Policy

The AI Race Between US and China Is a Self-Fulfilling Narrative

New research shows how claims about competition drive infrastructure decisions that then validate those same claims, creating a recursive policy trap.

Omega Editorial · November 16, 2026 · 3 min read

Key takeaways

  • Claims about U.S.-China AI competition drive policy decisions that create infrastructure, which is then cited as proof the race was real—a self-reinforcing loop.
  • The same technical objects (chips, model weights, data centers) carry contradictory meanings depending on whether they're framed by U.S. officials, Chinese state media, or technical practitioners.
  • DeepSeek's efficient training challenged the scale paradigm, yet both governments doubled down on arguments for more compute and infrastructure spending.
  • Technical communities on Reddit and Zhihu showed more skepticism than officials, noting details like CUDA dependency and routine distillation practices that official narratives ignored.
  • The race framing forecloses questions about who benefits from AI development and who bears its environmental, labor, and social costs.

The recursive loop of AI competition

When DeepSeek released its R1 language model in January 2025, U.S. Senate Democratic Leader Chuck Schumer called it AI's "Sputnik moment." Chinese state media celebrated it as proof that China had leapfrogged ahead. The same technology carried two contradictory stories: a threat requiring containment and a breakthrough demanding celebration.

A new comparative study examining government statements and public technical discussions in both countries reveals something more fundamental than who's ahead. The "AI race" functions less as an objective description of reality and more as a narrative that authorizes specific policy choices—export controls, infrastructure subsidies, restrictions on open models—which then become physical facts that appear to validate the original competitive framing.

Researchers analyzed how U.S. officials, Chinese state media, and technical communities on Reddit and Zhihu discussed DeepSeek after its release. The comparison, first reported by TechPolicy.Press, exposes what they call "discursive-material articulation": claims about competition drive spending and restrictions, which create infrastructure, which is then cited as proof the race was real all along.

Three contested narratives

The study identifies three circuits where technical objects become political symbols:

Sovereignty through chips. DeepSeek trained on Nvidia H800 processors—chips built to reduced specifications to comply with U.S. export controls. A House Select Committee report treated this as evidence of leakage requiring tighter restrictions. Chinese official media framed the same hardware as proof that sanctions only spur domestic innovation. Engineers on Zhihu noted the model still depended on Nvidia's CUDA software. Reddit users tracked Nvidia's $600 billion market value loss. The chip's meaning shifted depending on who held it.

Scale as self-justification. DeepSeek achieved competitive performance with far less compute than U.S. models were assumed to require. Rather than questioning the scale paradigm, U.S. officials argued efficient Chinese models made American energy dominance and faster data center buildout more urgent. Chinese state media initially criticized U.S. "brute-force scaling," then invoked the Jevons paradox: cheaper models attract more users, requiring more total compute. The case for scale survived evidence against it.

Openness as weapon or commons. When DeepSeek posted model weights to Hugging Face, U.S. officials described it as a national security threat. Chinese media called it a gift to the world. This summer, a former White House AI adviser warned that open-weight models lead to "full AI communism." Days later, Nvidia, Microsoft, and Meta signed a letter arguing open weights drive "American sovereignty and prosperity." The same file carried opposite meanings, both judged by whether they helped the U.S. lead.

Why it matters

Policymakers typically evaluate export controls, data center subsidies, and open-weight regulations by a single test: whether they keep one country ahead. This framing forecloses basic questions about what gets built, for whom, and at what cost. Once chips are fabricated, grids extended, and weights distributed, reversing the infrastructure loop becomes nearly impossible.

The engineers and hobbyists in the study shared neither government's certainties. They tracked technical details official accounts omitted and provided deflationary grounding for breathless claims. Their skepticism suggests the race narrative serves political and investment interests more than it describes technical reality.

The preoccupation with winning leaves unexamined who defines victory, who benefits from it, and who bears its costs in labor exploitation, environmental degradation, and economic precarity.

This analysis was originally published by TechPolicy.Press, drawing on research comparing official and public discourse in both countries following DeepSeek's release.

  • #us-china relations
  • #ai policy
  • #export controls
  • #deepseek
  • #geopolitics
  • #open source ai

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Policy