Published: May 14, 2026
A recent policy analysis sets out two possible states of the global AI competition in 2028 between the United States and the People’s Republic of China (PRC). The authors argue that access to compute — the advanced semiconductors needed to train and run frontier AI models — is the decisive input, and that democracies led by the United States and its allies currently hold a significant advantage. The paper illustrates how policy choices made today could determine which scenario materializes.
Why leadership matters
The report emphasizes that where the most capable systems are built will shape the rules and norms governing AI deployment. If authoritarian regimes that treat AI as a tool of repression or military advantage set the frontier, the risks to civil liberties, human rights and international stability grow. The Kínai Kommunista Párt (CCP) already uses AI tools for censorship, repression of dissidents, cyber operations and military modernization; frontier AI could make those capabilities cheaper, more pervasive and more sophisticated.
Frontier AI also accelerates other critical technologies (semiconductors, biotech, advanced materials), so a lead in AI can compound into broader strategic advantages.
How democracies built their lead
The analysis cites two main drivers of the compute lead:
- Commercial innovation: decades of R&D and engineering by companies such as NVIDIA, AMD, Micron, TSMC, Samsung and ASML across democracies.
- Public policy: sustained, bipartisan export controls and other measures by multiple U.S. administrations to limit PRC access to the highest-end AI chips and semiconductor manufacturing equipment (SME).
Current estimates indicate a widening compute gap. One analysis cited finds that Huawei will produce about 4% of NVIDIA’s aggregate compute in 2026 and 2% in 2027. Another study estimates that, if U.S. and allied restrictions hold, the United States could have approximately 11 times more compute accessible to its AI sector than China’s.
Two channels keeping PRC labs close to the frontier
Despite the compute disadvantage, PRC AI labs have remained relatively close to U.S. frontier models through two main workarounds:
- Illicit and evasive compute access: smuggling of chips into China and remote use of export-controlled chips via data centers outside PRC jurisdiction. The report cites a criminal case alleging diversion of $2.5 billion worth of servers containing advanced U.S. chips to China.
- Illicit model access and distillation attacks: generating numerous fraudulent accounts to access U.S. frontier models, harvesting their outputs, and training local models that mimic capabilities. Major industry actors — OpenAI, Google, Anthropic and the Frontier Model Forum — have publicly condemned distillation attacks.
Safety assessments show worrying patterns: the Center for AI Standards and Innovation (CAISI) found DeepSeek’s R1-0528 model complied with 94% of overtly malicious requests under a common jailbreaking technique, versus 8% for U.S. reference models. An independent review found Moonshot’s Kimi K2.5 failed to refuse CBRN-related requests at a much higher rate than U.S. frontier models.
The two 2028 scenarios
Scenario 1 — Democracies hold and expand a commanding lead (12–24 months):
- Stronger export controls, tighter enforcement against smuggling and foreign data-center accesses, and measures to detect and deter distillation attacks keep U.S. and allied compute advantages intact.
- U.S. frontier models are 12–24 months ahead of PRC models in intelligence. That gap gives democracies time to set norms, push global adoption of trusted AI infrastructure, and reap economic and security benefits.
- American AI underpins critical industries (cybersecurity, finance, healthcare, life sciences), strengthens defenses, and makes the democratic coalition more attractive to global partners, reinforcing the lead.
Scenario 2 — CCP-controlled ecosystem is neck-and-neck:
- If loopholes persist and distillation attacks continue, PRC labs can remain only months behind U.S. frontier models, aided by overseas compute access and other evasions.
- Beijing’s aggressive domestic adoption policies (“AI+”) and state-led deployment allow near-frontier systems to be integrated at scale across economic, military and technological domains, offsetting any small intelligence gap.
- The CCP’s integration of AI into cyber and military capabilities would pose a serious national security challenge: democracies may lose the security advantages they had despite having developed the technology first.
Policy recommendations to secure democratic leadership
To favor the first scenario, the paper endorses several policy directions:
- Close loopholes: increase enforcement budgets and tighten controls to prevent smuggling, shore up SME export controls, and limit remote access to export-controlled chips in foreign data centers.
- Defend innovations: legally and technically deter distillation attacks, clarify that such attacks are illegal, and enhance threat-intel and technical sharing between U.S. labs and government.
- Champion trusted American AI exports: promote adoption of democratic-aligned AI hardware and models abroad to deny the CCP low-cost global footholds.
Conclusion
The authors conclude that the United States and its allies today possess both the most capable frontier AI models and the most advanced inputs to AI, creating a substantial advantage. If access to that technology is defended through policy and enforcement, democracies can sustain and extend their lead into 2028. If not, the advantage could be eroded and a risky near-parity could arise. The paper stresses that decisions made by policymakers in the near term will be decisive for where the AI competition stands in 2028.
Note: the text references a January 2026 House vote in which the House passed a bipartisan bill 369–22 to close one of the cited loopholes; that bill had not yet passed the Senate at the time of publication.
