Is the US Losing Its Edge in AI?

In reports to the US government, the companies warned that America's advantage in AI is narrowing as advanced models like China's DeepSeek R1 grow.

These reports were submitted in March 2025 in response to a call for comments on the government's AI Action Plan.

Increasing presenceThe Rise of AI in China

DeepSeek R1, an AI model from China, has attracted the attention of American developers. OpenAI sees DeepSeek as evidence that the technology gap between the US and China is gradually narrowing.

The company describes DeepSeek as “state-subsidized, controlled, and freely available,” and expressed concerns about China’s ability to shape the development of global AI.

OpenAI even compared DeepSeek to Chinese telecommunications giant Huawei, warning that Beijing's regulations could allow the government to force DeepSeek to interfere with sensitive systems or critical infrastructure.

OpenAI also expressed concerns about data privacy, noting that the requirement to share information with the Chinese government could strengthen the state's surveillance capabilities.

Anthropic, another US AI company focused on biosecurity, stressed that DeepSeek R1 “can answer most questions about bioweapons, even if it is designed to avoid dangerous content.” This is in contrast to the strict safety protocols applied in US-developed AI models.

Competition isn’t just about DeepSeek, however. Chinese search giant Baidu recently launched two powerful AI models, Ernie X1 and Ernie 4.5, to challenge the most advanced Western systems.

According to Baidu, Ernie X1, a reasoning model, has the same performance as DeepSeek R1 but at half the cost. Meanwhile, Ernie 4.5 costs only 1% of OpenAI's GPT-4.5 but outperforms it on some benchmarks.

Both OpenAI and Anthropic see this race as not just technological but ideological, a competition between “democratic AI” based on Western values ​​and “authoritarian AI” subject to state control.

However, the recent successes of Baidu and DeepSeek show that cost and accessibility can play a decisive role in the global spread of AI, transcending ideological differences.

Concerns about AI security and infrastructureAmerica

US tech companies have also expressed concerns about security and infrastructure challenges associated with AI development.

In its report to the government, OpenAI focused on the risk of Chinese government influence over models like DeepSeek, while Anthropic highlighted the biosecurity threat posed by AI's increasingly advanced capabilities.

Anthropic revealed that their Claude 3.7 Sonnet model has shown significant improvements in supporting bioweapons development, highlighting the dual nature of advanced AI systems.

At the same time, the company also pointed out loopholes in US export control policies. While Nvidia's H20 chips are still compliant with export regulations, they still have strong performance in text generation, a key element in reinforcement learning.

Therefore, Anthropic calls on the government to further tighten controls to prevent China from gaining a technological advantage.

Google, meanwhile, took a more balanced view. The company acknowledged the security risks but warned against the risk of over-regulation.

According to Google, overly strict export controls could undermine US competitiveness, especially for domestic cloud service providers and AI development companies.

Instead of imposing broad restrictions, Google proposes targeted controls that protect national security without hindering industry growth.

OpenAI, Anthropic, and Google all stressed the need for increased government oversight of AI security. Anthropic called for expanding the AI ​​Safety Institute as well as strengthening the role of the National Institute of Standards and Technology (NIST) in assessing and mitigating AI-related threats.

Economic competitiveness and energy demand

Reports to the US government also highlight the role of economic factors in the development of AI. Anthropic warns of infrastructure challenges, predicting that by 2027, training an advanced AI model alone could consume up to 5 gigawatts of electricity.

To address this, the company proposes building 50 gigawatts of dedicated AI power capacity and speeding up the transmission line approval process.

Meanwhile, recent announcements from Baidu highlight the importance of cost-effective AI. Models like Ernie 4.5 and X1 are said to cost a fraction of their Western counterparts, with token processing costs significantly lower than OpenAI’s current models.

This pricing strategy could put pressure on US AI companies to reduce costs to maintain a competitive edge. OpenAI describes this as an ideological competition between Western and Chinese AI, arguing that a free-market approach would spur innovation and deliver greater benefits to consumers.

Google, in its reports, focuses on more practical policy recommendations. The company calls on the federal government to increase investment in AI research, expand access to government contracts for tech companies, and simplify export controls to support the growth of the industry.

Management strategiesWHO

A unified strategy for AI regulation emerged as a common thread across all three submissions. OpenAI proposed a regulatory framework overseen by the Commerce Department, warning that fragmented regulations across states could accelerate the trend of AI development going offshore.

The company also supports a tiered export control model that would allow democratic countries broader access to US AI technology while restricting access for other regimes.

Meanwhile, Anthropic called for tighter export controls on AI hardware and training data, stressing that even small improvements in model performance could give China a significant strategic advantage.

Google has taken a different approach, focusing on copyright and intellectual property issues. The company argues that its current policy, based on the principle of “fair use,” is fundamental to AI development, and warns that overly strict copyright rules could weaken the competitiveness of American companies against Chinese rivals.

All three companies emphasize the need to deploy AI more quickly in the public sector. OpenAI proposes removing administrative barriers to AI testing and procurement, and supports Anthropic’s campaign to simplify the process of acquiring technology from federal agencies.

Google also agreed with these reforms, emphasizing the importance of improving interoperability in government cloud infrastructure to promote AI development more effectively.

Maintain competitive advantage

The submissions from OpenAI, Anthropic, and Google all reflect a shared concern about protecting America's leadership in AI, especially as China increases competitive pressure.

The rise of DeepSeek R1 along with advanced models from Baidu not only poses technological challenges but also puts pressure on cost and accessibility.

As AI continues to advance at a rapid pace, the U.S. government will face the dilemma of balancing national security, economic growth, and maintaining leadership in the global technology race. This will likely remain one of the most important policy challenges in the coming years.