President Donald Trump unveiled his AI Action Plan last week, an ambitious and strategically framed document that signals artificial intelligence is no longer a niche issue for technocrats. It has become the defining arena of great-power competition.
As AI has become more deeply embedded in governance, a critical question has emerged: Will this revolutionary technology tip the scales in favor of authoritarian regimes or empower democracies? History offers no easy answers. Past innovations have demonstrated both emancipatory and repressive potential. Theoretically, AI could enhance transparency, participation, and accountability.
Theory, however, is conjecture. There are underlying authoritarian advantages at a cognitive and structural level that cannot be wished away.
AI competition is not merely a race for innovation—it is a contest of governance models.
Autocracies—particularly China—are poised to benefit disproportionately from AI’s capabilities: pervasive surveillance, granular social control, and predictive state planning. It is time the United States openly acknowledges this truth.
The U.S. may lead in model development, especially with a deregulatory bent, but legal, institutional, and normative constraints hinder large-scale deployment. Preserving a strategic edge, then, demands realigning technology with national purpose and actively curbing China’s advances.
Without institutional agility, we risk becoming laboratories of invention but graveyards of power. Values matter, but they cannot defeat realpolitik.
Favoring China
A hard-nosed approach to AI geopolitics is essential, but so is examining the cognitive foundation beneath it. AI’s promise is inseparable from its epistemic architecture. These systems do not grasp meaning; they recognize patterns, not causes; predict behavior, not understand it. This mirrors the authoritarian mindset: outcomes over principles, efficiency over deliberation, compliance over comprehension. AI systems function best in environments where data flows are centralized, objectives are clearly defined, and behavior is treated as something to be monitored, modeled, and managed.
It is this very nature that gives China’s regime, which is obsessed with prediction and preemption, as well as surveillance and suppression, an advantage. While what AI can do (capability) gives autocracies an advantage, it is what they can do with AI (applicability) that cements it.
By emulating AI itself—relentlessly goal-driven and unencumbered by constraints—China is redefining technological power. Its edge lies not just in capacity, but in coherence: as a centralized, securitizing regime, it can align priority and policy with unmatched discipline.
The U.S., by contrast, is hampered by diffuse authority and bureaucratic competition. It cannot match China’s directive velocity without betraying its democratic character. Recognizing this asymmetry is not defeat—it’s a strategic imperative.
As the late Henry Kissinger polemically recognized, the rise of AI may even signify the end of the Enlightenment. Although we often frame geopolitics as a contest of ideas—with truth prevailing through persuasion—these battles are frequently shaped by structural dynamics. In recent centuries, technological innovation has conveniently reinforced Enlightenment ideals. But that alignment can no longer be assumed.
For one, Kissinger noted, the printing press became “the technological advance that most altered the course of modern history” in the 15th century, allowing “the search for empirical knowledge to supplant liturgical doctrine.” This is a valuable philosophical insight, as the printing press enhanced thinking, not supplanted it.
Yet the structural impacts are also notable. While some in the West thought we reached the “End of History” because our ideas won, the truth is that structures already in place predetermined the rise of our ideas. Just as French dramatist Louis-Sébastien Mercier hailed printing as “the most beautiful gift from heaven” before the French Revolution, warning tyrants to “[t]remble before the virtuous writer!” tyrants today may very well view AI as their gift and command us to tremble in return.
[Printing] has another excellence; it is the formidable bridle to arbitrary power, by making public its leaft encroachment, by suffering nothing to be concealed, and by eternizing the vices and even the weakness of kings.
How the tables turn, right? When Kissinger explained how the Age of Reason replaced the Age of Religion, alluding to the rise of “[i]ndividual insight and scientific knowledge,” it is hard to ignore how with a cognitive shift came a structural one. It is little wonder that replacing “faith as the principal criterion of human consciousness” facilitated democratization. Perhaps due to the hubris of great powers, or perhaps because our memories aren’t like ChatGPT’s, we observe our current order’s state and fail to recognize the patterns. Through Kissinger’s evolutionary lens, we see how AI—and the internet’s rise—prefigures a decline in the social conditions that once underpinned democratic success.
While we may increasingly think these conversations are lovey-dovey, perhaps a symptom of Kissinger’s point, they are anything but trivial. Introspection is a strategic virtue—especially when confronting Sun Tzu’s China. Policymaking must absorb, not ignore, the study of shifting social conditions. In an age of accelerating technological disruption, these dynamics are too consequential to leave to cultural critics.
Fighting Back
With the cognitive-structural lens established, it is important to look forward. Philosophical reflection is necessary, but it must not breed paralysis. Thinkers like Yuval Noah Harari have illustrated how AI benefits tyrants—how it thrives on centralized data, enables surveillance, and risks rendering segments of the population useless. However, as authors like Ben Buchanan and Andrew Imbrie rightly caution, we must not let “Cassandras”—or “evangelists” or “warriors”—dominate the conversation.
While China possesses an applicability advantage, it is equally important to recognize our structural strengths. After all, one hopes there is little appetite to abandon republican values—making it necessary to play to our strengths. Awareness of the authoritarian advantage must not lead to inaction.
One of the most concrete advantages we hold in the AI race is dominance in raw computational (or compute) power—the vital hardware required to deploy cutting-edge models, an area where the U.S. and its partners still have command over. Measures like chip tracking and the multilateral application of export controls help preserve this edge, ensuring critical components remain out of reach for China.
We currently host the most powerful AI supercomputers on the planet. One such system, the Colossus cluster, is reportedly in the process of doubling its capacity to a staggering 200,000 NVIDIA Hopper GPUs. By contrast, China’s DeepSeek cluster, much of which was built before export bans took effect, is estimated to contain just 10,000 to 50,000 chips. Future Chinese data centers are effectively limited by less-efficient hardware.
We must not conflate compute hegemony with AI dominance though, especially with China’s unfettered access to data. As competition gets harsher, compute’s tangibility grants us a real advantage. Yes, the release of China’s DeepSeek’s V3 and R1 large language models may have triggered a 3.4% drop in Nasdaq 100 futures, but this should underscore that innovation requires more than compute, rather than negate its crucial position. Even DeepSeek founder Liang Wenfeng has stated that the American “embargo on high-end chips” constrained DeepSeek’s growth.
This advantage is no accident. While the Chinese Communist Party prioritizes “stability maintenance” (weiwen), the U.S. excels at experimentation. Autocracies may better exploit AI at parity, but we still hold a tactical lead: authoritarian advantages materialize only with capability. Here, we have room to maneuver.
This isn’t to say that the cognitive-structural issue vanishes when we focus on capability. But retaining an edge creates the space to confront it on our terms. In the meantime, America must act deliberately to close the applicability gap through a whole-of-nation strategy.
Often overshadowed by debates over chips and data, an element of such strategy must entail talent. While the U.S. still leads in hosting top researchers, China is catching up, tripling the U.S. in producing elite AI specialists in 2022. Ideally, talent should be developed at home. But AI, unlike most industries, requires top-tier expertise—and right now. With China producing a huge share of global AI talent—AI has been its most popular new major for years—the U.S. is in a difficult position. Until it makes serious investments in domestic talent pipelines, it will need some foreign talent, much of it Chinese. This makes tight security safeguards essential, and should make talent production a priority.
Importantly, Washington should not entertain the misguided notion that China can somehow be made dependent on U.S. technology. This illusion ignores the very strategy that enabled China’s rise: learning from the West, then replacing it. The U.S. must work with allies to close export loopholes and deny Beijing access to cutting-edge innovation. It should also consider banning Chinese-developed platforms like DeepSeek and urge partners to follow suit, thereby limiting China’s ability to train on open-access, high-quality data. Beijing blocks its citizens from accessing our leading AI systems. We should respond in kind.
Capability gives us the tools, but applicability determines whether those tools will gather dust. We must think like a chess player, not a sprinter; we must not sacrifice strategy for speed. We must move deliberately, leveraging our compute, revitalizing our talent, and binding partners in shared purpose. If we wish to lead, we must not only invent the future. We must prove we can govern it.