Study: the discourse on the "AI arms race" between the US and China is largely a construct, not a description of reality
An analysis in Tech Policy Press shows that the DeepSeek R1 model was framed in the US as a "Sputnik moment," and in China as "coming from beyond the horizon." The authors argue that the "AI arms race" is a narrative that legitimizes spending and export restrictions, not a documented fact.
Authors Shuxian Liu and David Gray Widder, writing in Tech Policy Press, compared how the US government, Chinese state media, and public discussions on Reddit and the Chinese platform Zhihu talked about the DeepSeek R1 model after its release in January 2025. They found that the same model was interpreted in completely contradictory ways: Senator Chuck Schumer, in the US Senate, called it a "Sputnik moment" and a warning of a loss of lead, while the Xinhua agency described it using the term héng kōng chū shì ("coming from beyond the horizon") as evidence that China had taken the lead. According to the authors, the same pattern repeats with every subsequent major Chinese model, including Kimi K3 from Moonshot AI, released this July.
The study's main thesis is that the "AI arms race" is not a fact that the discourse merely describes, but a construct that serves to legitimize specific decisions — spending, export restrictions, and institutions. These decisions subsequently become a "fact on the ground," which is then cited again as evidence that the race exists, and the cycle repeats; the authors call this process "discursive-material articulation." As an example, they cite the Nvidia H800 chip, on which R1 was trained: a House of Representatives committee report accuses Nvidia of circumventing export controls and diverting chips through Singapore, while Chinese state media claim that the sanctions only boosted domestic innovation and "broke the technological blockade."
On the topic of computing power, the study describes how R1 achieved competitive performance with significantly less compute than had been assumed for US models. Both governments nevertheless stuck to the narrative that more computing power is necessary — US officials argued for the need for energy dominance and faster data center construction, while Chinese media pointed to the Jevons paradox (cheaper models attract more users and thus more overall demand for compute). According to the authors, users on Reddit and Zhihu described this narrative as a game serving primarily investors and infrastructure operators.
The article further discusses a third narrative, model openness, but this part of the text was not available in the source. See the source article for details.
Why it matters
The study shows that decisions on export controls, data center investments, and public AI spending rely not only on technical facts but also on framing in media and politics, which itself becomes an argument for further steps. For anyone following reports on "leadership" in AI between the US and China, this means the need to distinguish between documented technical progress and the narrative that specific actors (governments, companies, investors) use to defend their interests.
Two audiences, two different impacts
What this means
For individuals
Reports on who is "leading" in AI between the US and China (including models such as DeepSeek R1 or Kimi K3) should be read with the awareness that the same technical event can be framed in opposite ways in different countries and serve different political or economic interests.
For a business
Companies deciding on investments in AI infrastructure, chips, or responses to export restrictions should treat narratives about an "AI arms race" as partly purpose-built constructs, not as an objective measure of technological lead, and should verify underlying data independently of political framing.
StrategyCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.