By Javier Surasky
The way in which the strategic competition between China and the United States to lead the field of artificial intelligence (AI) is decided constitutes one of the key issues in analyzing the reconfiguration of the international order. At times, this “race” has been described as a competition for digital technological supremacy. Still, the data-driven analysis reveals a clearly more complex and multilinear scenario, framed within a game of interdependencies in which both countries have much to gain and much to lose, and where each moves its pieces on the global board from institutional bases, economic structures, and industrial priorities that differ yet remain interconnected through supply chains. Knowledge flows that neither can independently control.
Since the
early 2010s, scientific dynamism has increasingly been measured by the
production of academic articles in highly recognized journals. In this regard,
the AI Index Report 2025
explicitly identifies China as the principal contributor to global AI-related
publications, accounting for approximately 23.2% of all articles in the field
in 2023, compared to the United States’ 9.2%.
This
difference is reflected in the scientific impact measured by citations of
academic work: Chinese authors account for 22.6% of all AI citations worldwide,
followed by their European counterparts (20.9%), leaving the United States in
an “uncomfortable” 13%.
However,
when examining the hundred most-cited articles of the year, the United States
occupies a comfortable first place. In 2023, it accounted for half of the 100
most-cited articles, with China in second place.
This
illustrates how two innovation models interact in the realm of cutting-edge
scientific production: while China prioritizes volume and massification, the
United States favors talent concentration, science-industry alliances, and the
production of high-impact research.
The
business ecosystem, however, is changing rapidly. The United States still hosts
the world’s largest digital technology companies—such as OpenAI, Google
DeepMind, Anthropic, and Meta—but no one can ignore Chinese firms like Baidu,
Alibaba, Tencent, or Huawei. China’s corporate AI market is growing in
complexity and competition, driven by significant state investments and
promotional actions at both national and regional levels to establish
“preferential zones” for AI-related ventures, including tax incentives for
Chinese entrepreneurs.
Yet
contrary to widespread intuition, the OECD’s Main
Science and Technology Indicators report
shows that China allocated around 2.6% of its GDP to R&D in 2023, while the
United States dedicated 3.4%. When considering gross domestic expenditure on
R&D expressed in purchasing power parity (PPP)–adjusted dollars—combining
the R&D share of GDP with PPP-based GDP estimates—we find that both China
and the United States operate within similar orders of magnitude, each
sustaining R&D expenditure close to one trillion PPP dollars. That tells us
that both maintain high and balanced absolute levels of scientific and
technological effort.
This parity
is relevant for explaining the development of advanced computing infrastructure
in both countries. According to the
OECD’s ANBERD/BERD database, which reports investment in PPP-converted U.S.
dollars and focuses on three critical sectors for the AI economy—manufacture of
computers, electronic and optical products (C26), computer programming and
related IT services (CJ62 and J63), and business R&D in scientific fields
(M72)—distinct patterns are observed between China and the United States during
2020–2022. In China, business R&D expenditure in C26 reached USD 117.5
billion in 2022, compared to USD 107.5 billion in the United States. 2022 was
the year in which China surpassed the United States in this category.
BERD data
also show that the United States has greater diversification: investment in IT
services and related activities (J62 and J63) reached USD 72.3 billion in 2022,
while investment in sector M72 (business scientific R&D) amounted to USD
39.6 billion. No disaggregated information is available for China on these two
sectors. Still, the data suggest that China’s profile is more oriented toward
electronic manufacturing, whereas the United States emphasizes digital services
and high-value-added scientific activities.
U.S.
industrial policy has intensified since 2020 to avoid losing leadership. Under
the CHIPS and Science Act and related measures, the country announced more than
USD 500 billion in private-sector investments to expand domestic manufacturing
capacity, with the ultimate goal of tripling U.S. chip-manufacturing capacity
by 2032 and generating new jobs tied to the technology sector. Despite this,
the industry faces a projected talent shortage of 1.4 million skilled workers
by 2030, representing a challenge to the sustainability of this effort.
This
differentiation is also visible in the semiconductor sector. According to the State
of the U.S. Semiconductor Industry Report 2025, global semiconductor
sales reached USD 630.5 billion in 2024, and U.S.-headquartered firms captured
slightly more than half of that value, reflecting a dominant position in the
highest value-added segments of the production chain, including intellectual
property, chip design, and a substantial share of specialized manufacturing
equipment.
China, by
contrast, focuses its contribution on assembly, testing, and packaging stages
and stands as the world’s largest semiconductor market—largely due to the rapid
growth of its domestic demand. Nevertheless, the AI Index 2025 shows evolving
changes: China accounted for 69.7% of all AI patents granted worldwide in 2023,
far ahead of any other country. That aligns with a broader trend documented by
the World
Intellectual Property Organization, which indicates that China is the
leading source of patent applications filed under the Patent Cooperation Treaty
(PCT), an international agreement that enables an inventor or company to submit
a single international patent application to seek protection in multiple
countries simultaneously. Although the United States maintains strong positions
in patents related to specialized hardware, advanced algorithms, and high-value
software platforms, its absolute volume of AI patents remains lower than China’s.
This
contrast adds another layer, indicating that while China pursues a high-density
patenting strategy, the United States focuses on frontier niches.
Despite all
these elements, interdependence between the United States and China in digital
technologies, AI, and related sectors is a structural feature of their
reciprocal relationship. The global semiconductor value chain, for example,
depends on U.S. design, European equipment, and Chinese demand. According to
the U.S.
International Trade Commission, the United States has maintained a trade
surplus in semiconductors since the late 1990s, and in 2024, U.S. exports in
the sector reached approximately USD 57 billion, placing chips among its
leading export categories, while China remains a major importer of
semiconductors required to sustain both its manufacturing chains and domestic
market expansion.
Neither
U.S. export-control policies adopted since 2019 nor China’s technological
substitution strategies have eliminated the mutual dependence between these
countries in the production and development of digital goods and services.
However, they have intensified competition over energy and infrastructure,
pushing both countries to reassess their energy transition models. China is
moving toward establishing large computing parks associated with renewable
energy projects, while the United States combines tax incentives, weak climate
regulation, and corporate commitments to data-center decarbonization.
In another
key domain—talent—the United States remains an attractive destination.
According to the Survey of Earned Doctorates, 2595 doctoral degrees were
awarded in Computer and Information Sciences in 2024, 1772 of which were in
computer science. More than half of these doctorates were earned by
temporary-visa students (58% in computer science and 64% in artificial
intelligence), suggesting that the U.S. system remains globally recognized,
while also creating a structural dependence on international talent to sustain
its leadership in advanced research.
Moreover,
according to the AI
Talent Report 2025, academic programs relevant to AI (bachelor’s,
master’s, and doctoral degrees in computing, engineering, and mathematics) have
grown over the past decade. Still, at the doctoral level, dependence on
international talent continues to rise. The report also distinguishes between
software-oriented talent (models, algorithms, machine learning) and
hardware-oriented talent (semiconductors, accelerators, data centers),
emphasizing that the largest human-talent shortage is found in hardware
development. It concludes that labor demand in AI is growing faster than the
supply of graduates in both segments—an emerging structural bottleneck for U.S.
competitiveness.
Country-of-origin
data reinforce this pattern. According to the Survey of
Earned Doctorates (SED),
China and India were the main countries of temporary visa doctoral students in
U.S. science and engineering fields in 2024, with 6116 and 2482 doctorates,
respectively, meaning that a significant proportion of foreign talent in
knowledge-intensive scientific fields comes from these Asian countries.
As a
result, the U.S. doctoral training system in STEM—and especially in digital
disciplines—combines highly competitive research institutions with the
continuous influx of international students, forming a diverse talent base that
remains vulnerable to shifts in migration policy or in international mobility
dynamics.
In China’s
case, the scale of its higher-education system suggests a rapid expansion of
its potential technical-talent base. According to the China Statistical
Yearbook, approximately 1015 million postgraduate students, 10.47
million undergraduate students, and 5.53 million short-cycle program graduates
completed their studies in 2023—17 million, tertiary-education graduates in a
single year. Although the Yearbook does not provide a breakdown by discipline,
it is clear that even if only a small fraction enters STEM fields, the absolute
magnitude would position China to feed its educational system through domestic
demand alone.
Finally, we
cannot overlook the strategic competition between the two countries in the
diplomatic arena. The United States promotes market-driven AI governance
frameworks with minimal state intervention, whereas China favors a highly
centralized, state-directed development model and advances the so-called Global
Artificial Intelligence Governance Initiative, aimed at promoting and
protecting digital sovereignty, technological non-discrimination, and equitable
participation of developing countries. These nearly opposing visions do not
prevent both countries from agreeing on the need for minimum transparency
mechanisms, rules governing the military use of AI, and coordination forums to
address potential systemic risks.
This
overview shows that the competition for AI leadership cannot be confined to the
traditional molds used to define technological hegemony contests. The United
States retains advantages in advanced models, chip design, software,
manufacturing equipment, and high-impact scientific production; China leads in
publication volume, academic citations, AI patents, industrial adoption,
strategic manufacturing, and a self-sustaining university-based talent-creation
system. The result is an unstable, interdependent, and fragile equilibrium that
both powers appear willing to maintain despite their loud rhetoric of rivalry.
If China
and the United States manage to “adjust positions” in this field, the future
may lead to a shared global technological sphere, but if they fail, the world
may ultimately witness countries being drawn into a “digital cold war” between
two separate technological blocs, doubling systemic risks.
In the
short term, U.S. AI leadership is impossible without China, and Chinese AI
leadership is impossible without the United States. Both also require
contributions from the European Union (machinery, human talent, installed
capabilities), as well as from less digitally advanced countries (data, energy,
low-cost labor for tasks at the base of the chain, such as data labeling).
The AI race
is being run by intimate enemies who depend on each other.