For Americans old enough to remember, the USSR’s 1957 launch of Sputnik 1 was both a gut-punch and a source of awe. The established narrative of the United States as the uncontested technological leader was shattered in an instant, galvanizing a decades-long competition in space technology, defense, and education. It quickly earned a name: the Space Race.

The United States won that race convincingly twelve years later when Neil Armstrong and Buzz Aldrin set foot on the Moon. 

Earlier this year, when the Chinese lab DeepSeek released its DeepSeek–R1 open-source large language model, there were echoes of Sputnik — a comparison made by more than a few people. But unlike the Space Race, or the 20th-century race in which a stalemate was reached — the nuclear arms race — this new race isn’t about one technological category. Artificial Intelligence is the technology that unlocks all other technologies. 

It is possible that we will look back in two years and laugh at the “DeepSeek scare” when American labs release better models. As we’ve already seen in the intervening months, DeepSeek’s virality was no match for ChatGPT when the latter saw a frenzy of use to retouch photographs in the style of Japanese animators Studio Ghibli. 

Still, it highlights something important: We are in an Everything Race. AI is accelerating advancements in practically every industry: medicine, manufacturing, national security, education. So, whichever country wins the Everything Race wins… everything. The winner of the Everything Race may not be the country with the single most advanced model at a snapshot in time, but rather the one that orchestrates a continuous, well-supported pipeline of breakthroughs—and then efficiently deploys those breakthroughs in the real world, tangible to citizens.

The question for the United States in particular is how to “win” that race in a way that aligns with American values and interests, while also competing with rival nations’ own ambitions.

***

Sputnik led to a society-wide mobilization in the United States. But today’s political and cultural landscape feels more polarized, whether because of a new president in the White House, revamping of federal bureaucracy, animosity from the media to people in tech or high cost of living to normal americans––making it unclear whether the U.S. can or will mobilize in a similarly unified way.

For those who follow AI research closely, DeepSeek’s models are less of a “magical leap” and more a combination of cost-reduction and refinement of known methods. As Dario Amodei, CEO of Anthropic put it in his personal blog “DeepSeek released a model called "DeepSeek-V3" that was a pure pretrained model. Then weeks later, they released "R1.” DeepSeek-V3 was actually the real innovation and what should have made people take notice a month ago (we certainly did).” To DeepSeek’s credit, this required great innovation in a few categories. The model uses “chain of thought” reasoning to self-evaluate its performance. It uses reinforcement learning to guide itself, which is a way for one AI model to learn from another AI model. It also looks like DeepSeek used OpenAI’s models for distillation.

The new R1 is open-source, widely available, and offers refined reasoning capabilities. This is a clarion call: big leaps in AI can indeed be replicated once the underlying algorithms and architectures become known. “In the face of disruptive technology, a closed-source moat is temporary. Even OpenAI’s closed source approach hasn’t stopped others from catching up,” said DeepSeek’s CEO at The China Academy. “Our value lies in our team, which grows and accumulates know-how through this process. Building an organization and culture that can consistently innovate is our real moat.”

Secrecy affords protection only so long. Once any frontier lab, in any country, has a breakthrough, replication is just a matter of time, talent, and compute.

***

For both frontier AI CEOs and the general American public, this Chinese-made, open-source model landed with a jolt. The R1 launch showed that Chinese labs can produce a globally-relevant AI model. And then, to add insult to injury, Deepseek gave it away for free, directly competing with the subscription-based offerings of U.S. giants. What is an American lab to do?

Given the opacity of the Chinese government, it is difficult to separate rumor from reality surrounding the DeepSeek launch. But to understand DeepSeek’s sudden emergence, it helps to look at some of China’s policy choices. Over the last several years, the CCP has imposed regulations on finance, capping salaries and increasing scrutiny on hedge funds, with many observers speculating that these “crackdowns” are designed––or at least have the incidental effect––of nudging top-tier talent out of “non-strategic” fields and into strategic fields like AI. What is certain is that the Chinese government is investing heavily in AI—recently announcing a $137 billion commitment—and encouraging local governments to offer preferential loans and specialized infrastructure to “national champion” AI labs.

DeepSeek itself is a young company—spun out of, pointedly, a hedge fund—and launched about a year ago. Export controls on advanced GPUs and chip making equipment have been a central U.S. strategy to slow China’s AI progress. But the reality seems to be that Chinese labs can still acquire powerful hardware, if at a premium. Hardware leaks via intermediaries like Singapore and other routes still exist.

Export controls seem to have incentivized Chinese companies to develop more compute-efficient AI architecture. DeepSeek claims it trained R1 for around $5 million, a figure dwarfed by comparable U.S. models. Though some Western experts doubt the $5 million figure, R1 was clearly trained for a fraction of the cost––and with a fraction of the chips––as other frontier models.The DeepSeek debut suggests that, so far, US hardware control measures haven’t succeeded in decisively blocking Chinese development. 

Eventually, such controls may incentivize China to develop homegrown chips––turning what could, or should, have been a weakness for Chinese labs into a strength.

Another unexpected win for Chinese AI development has been on censorship. Americans would probably assume that a Chinese model would be heavily filtered in line with the goals of the CCP, while an American model would be more “free to think.” DeepSeek–R1 does incorporate content restrictions aligned with CCP priorities, from restricting content on “sensitive” historical events to pushing pro-China narratives. At the same time, it exhibits less of the political correctness that many American AI labs have adopted — but not none of it, as it seems to have been trained on US models.

Language control is now the norm in mainstream AI; the real question is always which topics get restricted and how strictly.

For U.S. policymakers and Silicon Valley leaders, the arrival of DeepSeek–R1 has underscored a prickly debate about “open-source” ethos versus national security for advanced AI. The American government, broadly speaking, does not want to share its best technology with China. The tricky part is how to maintain an open system in our own country while attempting to deny access to adversaries.

The open-source ethos behind DeepSeek–R1 heightens the tension between American tech giants and the public. Many see Big Tech as profit-driven monoliths hiding cutting-edge AI behind closed doors. When a Chinese firm hands out a free, open-source model, it can appear more “altruistic”—even if that narrative is oversimplified. Americans who resent Silicon Valley’s concentration of power may unwittingly embrace a Chinese alternative, ignoring potential national-security or data-privacy concerns. The New York Times reported in mid April that the Trump administration is debating barring Americans' access to its services.

Does open source help the average working class American? Well, yes. A freely available, high-performing model can be adopted, modified, and localized for specialized needs in education, healthcare, or research. It allows smaller American businesses to integrate cutting-edge AI without paying hefty fees to closed corporate labs.

***

At the nation level, what does “winning” look like in this race? Deepseek was a wake-up call that the Chinese development model actually seems to be working. For one, the centralized government approach seems to be working for AI development in China, where absolute focus is an advantage.  

But actions like China’s “reallocation” of finance talent into AI—whether deliberate or incidental—would be politically unthinkable in the U.S. We don’t forcibly limit salaries in one sector to push talent into another. An American flavor of centralization might look like the Apollo Program or the Manhattan Project—a bold effort to fund AI research, upgrade semiconductor manufacturing, reform visas to recruit global talent, and expand the energy infrastructure that large-scale compute demands. The U.S. has historically relied on private sector innovation over centralized state planning. The free flow of ideas and top-tier universities that attract global talent remain powerful assets. The U.S. can “brain drain” by offering world-class labs, a thriving startup scene, and higher salaries — without targeting Wall Street pay, mind you.

In particular for the U.S., is our objective in winning the AI race to improve global living standards through AI––to fulfill OpenAI’s original mission of “ensur[ing] that Artificial General Intelligence (AGI) benefits all of humanity”? Or is it to ensure American hegemony in strategic industries like defense and communications? 

Advanced AI has obvious military applications—from logistics to surveillance to autonomous weapons. And if the U.S. and China are nearing parity in large-model capabilities, then both sides may scramble to integrate these advances into defense systems.

But there is also a subtler dimension to what is at stake in an AI race: soft power. Once upon a time, American companies like Microsoft, Apple, and Google symbolized technological leadership worldwide: for both the West and East. Now, Chinese firms such as Huawei, ByteDance, and DeepSeek show that technological dominance is shifting, however slightly or slowly, Eastward—and at the very least diversifying.

For the U.S. to maintain influence, it must keep innovating and perhaps recapture the culture of open research that originally propelled it to the top. Ironically, as Western labs become more secretive, companies like DeepSeek can claim the moral and practical high ground with open-source releases. This tension between “national security” and “open innovation” will only grow more acute as AI’s importance expands.

At the end of the AI race, the unveiling of DeepSeek–R1 might fade from memory. But what is obvious, right now, is the accelerating pace of AI (both on the U.S. and Chinese side) and the urgent need for strategic clarity from the U.S.. The United States stands at a crossroads. Ready or not, the Everything Race has begun—and the question is whether America is ready to run.

The U.S. still holds enormous advantages: top universities, a culture of entrepreneurship, robust capital markets, and a proven ability to realign when crises arise. However, DeepSeek is a reminder that this lead isn’t guaranteed. China’s massive investment, large talent pool, and willingness to mobilize resources quickly show that it’s ready to compete head-on.

Ready. Set. Go.