The global stakes of the U.S.-China AI rivalry

A focus on “who’s winning the AI race” can overlook key concerns around safety, governance, and equity, experts tell Rest of World.

Rest of World via Firefly

By REST OF WORLD STAFF

12 JUNE 2025TRANSLATE

Since the launch of DeepSeek earlier this year, everyone from talk show hosts to heads of state has had a point of view on the so-called artificial intelligence race between the U.S. and China. The discourse is largely about timelines, compute power, and export controls, with little discussion of what AI leadership truly looks like, how the technology will be governed, and what adoption of AI means for people and nations worldwide.

Listen to story14 MINUTES

Learn more

0:00/14:571X

Rest of World spoke to a wide range of experts to explore the broader societal, economic, and ethical ramifications of the rapid evolution of the technology, and the AI race between the two global powers.

Timnit Gebru, founder of the Distributed AI Research Institute, U.S.

Everybody wants to be in the race rather than taking a step back and critically thinking about what claims are actually true versus not, and whether or not the path that we’re taking is actually good for innovation. I do believe that it’s a race to the bottom.

What is the hype? What’s reality? We see even with DeepSeek, there are claims about reasoning capabilities, and we’re not sure of some of these benchmarks. It’s unclear if a particular high performance on a particular benchmark actually means that these models are doing reasoning right, so there are a lot of misleading claims currently.

What worries me is that this is further going to be used to say that the U.S. has to win the race against China, and therefore companies and the U.S. should face absolutely no restrictions — which means more data theft, and absolutely no oversight into what these models are producing.

I do believe that it’s a race to the bottom.

Sameer Patil, director, Centre for Security, Strategy and Technology, Observer Research Foundation, India

Prevailing in the global AI race entails far more than merely achieving domestic leadership in innovation and industrial integration. It also encompasses the worldwide endorsement of technical standards for AI development and, perhaps more crucially, the underlying governance frameworks.

India, while not yet established as a major producer of AI technologies, is rapidly distinguishing itself as a prominent adopter. The nation remains, for the most part, a consumer rather than an originator of AI innovations. The path forward lies not in replicating American or Chinese paradigms, but in cultivating indigenous equivalents to platforms such as ChatGPT or DeepSeek. Fortunately, India possesses a vast reservoir of young, highly skilled engineers who can serve as the bedrock for future AI innovation.

Irene Solaiman, head of global policy, Hugging Face, U.S.

The race between the U.S. and China has been interpreted often as racing toward AGI [artificial general intelligence] or toward the tech stack. There is competition on deployment, a sort of race for the global ecosystem as Chinese-developed systems, especially open-weight models, are deployed worldwide. I see this race as a culmination of years of different cultural values. Models developed in China, such as DeepSeek and Qwen, do exhibit different valuesand have different content restrictions compared to those developed in the U.S.

For those affected by this race, there is increasing pressure to build and deploy not just good systems, but cheap, accessible AI. DeepSeek’s R1 model shifted narratives on open-source in both the West and China, with more need to compete on open-source — these cheap, powerful, yet compute-efficient models can be more easily deployed worldwide.

On Hugging Face, we’ve seen models such as Qwen and DeepSeek be popular simply because they’re accessible to use and run. Meta’s Llama team and the upcoming open-weight model from OpenAI show U.S. progress in meeting user, research, and commercial needs. I’ve also noted upticks in European and other ecosystems becoming motivated to compete in model training, especially catalyzed by DeepSeek’s rise. In many ways, this type of competition toward better, cheaper models can be good!

J.S. Tan, doctoral student at the Massachusetts Institute of Technology, U.S., and author of the Value Added newsletter

Since 2021, the Chinese state has tried to steer tech development away from “soft tech” (consumer-facing platforms like e-commerce, food delivery, and social media) and toward “hard tech” such as advanced manufacturing and renewables.The fact that it’s still the internet giants that have the real capacity to drive AI innovation and diffusion probably isn’t what Beijing had in mind. I imagine what winning the AI race means for the [Chinese] state is not about boosting AI-powered features on WeChat or other consumer apps, but more about industrial productivity.

Ramesh Srinivasan, professor of information studies at University of California, Los Angeles, and author of “Beyond the Valley”

The DeepSeek moment showed that there are completely distinct approaches that the two likely take toward AI. It showed that you can do so much with so few resources, while we are bludgeoning ourselves with these massive investments and this idea of a nationalist AI. The Chinese approach is incredibly less expensive.

This race is likely to further cement and widen economic inequalities.

We also have to ask the question of whether we want any of this. And how much of this is total over-speculated hype? Big tech has taken over, and who’s conspicuously left out is everybody else. Much of the world’s population is just ignored, as usual, outside of being receptacles for the collection of data without disclosure or compensation. It’s incredibly opaque, the way that it’s being set up and framed and proselytized. Political and economic power is directly tied to who holds the strings and power over these technologies.

Near term, we’re going to likely have a kind of normalization or universalization of likely discriminatory systems. This race is likely to further cement and widen economic inequalities and create more precarity than ever in terms of access to economic security or secure work.

Dang Nguyen, research fellow at the ARC Centre of Excellence for Automated Decision Making and Society, RMIT University, Australia

The idea of “winning” the AI race reflects a geopolitical fantasy more than a measurable end point. The race is not just about building better models; it’s about exporting the conditions under which those models are built, maintained, and made profitable.

In that sense, “winning” isn’t a universal good. We shouldn’t be spectators of a fight where we hope the good guy wins, whatever that means. Being ahead in the race means dominating not only AI R&D pipelines but also the infrastructural and economic architectures that tether other countries — especially those in the Global South — to extractive platform dependencies, outsourced annotation work, and compromised sovereignty. The metrics of “progress” here are deeply entangled with asymmetries in compute access, data flows, and regulatory influence.

This can marginalize smaller players, deepen existing inequalities, and foreclose alternatives that don’t conform to U.S. or Chinese visions.

For countries in Southeast Asia, Africa, and Latin America, being caught in this rivalry often means contending with an influx of AI-linked infrastructure: cloud computing, data centers, startup investment. There’s often little room to shape AI deployment in ways that reflect local social, economic, or political realities. Governments and companies in these regions are increasingly encouraged to align with the logics of scale, optimization, and platform integration — often at the expense of public interest tech or more grounded, socially embedded uses of AI. This can marginalize smaller players, deepen existing inequalities, and foreclose alternatives that don’t conform to U.S. or Chinese visions of AI governance.

The metaphor of the AI race … misleads us into thinking that ultimately the future of AI is about who builds the most powerful models. In reality … it’s about how a race between two powers reorganizes the rest of the world into a supply chain of data, labor, compliance, and aspiration.

Aakrit Vaish, former adviser to IndiaAI Mission and founder of Haptik, a conversational AI company

To be a true player in this race means owning the underlying stack — compute, data, and foundational models — and using it not only for domestic growth but also as a global export engine.

Unsurprisingly, the U.S. and China are going all in. It’s tempting to say India should be competing at the same level. But that’s not a fair comparison. Our per capita income is a fraction of these countries’, which means our national priorities are — and must be — different: food, housing, health care, infrastructure. So today, India is not a core competitor in the AI race. But it is a key participant as a consumer and developer.

As a consumer, India’s scale makes it one of the largest markets for AI adoption — 1.5 billion people, 1 billion internet users, and deep digital penetration. As a developer, India’s unmatched tech talent is building AI solutions that power companies across the globe. We may not be building foundational models or owning large-scale compute yet — but we are in the game. And over time, with the right investments and strategic clarity, we can deepen that role.

Kashifu Inuwa Abdullahi, director-general, National Information Technology Development Agency, Nigeria

AI leadership is not about compute capacity alone; it’s about inclusion, relevance, and responsible governance. For Nigeria, the priority is building talent, enabling small language models that reflect our local context, and creating frameworks that protect our data and values. We see AI as an opportunity to strengthen public services, drive innovation, and ensure Africans help shape the global direction of this technology.

Leadership is not about compute capacity alone; it’s about inclusion, relevance, and responsible governance.

Jeffrey Ding, assistant professor of political science, George Washington University, U.S., and author of “Technology and the Rise of Great Powers”

When it comes to assessing whether the U.S. or China is ahead in the AI race, the first step is to clarify what winning the race actually means. Technological leadership in AI is about which country can diffuse and spread this general-purpose technology across its entire economy, so as to achieve broad-based productivity growth.

Under this framework, it’s not surprising that China has frontier firms like DeepSeek that are close to the innovative frontier, but the size of the gap in innovative capacity is not the most important factor. … The U.S. is very well positioned to win the marathon to diffuse AI at scale.

Sayash Kapoor, doctoral student at Princeton University’s Center for Information Technology Policy, U.S., and co-author of “AI Snake Oil”

The technical gap [between the two countries] is actually quite small. Chinese AI companies are, at most, 6–12 months behind leading U.S. companies in terms of model capabilities. This shouldn’t surprise anyone — AI knowledge proliferates quickly across borders, and there are many Chinese companies iterating quickly in the space.

Much of AI analysis focuses on the capabilities of AI models as a shorthand for which country is ahead. This is often a proxy for which country would be best positioned to realize the benefits of AI. But the adoption of AI across the economy is far more important for realizing its impact than simply building better models. AI analysis should be a lot more concerned with diffusion than innovation alone.

Paul Triolo, technology policy lead and partner, DGA-Albright Stonebridge Group, U.S.

The narrative of a race between the U.S. and China can only go so far. AI is an enabling technology, and at the economic level, will eventually permeate all aspects of business and trade, with no one really thinking about who is winning a race. Is anyone winning the race for electricity, 5G, or office software?

But it is a great power competition and increasingly, some are talking about whether a country — not a company — that gets to artificial superintelligence will have a decisive strategic advantage that will facilitate dominance in military, economic, and every other sphere of interaction. Getting there first, this paradigm holds, will be a zero sum game, winner takes all.

Most countries and companies can only stand back and watch as the elephants battle, hoping not to get crushed

It is not clear that China wants to play this game, while the U.S. and its national security establishment clearly does. For Beijing, AI is thought about primarily as a tool for economic growth, not military or economic dominance or hegemony. While the U.S. national security establishment, working closely with the leading AI labs, continues to signal that it is an absolute priority to deny China getting to advanced AI first.

This is causing friction with Beijing at many levels, resulting in major collateral damage around the world, and ultimately raising risks around Taiwan and global efforts to raise guardrails around advanced AI models and applications based on them. Businesses in both countries are caught in the middle. In the rest of the world, most countries and companies can only stand back and watch as the elephants battle, hoping not to get crushed in the bargain.

Alvin Wang Graylin, tech entrepreneur and digital fellow, Stanford University, U.S.

There is no AI war or race yet; if we continue to act the way we do, there will be a war. The belief that the U.S. is five years ahead of China, and that if we put sanctions, then we can stop their progress and we can be ahead forever — that is a dangerous myth that just isn’t true and is not executable, and is not even desirable.

By perpetuating an AI race, we will exacerbate the constraint on compute in the world, limit the amount and representativeness of data used to train, resulting in a suboptimal and biased understanding of our world, increase the danger of rogue actors’ misuse, slow down the diffusion of benefits, and raise the potential for escalation of this race into a full-out kinetic or nuclear war.

Do we really need to go as fast as we can? We are 8 billion people, we are all tied together. If one of us fails, all of us fail. The better model is collaborative acceleration: We work together, we share resources, make sure the benefits are societal, focus on safety over speed, and think long-term. If we do that, we’re going to get there a lot faster. In the long-term game, the optimal decision is to cooperate. There can be no winner and so many ways for us all to lose. The choice is obvious, but do our political leaders have the will to seek global benefit over personal gain? 

Reporting by Rina Chandran, Ananya Bhattacharya, Kinling Lo, Lam Le and Damilare Dosunmu.

Source: http://www.restoftheworld.org