding strength in automotive can extend into smart factories, autonomous logistics, and next-generation mobility systems. In each case, AI functions as a force multiplier improving productivity, safety, and quality within sectors where skills, supply chains, and institutions are already deeply embedded.

In both instances, AI strengthens existing comparative advantages rather than displacing them, laying out a stable foundation for future growth.

From today’s specializations to tomorrow’s growth

However, economic diversification doesn’t mean abandoning core strengths; it means upgrading and branching out from these. Our prior research shows that diversification tends to follow related pathways: Countries move into sectors that share capabilities, inputs, institutions, and knowledge with what they already produce. Countries can use AI systems to accelerate that process by improving performance in existing sectors and linking existing capacities to adjacent activities that reuse the same skills, data, infrastructure, and supply chains and create higher value. AI can accelerate that process by improving productivity in current sectors and revealing adjacent activities that reuse similar skills, data, supply chains, and institutional capacities.

In our earlier work, we modeled how AI specializations statistically co-occur with goods and services specializations over time, creating what we call a progression network. The key finding is that countries rarely leap into entirely unrelated sectors. Instead, new specialization emerges from sectors that are “nearby” in capabilities. AI does not erase path dependence; it clarifies the next steps along it.

Operationally, this logic involves two steps. First, countries should identify which technical AI capabilities, such as computer vision, natural language processing, or robotics, can directly raise productivity, quality, or safety in their current specializations. Second, policymakers should assess which adjacent sectors share skills, data, infrastructure, institutional capacity, and supply chains. Diversification over a five- to 10-year horizon is realistic because it matches existing workforce training cycles, supplier-development timelines, and the time needed to build standards, data pipelines, and deployment capacity.

Several country-specific AI applications illustrate such logic.

• Bangladesh has long specialized in garments and can enable smart textiles by embedding sensors and applying computer vision-based AI for quality control, traceability, and real-time performance monitoring, which would open pathways into technical fabrics and higher-value apparel services.
• Kenya’s agri-processing base naturally extends into AI-enabled supply chains, where data analytics and optimization tools improve storage, logistics, and market access while enabling adjacent services in agri-finance and digital trade facilitation.
• Indonesia can evolve from traditional shipping toward automated port and logistics operations, using machine learning to predict bottlenecks, coordinate multimodal transport, and support new services in maritime analytics and trade logistics.
• Mexico provides a complementary case of connected diversification. By building domestic capability in fintech, insurance technology, and cybersecurity, and by selectively procuring proven AI tools through partnerships and public-sector adoption, Mexico can embed secure payments, risk analytics, and automation into tourism, oil-linked services, and agri-food supply chains. That combination of domestic development and targeted adoption upgrades existing industries and creates feasible adjacent opportunities in logistics technology, industrial automation, and data-driven services beyond Mexico’s traditional export base.

The broader lesson is that diversification emerges from observed patterns of specialization—not abstract ambition. By examining how AI, goods, and services already cluster together, policymakers can see where innovation is most likely to succeed next. Rather than inventing entirely new sectors from scratch, national strategies should focus on connecting what already exists, using AI as the connective tissue. This is how countries move from today’s economic portfolio to tomorrow’s aspirations: incrementally, strategically, and with discipline. Regional cooperation can accelerate this path by pooling compute, harmonizing sectoral standards, and building shared data commons for priority sectors.

Ground strategy in existing economic strengths, not just stacks

If there is one takeaway from the paths of global AI development, it is this: AI is not a sector to build; it’s a capability to deploy. The countries that are getting closest to this approach aren’t starting from abstract AI principles or stacks. They’re embedding AI into their industrial strategies. Singapore doesn’t treat AI as a standalone sector. It integrates the technology explicitly into public services, education, and infrastructure, using “National AI Projects” and implementation pathways rather than treating AI as a standalone sector. The European Union tailored its AI rules to real use cases like robotics, health, and green manufacturing. At the same time, Europe illustrates the tension: use-case depth and “stack” aspirations are happening in parallel. Initiatives associated with “EuroStack” frame a push for greater digital sovereignty across layers of infrastructure from cloud to AI while EU programs, such as the Chips Act and EuroHPC AI Factories target specific bottlenecks (e.g. chips, compute access, applied enablement). The key distinction is that these stack elements are most defensible when they are selective and strategically focused on constraints that block sectoral deployment rather than treated as the national strategy itself.

By contrast, many national plans, especially those framed as AI leadership agendas lean heavily toward horizontal enablers (e.g. research, education/workforce, compute). For example, the Trump-era U.S. AI plan emphasizes skills, R&D, and national capacity-building; those are necessary, but without stronger sectoral pathways, they can under-specify how AI translates into competitiveness in particular industries. What these examples show is that the next generation of national AI strategies must move beyond compute capacity and regulatory frameworks

Figures 1 and 2 imply that advantage comes from deployment capacity in specific sectors, not generic stack-building. That deployment capacity depends on what we have called cognitive infrastructure—the intelligence layer that links data, people, and workflows across the real economy. This means developing local datasets, growing novel domain knowledge and expertise, building AI-literate talent, embedding institutional memory into software, and setting sector-specific safety and benchmarking standards.

This kind of cognitive infrastructure can turn a country’s existing expertise into competitive, AI-enabled products and services. It works effectively when countries try not to do everything at once but strategically connect AI into their economic and social assets and strengths.

The future of AI is not monolithic, and national strategies shouldn’t be either. Countries face a choice. They can pursue emblematic AI leadership measured in models trained, chips procured, and regulations passed, or they can pursue economic leadership measured in productivity, resilience, and the ability to adapt existing strengths to new realities. Countries should adopt two or three priority verticals for sectoral deployment, create sector-specific benchmarking regimes, and use public procurement to pull AI into priority workflows. This specialization will allow countries to focus on building workforce pipelines aligned with those verticals. Countries should coordinate on interoperability standards and evaluate them to avoid fragmentation and duplicative spend.

The latter path requires grounding AI in sectoral knowledge, institutional memory, and real economic constraints. But it is the only path toward turning intelligence into a lasting advantage.

In the end, the question for national AI strategies is whether the technology makes the economy more capable of learning, adapting, and competing over time.

The insights above draw from a dataset curated via the Quid platform, a semantic AI tool that combines natural language processing with investment and firmographic metadata from Capital IQ and Crunchbase. Quid searches and clusters over eight million global public and private company profiles, using Boolean logic and its proprietary LLMs to extract themes and topics. It identifies distinct sectors through keyword-based network clustering, including categories like “Generative AI,” “AI for Healthcare,” and “Data Management.” This data reveals who is building what kind of AI and where.

Search parameters and scope:

• Companies tagged with “AI,” “machine learning,” “deep learning,” or “generative AI”
• Investment event types: private placement, M&A, IPO, corporate VC
• Minimum threshold: companies with over $1.5 million raised between 2014 and 2025
• Company HQ country, founding year, sector, and status (active/inactive) included
• Up to 7,000 most relevant firms visualized per cluster