Losing lead but not race as Washington adapts to China’s AI rise Analysis by Foreign Affairs

The United States’ lead in artificial intelligence (AI) race with its main competitor, China, is slipping. In a Foreign Affairs analysis, the publication warns that while Washington has so far succeeded in maintaining its AI dominance through innovation and export restrictions, it must now prepare for a world where that lead may no longer hold. The article urges a shift in strategy—from simply trying to win at all costs to building a resilient AI ecosystem that can thrive even if China catches up or pulls ahead.

Back in October 2024, US National Security Adviser Jake Sullivan cautioned that the United States risked “squandering [its] hard-earned lead” if it failed to deploy AI more comprehensively. The second Trump administration doubled down with an executive order aimed at “sustain[ing] and enhanc[ing] America’s global AI dominance.” Washington’s two-pronged strategy—constraining China through export controls and boosting domestic innovation—has helped American firms like OpenAI, Google, Meta, and others remain global leaders. Their models have outperformed Chinese counterparts in both capabilities and market share.

But the gap is narrowing. Recent advancements by Chinese tech giants like DeepSeek, Baidu, Alibaba Cloud, and Tencent indicate that China is quickly catching up. While US models have grown more powerful and accurate by reducing hallucinations, handling multimedia, and demonstrating complex reasoning, China has responded with government-backed research, close tech-state coordination, and massive investments in data infrastructure. Some Chinese models have now matched the performance of top US systems.

Moreover, China is outpacing the US in integrating AI into real-world applications. In Beijing, Xiaomi uses over 700 AI-guided robots to manufacture electric vehicles in under 80 seconds each. Across Chinese cities, AI is widely used in law enforcement, traffic control, healthcare, and education. This deployment edge, combined with growing technical parity, suggests that America’s era of unchallenged AI supremacy may be ending.

Meanwhile, US efforts to stifle China’s access to advanced chips have not fully worked. China has stockpiled semiconductors, used shell companies to bypass restrictions, and accelerated domestic chip development. Its engineers are also innovating on software that maximizes the efficiency of existing hardware. As Foreign Affairs notes, the playing field is leveling.

Yet America is not out of the game. US firms still dominate the cloud computing sector, with AWS, Azure, and Google Cloud accounting for over 60 per cent of the global market—critical for training large models. Major investments continue: OpenAI, SoftBank, and Oracle recently launched a $500 billion infrastructure project called Stargate. The White House is also preparing a new national AI Action Plan. But continued innovation may not be enough if the pace slows, federal funding drops, or top talent migrates elsewhere.

What should Washington do if it finishes second? Foreign Affairs suggests a smarter path: embracing the possibility of AI multipolarity while safeguarding US interests. One key recommendation is to shift how AI models are evaluated globally. Current benchmarks focus on capabilities like reasoning and language understanding. But newer frameworks should also consider transparency, affordability, adaptability, and openness. Promoting these richer evaluation standards could give US models an edge in emerging markets.

Another priority is making it easier for users to migrate between AI models. If global customers feel locked into either US or Chinese ecosystems, switching becomes costly. By standardizing application programming interfaces (APIs) and minimizing software and hardware changes during migration, Washington can make American AI offerings more flexible and appealing.

To further reduce dependency on any one system, US firms can use an “intermediate abstraction layer”—software that sits between the application and foundational AI model. This allows apps to quickly switch models if needed, protecting them from disruptions, espionage, or degraded performance in foreign models.

Still, using Chinese models—as they become more powerful—poses real risks: biased outputs, privacy violations, or system failures. To mitigate this, it is recommended to create adjudication systems. These would compare the outputs of trusted and untrusted models, warning users of errors or threats before harmful information spreads. While costly and slower, such tools are essential for high-stakes sectors like healthcare, finance, or infrastructure.

Finally, the article urges Washington to develop smarter data-sharing policies. Blanket bans on sharing US data with foreign models may backfire. In cases where Chinese AI tools outperform American ones—say, in medical diagnostics—it might make sense to use them, provided proper safeguards are in place. Techniques like anonymization, masking, and differential privacy can reduce the risk while still benefiting from advanced capabilities.

Preparing for second place is not the same as conceding defeat. Rather, it is about ensuring that the United States can still shape and benefit from the global AI revolution—whether or not it stays on top.

By Sabina Mammadli