Why Do Americans Think China Leads the Global AI Race?

Key Takeaways

• Pew found that 36% of Americans see China as ahead in AI, compared with 12% for the United States.
• Measured leadership remains divided: the United States leads in capital, but China excels in patents and adoption.
• Most Americans expect AI to widen the economic gap between richer and poorer countries.

What the Pew Survey Reveals About the Global AI Race

On July 23, 2026, the Pew Research Center survey reported that 36% of U.S. adults believe China is more advanced than the United States in developing artificial intelligence (AI). Only 12% say the United States is ahead, 18% judge the countries about equal, and 33% are unsure.

Pew surveyed 3,488 adults from June 22 through June 28, 2026, through its American Trends Panel, a probability-based panel designed to represent the U.S. adult population. The result is striking because public perception does not match a simple reading of capital flows, frontier-model output, or corporate strength. It reflects a global AI race that Americans view through a mixture of news coverage, personal experience, political identity, and broader beliefs about national competitiveness.

The uncertainty figure deserves as much attention as the headline comparison. One-third of adults do not know which country leads. AI leadership cannot be observed through a single consumer product or national statistic. It includes advanced chips, computing capacity, electric power, research talent, scientific papers, patents, model performance, venture funding, cloud platforms, industrial deployment, defense applications, standards, and access to foreign markets. The public receives fragments of that picture, often after a Chinese model launch, a U.S. investment announcement, or a political speech about technological competition.

Pew also found that 43% of Americans consider it extremely or very important for the United States to become the world leader in AI development. Another 34% call leadership somewhat important, leaving 22% who consider it not too important or not important at all. Support for national leadership is broad, but intensity differs sharply by age, education, gender, and party.

The findings capture a public that sees strategic stakes but lacks agreement about the score, the purpose of leadership, or the policies needed to achieve it. They also fit a broader shift in how governments describe AI. The technology is increasingly treated as industrial capacity rather than a narrow software category.

The United States has organized federal policy around innovation, infrastructure, international diplomacy, and security through America’s AI Action Plan. China has combined state planning, domestic technology development, industrial deployment, research, and support for national companies. Other countries are pursuing forms of sovereign AI, seeking enough computing power, data control, talent, and regulatory authority to reduce dependence on either power.

New Space Economy’s examination of the AI cold war and sovereign AI describes this broader contest over computing infrastructure, energy, chips, data, security, and state capacity.

Pew’s numbers should not be read as a technical ranking. They measure public belief. That distinction matters because perceptions shape elections, spending priorities, regulation, education policy, export controls, immigration debates, and tolerance for large energy and data-center projects. A country may lead in measurable resources yet lose public confidence if people repeatedly encounter stories about a rival closing the gap.

Why China Appears More Advanced to Many Americans

China’s recent progress has been highly visible. Chinese laboratories and companies have released capable models at lower prices, often with open weights that developers can download, modify, and deploy. DeepSeek, Alibaba, Z.ai, Moonshot AI, ByteDance, Tencent, and Baidu have demonstrated that U.S. firms do not possess an uncontested claim to high-performing AI.

The Stanford University 2026 AI Index found that the performance gap between leading U.S. and Chinese models had narrowed to 2.7% as of March 2026. Models from the two countries had traded places near the top of prominent rankings since early 2025. Stanford still found that the United States produced more top-tier models and higher-impact patents, but China led in publication volume, citations, overall patent output, and industrial robot installations.

That narrowing gap is easier for the public to understand than investment statistics. Consumers see model demonstrations, coding results, image generation, translation, and low prices. They do not directly see private capital formation, specialized chip supply, data-center construction, or research migration. A Chinese model that reaches near-parity at a lower advertised cost can create an impression of national leadership even when U.S. companies retain advantages in revenue, cloud distribution, chip design, and access to private finance.

Open-weight distribution strengthens that effect. A model can gain influence without producing the highest benchmark score if it becomes easy to use, inexpensive to adapt, and attractive to companies or governments that prefer local control. New Space Economy’s analysis of Chinese open-weight AI models explains how lower-cost releases can weaken the pricing power of closed U.S. systems and broaden Chinese technical influence.

An open-weight model makes its trained numerical parameters available for download. It may still withhold its training data, development code, evaluation sets, or other components needed to reproduce the complete system. Open weight should therefore not be treated as identical to fully open-source AI.

The Council on Foreign Relations has argued that U.S.-China competition for international AI markets is intensifying as governments seek access to chips, models, cloud services, and technical support. A Chinese model does not need to win every performance comparison to reshape the geopolitical contest. Broad distribution can attract developers, local-language adaptations, derivative models, and foreign institutional users.

Patent activity adds another visible measure. The World Intellectual Property Organization reported in July 2026 that more than 56,000 generative AI patent families were published during 2024 and 2025, exceeding the cumulative output from 2014 through 2023. China-based inventors accounted for more than 43,000 of those patent families.

Patent counts do not measure commercial value, scientific quality, or actual deployment. One patent family can prove economically valuable, and thousands of others may have limited commercial effect. The volume still reinforces the belief that China is producing AI-related intellectual property at scale.

China also benefits from a broader narrative about infrastructure. Large power projects, industrial policy, domestic manufacturing, robotics, telecommunications, and state procurement make AI appear connected to national production rather than confined to a group of technology companies.

A Boston Consulting Group comparison published in June 2026 placed the United States ahead across capital, talent, intellectual property quality, data, energy, and computing resources. The analysis also found that China continued to close the computing gap and translate its intellectual property strengths into rapid frontier-model development and industrial adoption.

Public perception may reflect a broader concern that the United States is losing ground in science, manufacturing, education, or infrastructure. Pew’s question does not isolate the reason each respondent selected China, so the survey cannot establish why people answered as they did. It can show that the belief crosses party lines, gender groups, and AI-use categories.

More Republicans and Democrats identify China as the leader than identify the United States. AI chatbot users and nonusers also give China the same 36% figure. Direct exposure to chatbots lowers uncertainty, but it does not produce a broad belief that the United States is ahead.

Where the United States Still Holds Measurable Advantages

The United States retains substantial advantages that complicate the public verdict. The 2026 AI Index estimated that U.S. private AI investment reached $285.9 billion in 2025, compared with $12.4 billion in China. Stanford cautioned that private investment figures understate Chinese spending because government guidance funds and public financing operate differently.

Even with that limitation, the reported U.S. capital advantage remains immense. Stanford counted 1,953 newly funded U.S. AI companies during 2025, more than 10 times the count in the next country. Large American technology companies can also fund models, data centers, custom chips, power contracts, acquisitions, and distribution systems without relying solely on outside venture capital.

American companies occupy powerful positions throughout the computing chain. Nvidia dominates advanced AI accelerators. Microsoft, Amazon, and Google operate global cloud platforms. Anthropic, Google, OpenAI, Meta, and xAI develop frontier systems or widely used open-weight models. U.S. universities remain central to AI research, and the country continues to attract scientists, entrepreneurs, engineers, and graduate students.

The talent advantage is not guaranteed. Stanford reported that the number of AI researchers and developers moving to the United States had fallen 89% since 2017, including an 80% decline during the most recent year covered by its 2026 analysis. Immigration rules, research funding, corporate hiring, academic opportunities, and the growth of AI centers outside the United States will influence whether this decline continues.

Model production offers another measure. The United States produced more top-tier models than China in the 2025 data summarized by Stanford, even though the performance difference had narrowed. American leadership is strongest when measured through the combination of frontier laboratories, advanced chip design, cloud distribution, private capital, enterprise software, and global commercial reach.

China’s advantages appear more strongly in publication volume, patent volume, industrial robotics, domestic deployment, and state-supported scaling. A single winner label obscures these separate dimensions.

The AI race also depends on physical infrastructure. New Space Economy’s review of the Alberta AI data-center strategy shows how AI infrastructure includes energy, construction, cooling, networking, data, specialized labor, land, and regulatory approval.

America’s AI Action Plan treats power generation, transmission, permitting, and data-center construction as strategic requirements. Its three pillars cover accelerated innovation, expanded infrastructure, and international leadership. That framing confirms that software performance alone cannot secure national advantage.

Infrastructure constraints could become decisive. Training and operating advanced models requires large computing clusters, reliable power, high-capacity networks, water or alternative cooling systems, and access to specialized semiconductors.

The Organisation for Economic Co-operation and Development reported in July 2026 that AI investment and computing capacity remain concentrated in a small group of countries and companies. That concentration supports U.S. and Chinese power, but it also creates bottlenecks and dependency for smaller economies.

The United States hosts many of the world’s AI data centers and designs many of the most advanced processors. Manufacturing remains internationally dependent. Taiwan Semiconductor Manufacturing Company fabricates most leading AI chips, and advanced packaging, memory, lithography, networking, and semiconductor equipment rely on suppliers in several countries.

This interdependence means that national AI leadership is partly built on international supply chains. Export controls can restrict access to advanced equipment, but they can also encourage domestic alternatives, inventory accumulation, supplier substitution, and new commercial partnerships.

Some companies have proposed moving selected computing workloads into orbit, linking the AI race to launch capacity, satellites, energy systems, and communications. New Space Economy’s assessment of the orbital AI market explains why such concepts face severe limits involving launch cost, radiation, heat rejection, repair, debris, insurance, regulation, and the rapid obsolescence of computing hardware.

Their appearance still demonstrates how demand for computing has expanded the strategic conversation beyond algorithms. AI leadership now includes energy policy, industrial construction, communications networks, and access to advanced manufacturing.

Why National AI Leadership Matters to Americans

Pew found that 77% of Americans view U.S. leadership as at least somewhat important, combining the 43% who say it is extremely or very important with the 34% who say it is somewhat important. That level of concern reflects the range of outcomes attached to AI: economic growth, military capability, cybersecurity, scientific research, productivity, media influence, health care, education, public administration, and control of technical standards.

The Trump administration has used direct competitive language. America’s AI Action Plan describes the United States as being in a race for global AI leadership and organizes federal policy around reducing barriers, building infrastructure, and extending American technology internationally.

That policy may appeal to respondents who see AI leadership as an extension of national power. It may concern respondents who fear that competition will weaken safety, labor protections, privacy, environmental review, or international cooperation.

Party differences are real but not absolute. Among Republicans and Republican-leaning independents, 54% say U.S. leadership is extremely or very important. The figure is 34% among Democrats and Democratic-leaning independents. Yet 30% of Republicans and 43% of Democrats say China is more advanced.

The same people can believe that leadership matters and believe that the United States is behind. That combination can produce support for public investment, export controls, domestic manufacturing, education programs, or faster permitting, depending on political values.

Leadership can mean different things. One definition prioritizes the best models. Another focuses on the widest adoption. A third emphasizes profits, patents, military capability, scientific discovery, social benefit, or the ability to set rules.

New Space Economy’s review of public concerns about AI identifies job displacement, misinformation, surveillance, discrimination, energy use, weak accountability, and loss of human control as prominent issues. A nation could lead commercially yet fail to earn trust at home.

International influence may depend less on possessing the strongest model than on offering affordable systems, chips, cloud services, technical training, local-language tools, and financing. China’s open-weight and lower-cost offerings can appeal to countries that lack the funds or bargaining power to purchase premium U.S. systems.

U.S. companies retain global brands, extensive cloud platforms, developer communities, and enterprise relationships. Price, licensing terms, data-residency requirements, export controls, political conditions, and access to local support still affect adoption.

National leadership therefore has at least two audiences. Domestic voters want economic gains, safety, employment, and accountable institutions. Foreign governments want access, affordability, control, security, and the freedom to choose suppliers. A strategy that addresses only frontier-model performance may fail both audiences.

How Demographic Differences Shape the Survey Results

Age produces the largest gap in the importance question. Only 26% of adults ages 18 to 29 say U.S. AI leadership is extremely or very important. The share rises to 37% among adults ages 30 to 49, 51% among those ages 50 to 64, and 56% among adults age 65 or older.

Older Americans are more than twice as likely as the youngest adults to place high importance on national leadership. Several interpretations are possible, but the survey does not establish causation.

Older adults may be more familiar with Cold War competition, industrial rivalry with Japan, the space race, or debates over American decline. Younger adults may view AI through employment, education, creativity, surveillance, climate, and corporate power rather than national competition. They may also feel less attachment to a binary contest between the United States and China, even when they use AI tools more often.

Gender differences appear in both perception and importance. Men are more likely than women to say the United States leads, 18% compared with 7%. Women are more likely to say China leads, 41% compared with 31%, and they are more likely to be unsure, 37% compared with 29%.

Half of men call U.S. leadership extremely or very important, compared with 36% of women. These findings describe opinion differences, not technical knowledge, intelligence, or ability. Pew did not claim that one group evaluates AI more accurately.

Education also changes the response. Half of adults with at least a bachelor’s degree say leadership is extremely or very important, compared with 39% among those with some college education or less. College graduates are also more likely to expect AI to widen inequality between countries, 62% compared with 46%.

Greater exposure to technology, economics, international affairs, or workplace automation may contribute, but Pew’s data cannot identify one explanation.

AI chatbot use reduces uncertainty without changing the China figure. Among users, 28% are unsure who leads; among nonusers, 38% are unsure. Users are more likely to select the United States, 16% compared with 9%, yet 36% of both groups choose China.

Direct experience may make people more willing to form an opinion, but it does not produce a broad pro-American judgment. A person can regularly use an American chatbot and still believe China has stronger national momentum in manufacturing, infrastructure, robotics, or government coordination.

These divisions have policy consequences. An AI strategy framed mainly as competition with China may mobilize older voters and partisan constituencies, but it may fail to address the concerns of younger adults. Policies centered on jobs, affordability, education, privacy, energy costs, and public accountability may reach groups that do not respond strongly to national-leadership language.

New Space Economy’s discussion of AI ethics in 2026 connects these domestic concerns to international competition, infrastructure ownership, and state policy.

Why Americans Expect AI to Widen Global Inequality

A majority of Americans, 51%, say AI will increase the gap between rich and poor countries. Only 7% expect it to reduce the gap, 16% expect little difference, and 25% are unsure. Democrats are more likely than Republicans to predict a wider gap, 60% compared with 45%. AI chatbot users are also more likely to expect increased inequality, 55% compared with 47% among nonusers.

That concern has a strong economic basis. Advanced AI depends on expensive computing systems, reliable electricity, high-speed connectivity, skilled workers, digital data, research institutions, and companies able to deploy technology at scale.

Countries that already possess those resources can develop models, attract investment, automate high-value work, and export AI services. Countries with weak electrical grids, limited broadband, small research budgets, or high cloud costs may become customers rather than producers.

The United Nations Trade and Development report warned in 2025 that AI could deepen international inequality unless more countries gain infrastructure, data, skills, and participation in governance. UN Trade and Development projected that the global AI market could increase from $189 billion in 2023 to $4.8 trillion by 2033.

The forecast does not guarantee that the market will reach that figure. It demonstrates the scale of the economic activity that could become concentrated in countries and companies that control computing infrastructure, models, intellectual property, and distribution.

The OECD productivity analysis reached a related conclusion. AI could improve productivity in low-income and lower-middle-income economies, but results depend on complementary investments, skills, institutional quality, connectivity, financing, and access to computing.

Falling model prices help, yet inexpensive access to an application does not create domestic research capacity, local data governance, or ownership of the underlying infrastructure. A country may gain useful services without gaining the companies, intellectual property, jobs, and tax revenue produced by the technology.

Open models can reduce some barriers. Governments, universities, and companies can adapt them without paying for every use through a closed service. China’s model-distribution strategy may gain support in developing markets for that reason.

Open access does not eliminate hardware costs, technical staffing, cybersecurity requirements, software dependencies, or reliance on foreign chips and cloud providers. It can shift the point of dependency rather than remove it.

Middle powers face a related problem. Canada, France, Germany, India, Japan, South Korea, the United Kingdom, and Gulf states possess money, talent, energy, research institutions, or industrial strengths, but most cannot reproduce every layer controlled by the United States and China.

Canada’s AI for All strategy, launched in June 2026, seeks to expand domestic computing infrastructure, increase adoption, support Canadian companies, strengthen public trust, and retain greater national control over AI capabilities. New Space Economy’s analysis of the Canadian AI strategy connects those goals to compute access, public-sector procurement, skills, safety, and industrial development.

Similar national programs show that the global AI race is producing more than two strategies, even if two countries dominate the public story.

Global inequality is not predetermined. Shared research, lower-cost models, regional computing centers, open technical standards, education, reliable energy, and procurement policies can spread benefits. The decisive question is whether countries can use AI to build local capability or remain dependent on services, infrastructure, and rules controlled elsewhere.

What the Survey Misses When Leadership Becomes a Single Ranking

The phrase “AI leader” encourages a scoreboard that may be too simple for the technology. Stanford’s evidence places the United States ahead in private investment and the number of top-tier models, but it places China ahead in publication volume, patent volume, citations, and industrial robot installations.

The top model-performance gap had narrowed to 2.7% by March 2026. WIPO’s patent data favors China in quantity, though patent totals do not establish quality or commercial success. BCG’s six-factor comparison favors the United States, yet its analysis acknowledges that the two systems draw strength from different institutions.

A useful assessment separates at least four forms of leadership.

Scientific leadership concerns research quality, talent, publications, and discovery. Commercial leadership concerns investment, revenue, platforms, customers, and global distribution. Industrial leadership concerns deployment in factories, logistics, energy, transportation, robotics, and public services. Governance leadership concerns standards, safety practices, public legitimacy, regulation, and international cooperation.

A country may lead in one category and trail in another.

The global AI race framing can motivate investment, but it can also produce poor decisions. Governments may subsidize projects with weak economics, weaken oversight to accelerate construction, restrict research exchange, or treat every foreign model as a security threat. Companies may use national rivalry to obtain public support for data centers, power generation, chip factories, and favorable regulation.

Competition can stimulate research and infrastructure. It can also narrow debate around who is ahead rather than who benefits.

The framing understates the role of private firms. Frontier laboratories command computing resources and technical talent once associated mainly with states. Their decisions about model access, prices, safety testing, licensing, and foreign markets can alter national power.

Governments influence those firms through procurement, export rules, antitrust enforcement, immigration policy, energy approvals, research funding, and national-security restrictions. They do not fully control them.

Public opinion may become more informed as AI use spreads. Stanford reported that generative AI reached 53% global population adoption within three years, though adoption differed sharply by country and income. The United States ranked 24th in the measure used by the 2026 AI Index, at 28.3%, despite its investment and model-development strength.

Direct experience could make people less uncertain, but experience may also increase concern about jobs, fraud, privacy, reliability, or unequal access. New Space Economy’s coverage of the main AI risks in 2026 explains how risks now extend from model behavior into cybersecurity, labor, energy infrastructure, information integrity, and institutional accountability.

The most informative part of the Pew survey may be the coexistence of three beliefs: China appears ahead, U.S. leadership matters, and AI is likely to widen international inequality.

Together, those beliefs create pressure for a national response but do not supply a shared definition of success. Policy choices will differ depending on whether success means better models, cheaper access, stronger defenses, higher productivity, safer systems, or broader distribution of gains.

How Public Perception Could Influence the Next Phase of Competition

Public belief can change the resources available to governments and companies. If Americans think China leads, elected officials may find greater support for semiconductor policy, research funding, energy construction, workforce programs, and restrictions on technology transfers.

The same perception can produce anxiety-driven measures that cost more than they deliver. Sound policy requires measures that distinguish real capability gaps from symbolic competition.

Policies should identify the capability being strengthened, such as computing capacity, research talent, semiconductor production, grid reliability, model security, or small-business adoption. They should explain who receives the economic gains and who bears the costs. They should measure outcomes over time rather than treat spending, construction announcements, or model demonstrations as proof of progress.

Trust may determine whether the public accepts the physical expansion associated with AI. Data centers affect electricity demand, land use, water systems, transmission lines, and local tax arrangements. Semiconductor plants require large subsidies and long construction schedules. Training programs need schools, universities, employers, and immigration systems to work together.

National competition language can secure attention, but local consent depends on electricity rates, employment, environmental effects, land-use decisions, and credible oversight.

International policy needs similar precision. Export controls may slow access to advanced chips, but they can encourage domestic substitutes and redirect foreign buyers. Open models may spread influence, but they can also aid research and local innovation. Standards can improve safety, yet rules written by a small group of wealthy countries may reinforce the inequality that 51% of Americans expect.

The OECD’s July 2026 assessment of competition in AI markets emphasizes concentration across the value chain. Advanced processors, cloud infrastructure, data centers, foundation models, application platforms, and distribution channels can each become a source of market power.

The United States and China also share interests in limiting cyber misuse, accidental escalation, fraud, and the uncontrolled spread of dangerous capabilities. Competition does not remove the need for technical dialogue. A race without risk management can leave both sides less secure, even if one side wins a temporary benchmark or market advantage.

Pew’s survey captures a moment when Americans recognize the stakes but lack a common map. The most constructive response would treat public concern as a demand for transparent measures rather than proof that every policy labeled competitive deserves support.

Leadership will depend on technical performance, industrial use, economic distribution, public trust, and international relationships. No single percentage can settle that contest.

Summary

Pew’s June 2026 survey found a large perception gap: Americans are three times as likely to name China as the more advanced AI power as they are to name the United States.

The measured record is mixed. The United States commands far more reported private investment, more top-tier model production, leading chip and cloud companies, and strong global commercial reach. China leads in generative AI patent volume, has brought leading model performance close to parity, and is expanding influence through lower-cost and open-weight systems.

The survey also shows that AI competition is not understood solely as a technology contest. Age, gender, education, party, and personal use shape how people judge leadership and its importance. A majority expects the technology to widen inequality between countries, matching warnings from UN Trade and Development and the OECD about concentrated infrastructure, capital, and skills.

The deeper policy issue is democratic legitimacy. Governments can spend heavily and restrict technology in the name of winning, but public support may weaken if households face higher power costs, workers see insecure employment, communities distrust data-center projects, or benefits remain concentrated.

A lasting form of leadership will require measurable capability, broad economic participation, accountable institutions, and the ability to explain tradeoffs without turning every disagreement into a test of patriotism.

Appendix: Useful Books Available on Amazon

• AI Superpowers
• Chip War
• The Coming Wave
• The Worlds I See
• The Thinking Machine

Appendix: Top Questions Answered in This Article

What Percentage of Americans Think China Leads in AI?

Pew found that 36% of U.S. adults believe China is more advanced in AI development. Twelve percent select the United States, 18% say the countries are about equal, and 33% are unsure. The findings come from a survey of 3,488 adults conducted from June 22 through June 28, 2026.

Does the Evidence Show That China Is Definitively Ahead?

No single measure produces a definitive winner. The United States leads in reported private investment, top-tier company strength, advanced chip design, and cloud distribution. China leads in generative AI patent volume and has reduced the model-performance gap to a small margin on prominent evaluations.

Why Might the Public Think China Is Ahead?

Chinese companies have released competitive, lower-cost models that receive broad attention and are easy to test. Patent totals, industrial deployment, robotics, and visible infrastructure growth reinforce the impression. News about rapid Chinese gains may influence opinion more directly than less visible U.S. advantages in capital, cloud systems, and semiconductor design.

How Important Is AI Leadership to Americans?

Pew found that 43% call U.S. leadership extremely or very important, and 34% call it somewhat important. Twenty-two percent say it is not too important or not important at all. Older adults and Republicans express stronger concern than younger adults and Democrats.

Which Age Group Places the Most Importance on U.S. Leadership?

Adults age 65 or older show the strongest concern, with 56% calling leadership extremely or very important. The figure is 51% among those ages 50 to 64 and 26% among those ages 18 to 29. The survey does not establish the reasons for the age gap.

Do Republicans and Democrats Agree About Who Leads?

Both political coalitions are more likely to choose China than the United States. Republicans are more likely than Democrats to say the United States leads, but Democrats are more likely to identify China. Republicans also place greater importance on achieving U.S. leadership.

Why Does Private AI Investment Favor the United States?

The United States has deep venture-capital markets, large technology companies, global cloud providers, research universities, and established paths for financing startups. Reported private investment does not capture all Chinese state-backed spending, so direct comparisons require caution. The U.S. lead remains large under the available measure.

Why Does China Lead in Generative AI Patents?

Chinese companies, universities, research institutes, and state-linked organizations file large numbers of AI patents. Government priorities, industrial scale, and domestic competition support that activity. Patent volume shows inventive activity, but it does not prove that every patent has high value or produces a successful product.

Will AI Widen Inequality Between Countries?

A majority of Americans think it will. Countries with abundant computing power, reliable electricity, skilled workers, data, and investment can adopt AI faster and capture more value. Lower-cost models can help poorer countries, but infrastructure, governance, financing, and skills still determine whether access becomes local economic capability.

What Would Responsible AI Leadership Look Like?

Responsible leadership would combine strong research and infrastructure with fair competition, public accountability, worker preparation, secure systems, and wider access to economic gains. It would measure results instead of treating spending as success. International cooperation on safety and standards would remain necessary despite strategic rivalry.

Appendix: Glossary of Key Terms

Artificial Intelligence

Artificial intelligence refers to computer systems designed to perform tasks associated with human reasoning, perception, language, prediction, or decision support. AI includes techniques ranging from machine learning and computer vision to generative systems that create text, images, audio, video, or software code.

Global AI Race

The global AI race is a political and economic framing that compares countries by their ability to research, build, deploy, finance, and govern AI. The term often centers on the United States and China, although many other countries pursue national computing, data, talent, and industrial strategies.

American Trends Panel

The American Trends Panel is Pew Research Center’s probability-based survey panel of U.S. adults. Participants are recruited through methods intended to represent the national adult population, allowing Pew to estimate public opinion through published sampling procedures and statistical weighting.

Frontier Model

A frontier model is an advanced general-purpose AI system near the highest current level of capability. Frontier models often require large computing clusters, extensive data, specialized engineering, and substantial financing for training, evaluation, security, and operation.

Open-Weight Model

An open-weight model is an AI model whose trained numerical parameters are released for others to download and use. Access can support research and local adaptation, though the license may still limit commercial use, modification, redistribution, or selected applications.

Patent Family

A patent family is a group of related patent applications filed in more than one jurisdiction for the same or closely related invention. Counting families helps reduce duplication when comparing inventive activity across countries and patent offices.

AI Compute

AI compute refers to the computing resources used to train and operate AI systems, including advanced processors, memory, networking, data centers, software, electricity, cooling, and technical management. National access to compute affects research, deployment, security, and industrial competitiveness.

Sovereign AI

Sovereign AI is a policy approach through which a country seeks greater control over the computing resources, data, models, talent, and rules used for AI. Full independence is uncommon because chips, cloud services, research, capital, and supply chains cross national borders.

Export Controls

Export controls are government restrictions on the sale, transfer, or use of selected goods, software, technical information, or services. In AI competition, export controls often focus on advanced semiconductors, manufacturing equipment, computing systems, and technologies with national-security applications.

Generative AI

Generative AI refers to systems that create new content in response to instructions or examples. Outputs can include text, images, music, speech, software code, designs, video, or synthetic data. These systems are commonly built with large machine-learning models trained on extensive datasets.