3.1.1 AI Race and National Competition

Dr. Sarah Chen sits in a windowless conference room in Washington D.C., reviewing classified intelligence briefings on China's AI capabilities. She's been working in AI policy for eight years—first at DARPA, then the National Security Council, now as a senior advisor on emerging technology.

The briefing she's reading contradicts almost everything she thought she knew six months ago.

In December 2024, Chinese startup DeepSeek released an AI model that performed comparably to OpenAI's most advanced systems—but trained for a fraction of the cost, using older chips that U.S. export controls were supposed to have restricted. The model went viral. Within weeks, it captured 4% of global market share that OpenAI had dominated for years.

Sarah had assured policymakers that U.S. chip export restrictions would keep China at least two years behind. The restrictions were comprehensive, carefully designed, and aggressively enforced. American semiconductors—the most advanced in the world—were the bottleneck. Control the chips, control the AI race.

DeepSeek proved that assumption wrong. The Chinese hadn't waited for American chips. They'd innovated around the restrictions, developing training methods that achieved competitive results with less computational power. They'd turned a constraint into an advantage.

Sarah's job just got much harder. Because the question she now has to answer is: if export controls don't work, what does?

The U.S.-China AI race isn't a sprint. It's a marathon being run on multiple tracks simultaneously, with rules that keep changing and no clear finish line. And increasingly, it's not obvious who's winning.

The Two-Superpower Reality

By early 2026, the global AI landscape is dominated by two powers: the United States and China. But they're not competing on the same terms.

The United States leads in cutting-edge model development. OpenAI, Google, Anthropic—U.S. firms produce the world's most capable large language models. In August 2025, U.S. models captured approximately 93% of global LLM site visits. American AI research sets the pace. American companies define the state of the art.

But market share isn't everything. China's approach is different—less focused on consumer-facing chatbots, more focused on industrial integration, infrastructure, and manufacturing. Chinese companies lead in AI-powered surveillance systems, smart city infrastructure, autonomous vehicles, and industrial robotics.

And Chinese models are catching up fast. Site visits to China-based LLMs increased by 460% in just two months in late 2025. DeepSeek's breakthrough wasn't a fluke—it signaled a broader pattern of Chinese firms closing the gap through innovation under constraint.

The competition is fragmenting across many domains: advanced model development, computing infrastructure, global standards and governance influence, integration into physical systems, and energy availability for massive data centers. Having an edge in one area doesn't automatically translate into an advantage in others. As a result, it's increasingly plausible that Washington and Beijing could each emerge as leaders in different parts of the AI ecosystem rather than one side decisively outpacing the other across the board. This isn't the outcome either superpower wanted. Both are pursuing AI dominance. But what they're getting is AI specialization—a world where the United States controls some layers of the technology stack, China controls others, and no one has total superiority.

The Investment Arms Race

Money tells part of the story. In 2025, the United States spent approximately $470.9 billion on AI—nearly 63% of global AI spending. Tech giants like Google, Microsoft, and OpenAI are pouring resources into model development, chip manufacturing, and data center infrastructure. China invested $119.3 billion—a quarter of U.S. spending in absolute terms, but massive relative to its economy. The Chinese government is backing smart manufacturing, healthcare AI, and autonomous driving through the National AI Industry Investment Fund. State support allows coordination that U.S. firms, despite their scale, struggle to match.

The race is not merely bilateral, however. The European Union launched InvestAI at the 2025 AI Action Summit in Paris, mobilizing commitments approaching €110 billion from public and private sources—an effort less aimed at winning the AI race than at avoiding being left behind entirely. The United Kingdom invested $28.2 billion, focusing on AI safety research, healthcare, and public services, while Canada allocated $15.3 billion toward computing infrastructure and research capacity. India ranked seventh globally in total AI investment by 2025, committing $11.1 billion to domestic development, and Israel dedicated $15 billion with a particular focus on defense and cybersecurity. The UAE, Singapore, Norway, and South Korea are also investing aggressively, with smaller economies prioritizing AI infrastructure and talent development in ways that, relative to their size, rival the major powers.

These investments reflect a broader realization: AI is not just another technology sector. It is foundational—shaping economic competitiveness, military capability, and geopolitical influence for decades to come. Countries that fall behind risk being locked out of future prosperity. And the gap between leaders and laggards is widening. Nations that invested early in digital infrastructure, AI education, and government adoption—like Singapore, the UAE, and Norway—are pulling ahead of larger economies that were slower to act.

The Strategy Divergence

In 2026, the United States and China are doubling down on divergent strategies. The United States is betting on exporting its tech stack. In December 2025, President Trump signaled a shift by allowing Nvidia to export advanced H200 chips to China—a reversal of previous export restrictions. The underlying strategy is to let American hardware and software become the global standard, then leverage that dependence for geopolitical influence. U.S. firms dominate cloud computing, AI models, and semiconductor design. If the rest of the world builds on American infrastructure, the United States retains influence even if it doesn't control end-use applications. It's a bet on technological lock-in.

China is betting on open-source. By releasing powerful models like DeepSeek publicly, Chinese firms make their technology the default for developers worldwide—especially in countries that cannot afford proprietary American systems. Open-source also circumvents export controls: if the model weights are publicly available, restrictions on chip sales become less relevant, because anyone can run Chinese AI on whatever hardware they have access to. It's a strategy of ubiquity over exclusivity.

Meanwhile, middle powers are pursuing sovereignty. India's launch of a sovereign large language model at the AI Impact Summit in February 2026 signals a broader trend: nations seeking control over their AI infrastructure to strengthen domestic economies, protect national security, mitigate geopolitical shocks, and reflect national values. Sovereign AI isn't about competing with the superpowers—it's about not being entirely dependent on them. Countries like India, Brazil, and Indonesia are investing in domestic AI capabilities not because they expect to out-innovate Silicon Valley or Shenzhen, but because relying entirely on foreign technology is a strategic vulnerability.

The Energy Bottleneck

One factor that receives insufficient attention in discussions of AI competition is energy. Training and running AI models at scale requires enormous computational power, which in turn demands enormous quantities of electricity. Modern data centers operating frontier models consume as much power as small cities, and energy requirements grow as models scale in size and deployment. This makes electricity generation not just an infrastructure concern but a strategic one.

China has produced more energy than the United States since 2010, and it is expanding capacity faster. The country has invested heavily in both conventional and renewable generation, giving it a structural advantage in powering the data centers that AI requires. If China's access to advanced chips increases—whether through relaxed export controls, domestic semiconductor production, or continued innovation under constraint—its abundant and comparatively cheap energy supply could become the decisive factor in the long-term competition.

The United States faces a different challenge. Data centers are already straining electrical grids in key regions, from Northern Virginia to Phoenix. New power generation takes years to permit and build, and the clean energy capacity that AI companies need to meet their sustainability commitments is particularly slow to come online. Unlike chips, where U.S. firms retain design leadership, or talent, where American universities continue to attract global researchers, energy is the one key input where the United States faces a genuine structural disadvantage. If the AI race extends deep into the 2030s—as it almost certainly will—this asymmetry could matter more than any other single variable.

The Myth of Total Dominance

No country is going to "win" the AI race in the sense of achieving uncontested dominance. The technology is too complex, too multifaceted, and too globally distributed. Different nations are pulling ahead in different dimensions, and the overall picture looks quite different depending on what one chooses to measure.

Dimension	Current Leader	Basis
Advanced model development	United States	OpenAI, Google, Anthropic
Industrial AI integration	China	Manufacturing, smart cities, robotics
AI governance frameworks	European Union	AI Act, GDPR
Energy availability	China	Larger grid, faster expansion
Global LLM market share	United States	~93% of site visits (2025)
AI surveillance systems	China	Smart city and defense infrastructure
AI safety research	United States / UK	Leading research institutions
Open-source AI adoption	China (growing)	DeepSeek and derivatives

What emerges from this picture is not a unipolar AI order with one superpower on top, but a fragmented, multipolar ecosystem where different nations control different pieces of the value chain and no one has total superiority. The competition is fragmenting across domains faster than any one nation can dominate all of them. And because AI development is cumulative—building on open research, shared datasets, and global talent flows—even adversarial nations are inadvertently contributing to each other's progress. Chinese researchers publish papers that inform American models. American open-source frameworks enable Chinese applications. Israeli defense research feeds into commercial systems deployed globally. The boundaries are porous.

This doesn't mean competition isn't real or intense. It is. But total victory—where one nation has overwhelming AI superiority across all dimensions—is probably impossible. The race will produce leaders in different domains, not a single winner-takes-all outcome.

The 2026 Inflection Point

Policy experts are calling 2026 a hinge year—a moment when strategic decisions made now will shape the next decade of AI competition. Several questions are critical. Will the United States maintain export controls, or will it adopt a more permissive approach to exporting chips and models to allies and rivals? Will China succeed in building a domestic semiconductor industry capable of producing cutting-edge chips, or will it remain dependent on smuggled or legacy hardware? Will middle powers develop genuine AI sovereignty, or will they remain subordinate to superpower ecosystems?

The answers aren't clear. But the decisions are being made right now—in boardrooms, in ministries, and in classified briefings. AI is entering a decisive phase—one defined less by speculative breakthroughs than by the hard realities of governance, adoption, and strategic competition. As AI systems move from experimentation to widespread deployment, policymakers face mounting pressure to translate abstract principles into enforceable rules while managing the economic and security consequences of uneven adoption across countries and sectors. Growing friction between rapid advances in AI capability, unsettled policy debates, and intensifying geopolitical rivalry is likely to make AI one of the most consequential forces shaping global politics for decades to come.

Summary

The global AI race is not a sprint to a single finish line but a sustained, multidimensional competition in which no country is likely to achieve total dominance. The United States and China are the clear leaders, but they are competing on different terms: the United States holds an advantage in cutting-edge model development and global market share, while China leads in industrial integration, energy availability, and the rapid deployment of AI into physical infrastructure. Investment is expanding well beyond these two powers, with the European Union, United Kingdom, India, Israel, and numerous smaller nations committing substantial resources to build domestic AI capabilities and avoid strategic dependence on the superpowers.

American and Chinese strategies have diverged sharply. The United States is betting on technological lock-in through hardware and platform exports, while China is pursuing ubiquity through open-source releases that circumvent export controls and reach developers in markets the United States cannot easily serve. Middle powers, meanwhile, are investing in sovereign AI not to compete with the superpowers, but to reduce their vulnerability to them.

Energy is the most underappreciated constraint in the competition. China's structural advantage in electricity generation could prove decisive as the race extends into the 2030s, particularly as U.S. data center expansion runs into grid capacity limits. The result, across all these dimensions, is a fragmented, multipolar AI ecosystem in which different nations lead in different areas—and in which the decisions being made in 2026, from export policy to infrastructure investment to open-source strategy, will shape the competitive landscape for years to come.

Key Takeaways

The U.S.-China AI competition is multidimensional, not a single race: the U.S. leads in frontier model development (~93% of global LLM site visits in 2025), while China leads in industrial integration, AI-powered infrastructure deployment, and energy availability.
DeepSeek's 2024 breakthrough — training a competitive model at a fraction of the cost using restricted chips — demonstrated that export controls are insufficient as a containment strategy; algorithmic efficiency can compensate for hardware disadvantage.
Investment is expanding well beyond the two superpowers: the EU mobilized ~€110 billion at the 2025 AI Action Summit; the UK, India, Israel, UAE, and smaller nations are all building domestic AI capacity specifically to avoid strategic dependence.
The U.S. and China have diverged strategically: the U.S. bets on technological lock-in through hardware and platform exports; China pursues ubiquity through open-source releases that reach global developers while circumventing chip export restrictions.
Energy is the most underappreciated constraint: China has produced more electricity than the U.S. since 2010 and is expanding faster, providing a structural advantage in powering data centers that will compound as AI scales through the 2030s.
Middle powers are investing in sovereign AI — not to out-innovate the superpowers but to reduce strategic dependence on them; India's launch of a sovereign LLM in February 2026 signals this is becoming a mainstream policy posture rather than an aspiration.
Total dominance is probably impossible: AI development is too multifaceted and globally distributed for any single nation to control all dimensions, and the most likely outcome is a fragmented multipolar ecosystem where different nations lead in different parts of the value chain.

Sources:

Last updated: 2026-02-25