The United States is spending roughly $1 trillion annually on artificial intelligence development, while China invests 50 to 80 billion dollars per year and achieves performance within 2 to 3 percent of American benchmarks. This stark disparity has prompted a fundamental question among technology experts: Is the US chasing the wrong strategy in what many call the AI race?

Is the AI Race Actually a Race?

The dominant narrative in Washington and Silicon Valley frames AI development as a winner-take-all competition where the first nation to achieve artificial general intelligence (AGI), a system as intelligent as an average human, will gain decisive strategic advantage. But this framing may be misleading policymakers into wasteful spending and counterproductive restrictions.

Alvin Wang Graylin, Senior Fellow for Technology at the Asia Society and a researcher with 35 years of experience in AI and cybersecurity, challenges this assumption directly. He explained that the narrative assumes whoever reaches AGI first will use recursive self-improvement, where the system teaches itself to become exponentially smarter, eventually reaching artificial superintelligence that could dominate global systems. "I think it's a false narrative, and this narrative is actually driving us to make a lot of bad decisions in terms of overspending, in terms of over-indexing on controlling compute," Graylin stated.

"The United States needs to re-examine the race we think we're running, the game we think we're playing, and the strategy we've chosen. Because right now, the evidence suggests that our current strategy may be incomplete or over-concentrated on a single dimension of competition," Graylin said.

Alvin Wang Graylin, Senior Fellow for Technology at the Asia Society

Where Is the Real Competition Happening?

The current US strategy focuses heavily on scaling, building larger models and acquiring more graphics processing units (GPUs), the specialized chips that power AI training. But Graylin argues this approach misses where genuine innovation is occurring. Recent advances come from techniques like quantization, which compresses models to run on smaller devices, different memory management approaches, and using specialized chips for inference, the process of running a trained model to generate outputs.

These optimizations can deliver 5 to 10 times better performance improvements compared to simply adding more computing power, which follows logarithmic scaling laws. In other words, doubling compute capacity might only improve performance by a small percentage. The US is approaching physical limits on data center expansion, making pure scaling increasingly impractical.

Meanwhile, China is demonstrating capability in areas beyond raw model size. In June 2026, Beijing-based Z.ai released GLM-5.2, an AI model that drew admiration from Silicon Valley observers for completing complex tasks with minimal prompts at a fraction of the cost of American counterparts. The model ranks fifth on Artificial Analysis' intelligence leaderboard and second on Code Arena's front-end coding rankings.

Z.ai founder Tang Jie stated the model matches Claude Opus 4.8, Anthropic's advanced model, with goals to reach the capabilities of Claude Fable 5, Anthropic's most powerful publicly available model, by the first quarter of 2027. Some observers called it a "mini DeepSeek moment," referencing January 2025 when China's DeepSeek chatbot surprised the industry by achieving cutting-edge performance without access to advanced Nvidia chips that the US deliberately restricted.

Tang Jie

How Are the US and China's Advantages Actually Different?

Rather than a simple lead-or-lag scenario, the US and China possess distinct but complementary strengths. Kyle Chan, research fellow at Brookings Institution specializing in Chinese technology policy, explained that the US dominates the "virtual world" of research advancement, model development, and the software ecosystem surrounding AI. China's advantage lies in the "physical world" of manufacturing and supply chain innovation.

This division matters because it suggests the competition is not a single race but multiple parallel competitions across different domains:

• Model Development and Research: The US leads in advancing frontier AI research and creating large-scale language models through companies like OpenAI and Anthropic.
• Manufacturing and Supply Chains: China excels at optimizing production processes, managing semiconductor supply chains, and developing alternatives when access to Western chips is restricted.
• Cost Efficiency: Chinese AI developers are achieving competitive performance at significantly lower computational cost, suggesting superior algorithmic optimization.
• Open-Source Models: The availability of open-source versus closed models affects global adoption patterns and determines which systems run on consumer devices versus data centers.

What Recent Incidents Reveal About US Policy Uncertainty

Recent events suggest US AI policy is becoming erratic, potentially undermining American interests. In early June 2026, Anthropic released Claude Fable 5, a public version of its most advanced model, Claude Mythos, which had been withheld because it could detect software vulnerabilities and assist in hacking. Three days later, the US government ordered Anthropic to cut off access to both models for anyone who was not an American citizen. Unable to easily verify user nationality, Anthropic shut down the models worldwide.

The Trump administration cited a discovered "jailbreak" that could coax the model into producing dangerous information. However, some observers linked the shutdown to a February dispute when Anthropic declined to let the Pentagon use its AI without restrictions on surveillance and weapons applications, prompting Trump to order government agencies to stop using Anthropic products. The administration partially eased restrictions on July 1, 2026, allowing Anthropic to restore access to Fable.

Additionally, US Commerce Secretary Howard Lutnick told senior leaders at Dutch chip-equipment maker ASML that Washington suspected one of its extreme ultraviolet lithography machines, used to manufacture the most advanced processors, may have reached China in violation of export controls. ASML denied any breach, stating in an internal document that all 314 EUV machines in operation worldwide were accounted for outside China. Senior US officials told Bloomberg they had evidence ASML shipped specialty equipment linked to EUV systems to China, though they declined to make evidence public.

A third dispute erupted when Anthropic accused Alibaba Group of illegally accessing its AI model, sending Alibaba's Hong Kong-listed shares to a 16-month low and causing shares in other Chinese AI developers to fall more than 3 percent. Bloomberg Intelligence analyst Robert Lea said these episodes signaled that "Chinese AI models face an elevated risk of a US ban."

Why the "AGI First" Narrative May Be Misleading

Graylin emphasized that intelligence does not equal omniscience or omnipotence. Even a superintelligent system confined to a room with limited information cannot break into banks or disable power grids without access to real-world data, tools, and experimental capabilities. Breakthroughs require time, access to information, and the ability to conduct both thought experiments and physical experiments to validate new knowledge.

The "fast takeoff" concept, where AGI rapidly becomes superintelligence through recursive self-improvement, has driven US policy toward what Graylin calls "decisive strategic advantage" thinking. But this assumption lacks empirical support and has led to overprotection of AI models, making the US an unfriendly partner to the rest of the world and potentially driving innovation elsewhere.

The AI competition between the US and China is not a simple race with a finish line. It is a multidimensional contest where different nations excel in different areas. The US advantage in research and model development is real but may not translate to global dominance if China continues optimizing efficiency and manufacturing capabilities. Meanwhile, erratic US policy decisions risk alienating allies and pushing innovation into less regulated jurisdictions, potentially undermining the very strategic advantage American policymakers seek to protect.