This article was originally published in The Free Press.
On October 4, 1957, the USSR launched Sputnik, a 184-pound satellite, into Earth’s orbit. The satellite didn’t do much — it just “beeped” over radio waves. But those beeps sounded a wake-up call for the United States.
Fearful of falling behind its Cold War strategic rival, the federal government launched an expensive crash program to spur technological development. The U.S. went from a laggard in the space race to unified, fast-moving behemoth. By 1969, we had landed men on the moon, a feat the Soviets never accomplished. And more than that: The Sputnik moment began decades of across-the-board American dominance in science and engineering.
Now comes what many are calling a new Sputnik moment: the release of DeepSeek, a low-cost, high-performing Chinese-created artificial intelligence (AI) model. The analogy is a bit imprecise though — and probably understates the significance of last week’s event.
Rockets are bounded by the laws of physics and the scarcity and movement of materials — which is why we say that hardware is hard. The only constraint on software development is the human imagination. Small, far-flung teams can accomplish extraordinary things. If a rocket explodes, it takes 12 months to get the next one built. Software can instantly be replicated 7 billion times into every human’s pocket. And updated seamlessly.
Another difference: The space race (notwithstanding its many peace dividends) was primarily focused on the next weapons of war — ICBMs and the like. Building leading AI or, better yet, artificial general intelligence, is more like building the brain layer for everything — weapons, yes, but also “agents” for almost every kind of imaginable work, and potentially the source of almost all knowledge and information to be consumed by humans in the future.
Whoever controls AI will control the answers to questions like: What happened on June 4, 1989? How many genders are there? Did Covid-19 lockdowns work? What’s happening in Israel? What happens when you ask AI these questions depends on who created the AI, where it’s hosted, and what information it was trained on. Which is why every cultural warrior and government wants to control AI.
And, unlike rockets, AI models are fundamentally “black boxes” that can be observed, but not deterministically explained. AI models generate responses probabilistically, meaning they weigh countless possibilities and choose outputs based on likelihood rather than fixed rules. This makes them flexible but also unpredictable — and if adversaries control the models, they could manipulate the underlying probabilities to subtly distort truth, influence decisions, or spread falsehoods at scale. This makes the geopolitical risk of having an adversary control these “brains” real. The space race operated at the intersection of national pride and national defense; the AI race is about so much more — and progressing at breakneck speed.
Unlike rockets, AI is primarily bounded by math, compute (the processing power needed to train an AI model), and data.
For the math part, it’s largely vector math — linear algebra and multivariable calculus. China finished first in the International Mathematical Olympiad from 2019 to 2023, with a United States team of four Chinese Americans (out of six team members) finally besting China in 2024. China is very, very good at math.
There’s a joke that in every International Mathematical Olympiad, the top teams are China versus those of Chinese background. It’s why the allegations that DeepSeek is some kind of Communist psyop ring hollow. While it’s hard to confirm the exact amount of money spent or chips used to train the model, China is the world leader in human capital around this type of work, so it’s not surprising to see a tremendous advance come from that nation.
By making its model open-source, the Chinese hedge fund behind DeepSeek has confirmed how counterproductive the Biden administration’s pro-containment, pro-hegemony, anti-open source AI strategy had been.
Biden issued an executive order which sought to constrain compute under an arbitrary threshold, bar open source as an alleged threat to national security, and effectively allow regulatory capture by the biggest players. The administration and its enablers wanted to limit math, and in turn, limit code—but ended up just limiting America’s lead.
The apparent concern with an American open-source model was that the Chinese would copy it. DeepSeek has flipped the bit, so to speak: It’s the Chinese who have released something open source that now every American company is seeking to use or replicate, because of its incredible performance.
In that ironic sense, DeepSeek is an incredible gift to the American people. It’s not exactly Sputnik (for starters, it’s much more useful), but it could align our policy goals with reality and light a fire under both government and private-sector actors alike. In an encouraging start, President Donald Trump has thrown out all of Biden’s AI-focused executive orders.
By radically reducing the cost of cutting-edge AI, DeepSeek — helped by a regulatory regime in the U.S. that, unlike China’s, appreciates entrepreneurship — should usher in an explosion of AI applications.
Soon, the paradoxical question before us might be what to do when AI goes from scarce to practically free? This is because of the Jevons paradox, perhaps more accurately described as the Jevons effect. When James Watt introduced a more efficient steam engine in the late 1700s, one would have expected coal use to decrease, since less coal could now produce more output. Instead, steam-engine use skyrocketed as new applications for it emerged, and so did demand for coal.
The Jevons effect explains how cooling went from iceboxes to central air-conditioning, or how faster computing has only yielded a need for more computers by uncovering more uses.
AI is about to experience its own Jevons effect. It’s a new type of software that can handle almost any human labor involving “bits” (with atoms hopefully soon to come) — answering emails, conducting phone calls, writing documents, editing or searching photos, making diagnoses, anything.
It’s going to be inside everything, the basis for all software and systems of interaction between machine and human. Companies like Hippocratic AI that works as an AI nurse or Eve that automates part of a plaintiff’s law work use AI “brains” from others (like OpenAI, Meta’s Llama, Mistral AI, or now DeepSeek) to develop their products.
In principle, it’s no different from the way that textile mills, printing presses, or metalworks used steam engines and coal to power their factories. These brains keep getting cheaper, better, and smarter, with DeepSeek being the biggest leap yet.
And this is where America still has an edge that we have to keep. In 2018, 51,302 start-ups were created in China; that number plummeted to around 260 by 2024 due to Chinese regulatory overreach and an increasingly restrictive business environment under President Xi Jinping’s stubbornly Marxist rule.
If America is to win the AI race, we must embrace open innovation, incentivize entrepreneurship with low regulatory burdens, and ensure that the Jevons effect works in our favor. The future isn’t just about algorithms — it’s about the policies and culture that enable them to thrive.
