Understanding the AI Competition with China

The development and adoption of artificial intelligence is poised to revolutionize the commercial and military worlds, making it a key focus of the technological competition between the United States and China. While the United States is generally understood to have a lead in AI development, in part due to its access and control over the highest-end semiconductors used to train larger language models, China has declared its intention to become the global leader in AI by 2030. Indeed, the release of Deepseek r1 in January raised the prospect that China could close the gap with the United States with far less investment in AI infrastructure.

To better understand the nature of this competition, BENS recently spoke with the following experts:

---

How would you characterize the competition over AI development between the United States and China?

Anthony Bak: First of all, we’re competing across the AI stack. Meaning everything from the extraction of natural resources, to chip manufacturing, to AI model production, to AI use. That’s really the full scope of where we’re competing. China has various efforts, like Made in China 2025, where they want to have their own complete domestic AI supply chain.

I would say the U.S. hasn’t really had a coherent strategy—meaning that we focused on trying to slow China down in their AI model production by reducing their access to chips. To some degree, that was successful under the previous administration and created a little space for us. But I don’t think we did a good job of saying, “Well, what do we do with this space?” So we’ve slowed them down, but it’s not a permanent state. We’re not going to be able to keep them back long term, so we really need to take advantage of the space we’ve created, and that’s where how exactly we are using AI comes into play, and that’s where I think we need to focus more of our competitive energy. We’re developing AI so that it can be used to do things in the real world. Somehow, we’ve lost track of that.

China is also aware of this, which is why they have a directive to companies to use Deepseek in their products. They understand the issue but I’m not sure they are in the best position to execute on it. I am not a China expert, but I am also not incredibly confident that doing things based on top-down directives for political reasons is the way to get ahead. We don’t have to be in a panic that they’re ahead of us because a lot of their companies will do political box checking to use Deepseek without truly appreciating the ground-truth challenges for bringing AI to bear on valuable problems. We really need to focus on how our system of innovative entrepreneurship can bring AI to all of our industries, making operations more efficient and making better, more innovative products. On the government side, we have missions we are trying to accomplish, particularly in national security, and we can use AI to better enable those mission outcomes.

I will say that there are different ways of investing in AI. One of them is to invest the way that traditional IT infrastructure has done, which is by buying point solutions for specific problems, which leaves you with traditional challenges as well as data and organizational silos, leaving a lack of transparency around decision making. The most successful places for AI are places where we are investing organization-wide across its functions, or even across ecosystems to see what AI can do for them.

For example, in the Department of Defense (DoD), Palantir has supported the Maven Program which started as a kind of point solution running computer vision models to identify and track targets in imagery – the “limited” kind of AI project. When it was this kind of solution, there was a lot of “So what? Who cares?”. It didn’t matter until it started taking in more of the targeting workflow, target development, and bringing in intelligence. According to an independent analysis, the result was an improvement by a factor of a hundred, so what used to take 2000 people in a targeting cell can now be done with 20. When you can complete tasks typically required of 2000 soldiers on the battlefield now with 20, a lot of things change about how you operate. The product is continuing to expand – into logistics and into readiness information – which will bring further gains: 100x improvement is just the beginning! When I talk about investing in using AI, this is the kind of project that I mean, it starts with investing in infrastructure that supports the full AI lifecycle and supports many disparate use cases, not isolated point solutions – and building that out improves core mission outcomes that span across many silos. Organizations have to start with ambition and vision, but revolution is at hand.

Aaron Burciaga: The U.S.-China race for AI dominance is a dead heat, with no other global players—neither the UK nor Russia—coming close to matching their pace. If I had to pick a leader, I’d give the U.S. a slight edge, driven by our unmatched innovation in frontier AI models, defense tech, and integration of AI into national security systems. That’s what keeps us ahead, but it’s a razor-thin margin.

The U.S. holds a slight edge thanks to our leadership in frontier AI innovation and defense technology, our ability to produce cutting edge models, and to integrate AI and defense networks—that’s what keeps us in the driver’s seat. China’s closing the gap fast. Their cost-effective AI models, massive data reserves, and swift military applications make them a serious threat. More critically, China’s outpacing us in building the backbone for AI—state-driven programs pumping out talent and infrastructure, from semiconductors to data centers. Their 2017 AI plan set a clear target: global AI supremacy by 2030, per China’s 2017 AI development plan. They’re backing it with heavy investment, while we’re lagging on scale. Even if we poured resources into our talent pipeline today, we’d struggle to churn out enough PhDs in five years to match China’s output.

The U.S. is still the brain of AI—our creativity and tech ecosystem are unmatched. But China’s the muscle, with the infrastructure and workforce to scale fast. In a kinetic conflict where both militaries wield AI, muscle could tip the scales. To stay ahead, we need to double down on our strengths: innovation, talent development, and public-private collaboration. But we’ve got to fix our weaknesses—chip supply chain vulnerabilities and regulatory bottlenecks—fast. Without bold investment and streamlined policies, we risk losing our lead before 2030. The clock’s ticking, and China’s not slowing down. Let’s seriously think about what’s ultimately going to win in a kinetic conflict with both militaries leveraging AI.

While the AI revolution is digital there is significant infrastructure involved in AI development, from chip fabrication to the power needs of data centers. How is industry approaching the infrastructure needs for AI?

Anthony Bak: There are a lot of chip startups right now trying to accelerate AI in various ways, and I think that’s really important going forward. Innovation in chips and chip production are key and these startups can help unlock some of that innovation related to the chip infrastructure for AI.

Then there’s the issue of energy – AI uses tremendous amounts of it. Energy is a big cost driver for data centers and we need to bring the costs of that energy down while also making sure energy production can meet the demand.

Indeed, I think the extent to which we can commodify AI will be a driving force behind our success in the AI race. We shouldn’t want AI use to be esoteric and highly specialized or have prohibitive costs for its use. We should want AI to be broadly accessible, cheap, and extremely flexible in its potential uses. We should identify the factors that will drive these characteristics – lower energy costs, more competition in chip manufacturing, and preserving competition among models – and craft policies that will deliver the best outcomes for each.

Industry’s most important tool has been Silicon Valley, a kind of “social infrastructure” for innovation. It’s the global center for technology. It attracts researchers from all over the world—France, Germany, and yes, China. It attracts the world’s smartest people, which is a major advantage for us that we need to maintain. A lot of people around the world have asked, “How can I build my own Silicon Valley?” But no one has really been able to reproduce it. It’s a very complicated ecosystem. It involves government support, venture capital, a culture that accepts failure, and the fact that Stanford and Berkeley just happen to be right there. We haven’t even been able to move that ecosystem to other parts of the country. Seattle, Austin, Boston – they all want to be tech centers, but it’s not the same.

From my standpoint, this is about enabling a healthy AI ecosystem, not top-down like in China, but bottom-up. People are free to work on problems that matter to them, access capital, and talent etc., and that freedom motivates innovation and benefits everyone.

Aaron Burciaga: The AI revolution hinges on four pillars: research and development, talent, processing power, and energy. Industry’s approach to these infrastructure needs is a mix of innovation and improvisation, but we’re not moving fast enough to secure long-term dominance.

On research and talent, we can’t keep leaning on Ivy League pipelines—that’s too narrow. I’ve seen promise in efforts like the Blue Collar Initiative, now championed by the National AI Council, to bring applied AI training to community colleges. You don’t need a PhD for the yeoman’s work of AI—data operations, model tuning, and system maintenance are perfect for associate-degree holders. But we’re not investing enough in a national program to scale this workforce. China’s pumping out talent through their own elite institutions, and while their governance stifles the creative spark we take for granted, their sheer volume is a threat. Our edge lies in fostering critical thinking and innovation, but we need to broaden access to training or risk falling behind.

Processing power is another bottleneck. Industry is treating data centers like any other commercial venture—think Amazon leasing warehouses from third parties. Companies like AWS and Google rely on suppliers to build and operate data centers, which is efficient but lacks strategic coordination. Here in Loudoun County, “Data Center Alley” hosts a quarter of U.S. data center capacity—a massive single point of failure if disrupted. Where’s the plan for redundancy? Advances in semiconductors and quantum computing could reshape how we build these facilities, potentially turning today’s data centers into tomorrow’s relics. We need to think bigger—more resilient, decentralized infrastructure to stay ahead.

Energy is the final piece, and it’s a game-changer. Tech giants like Meta are striking deals with energy providers for nuclear power to fuel AI’s massive computational needs. This isn’t your typical corporate supply chain—when did Ford ever contract with a utility for something as critical as nuclear energy? It’s a new paradigm, and it’s exposing gaps in our grid’s capacity and reliability. If we don’t scale sustainable energy solutions fast, we’ll hit a wall.

Industry is innovating, but it’s fragmented. To win, we need a coordinated national strategy—more investment in accessible AI education, resilient data center networks, and robust energy infrastructure. Without it, we’re leaving our AI future vulnerable.

What are the most impactful ways the U.S. government and private sector can work together to preserve U.S. leadership in AI?

Anthony Bak: Sitting at Palantir, we often give the same answer: our request to the U.S. government is to be a good customer.

Buy commercial products from American companies. That supports the innovation ecosystem that produces those products. For example, Palantir Foundry, our enterprise data platform, was originally developed to support Palantir’s commercial customers and is now deployed within DoD. Other technologies like the Palantir Ontology were originally developed as part of our government work and are now distributed to all of our commercial customers. Just be a good steward of the ecosystem. Don’t compete on commodity capabilities; seek an alpha that is unique and particular to your organizational mission and build solutions to that on top of the commercial capabilities. In government acquisitions, a lot of progress has been made in prioritizing commercial item preference, both in law and in recent executive orders. But we’re still seeing pockets of resistance where products are still being built by the government that could have otherwise been purchased from private industry. This is bad for the government and bad for the American people.

Drones are a good example of much-needed ‘ecosystem’ thinking. The DoD needs drones but it’s hard to get drones in volume if you’re not buying from China. We’re not talking large high-end drones like Reapers costing millions of dollars, we’re talking little drones that might be single-use. So you need them cheap, and you need a lot of them. That’s a problem for us, and it’s a problem for our allies. I think that the DoD should be thinking not just in terms of exquisite drones that they’re purchasing for themselves but considering how we build a native drone industry that we can call upon the way Ford Motor Company was called upon to build tanks instead of trucks during World War II.

That’s a very Palantir answer, but I think it’s true. There are also other things the U.S. government can do. For example, in medical research, AI can have huge benefits—but gathering datasets is hard and expensive. There are privacy issues such as getting patient consent to use their medical information for research, technological requirements to provide tools for de-identification of data, and different data formats across hospitals. As projects expand and grow you run into (justifiable) limitations on data re-purposing.

Look at N3C as an example where the government serves as a convener and guarantor of privacy and data security, bringing together large datasets of health information and making them available to researchers in a privacy-protecting way. This ongoing work should be expanded to cover more diseases.

We need to expand on how we think about making data available to researchers so that we can achieve better and more ambitious outcomes. Historically, data sets were static. You’d gather them, publish them on data.gov, and that was it. There was no connection back to the mission or the people affected. Modern software allows us to maintain those connections. The data isn’t dead—it’s live, and its real mission data, not something cooked up for AI researchers. If you’re building models to work with medical records, they need to be trained and tested on real, complex, messy records. And then when you have a result, if the data is live, you can go help those patients or recruit them into a clinical study immediately. Live data lowers the bar between insight and action. That’s true for any agency mission, not just healthcare. To truly solve problems, you have to deal with the full complexity of the data, the problem that needs to be solved, and the impacted people.

There are many other areas where this applies: disaster relief, for example, where the government could gather data and spur innovation. Efforts like that provide both immediate benefit and focus for AI researchers. When problems are unsolved, it motivates real research. The government can support that through various programs—things like AI testbeds tied to agency missions and important datasets. But it’s not just about giving out the data. The key is connecting AI researchers to what agencies actually need to do with that data to ensure models are solving the right problems and enabling mission outcomes.

Aaron Burciaga: To keep our edge in AI, the U.S. government needs to act decisively, partnering with industry and academia to build a robust, future-proof strategy. Here’s how we can do it.

First, we need a national plan for AI infrastructure—data centers, energy, and chip production. Right now, projections are muddied by politics or corporate hype, not hard data. Agencies like the Department of Commerce or the Federal Energy Regulatory Commission should set clear, evidence-based targets: where to build, how much power we’ll need, and what redundancy looks like. Why are we piling data centers in places like D.C. and Austin while ignoring strategic locations like Kentucky or Nebraska? A dedicated federal task force should map demand and incentivize decentralized, resilient infrastructure to avoid single points of failure.

Second, talent is our backbone, and we’re not investing enough to sustain it. We need a national policy to channel university endowments—sitting on billions—into AI education, from community colleges to PhDs. Programs like the Blue Collar Initiative, which I’ve supported, show that associate degrees can train a workforce for data operations and model deployment. But this needs scale—federal grants, tax incentives, and partnerships to flood the pipeline with diverse, skilled talent. China’s churning out graduates at a staggering rate; we can’t afford to coast on our Ivy League laurels.

Third, the DoD’s AI leadership needs a reality check. Some folks running these programs are solid, but they’re not steeped in cutting-edge AI. We need robust workforce exchange programs—think DoD engineers embedded in tech startups, or industry experts advising Pentagon AI initiatives. One-off university consultants won’t cut it. Structured, ongoing collaboration between DoD, industry, and academia will keep our defense AI sharp and relevant.

Finally, we must prioritize digital sovereignty. China’s AI vision clashes with our values—centralized control versus open innovation. We shouldn’t share the same digital sandbox. When my company, ZETEC LLC, worked with Australia, they demanded sovereign solutions, so we built a fully Australian subsidiary. Why aren’t we doing the same? From chip supply chains to platforms like TikTok, we need policies that protect our tech ecosystem and ensure our AI solutions reflect American priorities.

The U.S. leads in AI, but that lead’s fragile. Without bold government action—strategic infrastructure, aggressive talent investment, real DoD-industry collaboration, and a commitment to sovereignty—we’ll lose ground to China. It’s time to move with purpose, not complacency.

---

The views expressed by the interviewees are their own and may not represent those of BENS or their employers. Comments have been edited for length and clarity.