Preparing for ‍Second ⁢Place:‍ U.S. Strategy in⁣ the AI Competition

Updated June 14, 2025

Technology executives and U.S.officials agree: The United States must win the artificial intelligence competition with China. Jake sullivan, than-National security⁣ advisor, warned in October 2024 that the U.S. risked⁢ losing⁤ its lead if it didn’t deploy AI⁤ more quickly. A Trump administration executive order declared ⁤the goal to “sustain and enhance America’s global AI ⁢dominance.”

Washington has pursued a ‍two-pronged strategy: restricting China’s‍ access to key technology and accelerating domestic innovation. Both administrations have favored light regulation,targeted investment in semiconductors and‍ energy,and encouraging AI adoption across the federal government,including defence and intelligence agencies.

These policies have helped American firms maintain their lead in market share and‍ model performance. However, recent breakthroughs by Chinese AI companies like deepseek and alibaba Cloud suggest the gap is narrowing. American supremacy is not assured.

Even while vying for dominance, Washington needs to plan for ‍a future ‍where China’s AI models are⁣ equally popular globally. Preparing for second place doesn’t mean repeating the failures of the 5G competition.⁤ By promoting frameworks that account for AI’s appeal in emerging markets, easing model migration, building comparison ⁤systems, and securely sharing data,⁢ Washington ⁢can still benefit from the AI revolution.

blown Lead

Until the summer of 2024, the U.S. seemed to have a winning formula: rapid innovation fueled by academic research, private sector capital, and light regulation. U.S. models like OpenAI’s GPT and Google’s Gemini showed ⁤significant advancements. AI tools reduced “hallucination,” gained image and ⁢sound capabilities, mastered complex tasks, and enhanced‍ reasoning.

Companies like Anthropic, xAI,⁣ and Meta developed larger models capable of mastering benchmarking tasks. As American models demonstrated speed and mastery, their global market share grew.

By the end of 2023, U.S. models outperformed Chinese counterparts by double-digit percentages in ⁢accuracy.But China ⁢has quickly closed ‍the gap through ⁣government initiatives, AI education, research investment, coordination between Beijing⁤ and the tech industry, and public investment in data centers and semiconductor manufacturing.

These efforts narrowed the performance gap to single-digit percentages by the end of 2024. DeepSeek and Qwen have matched⁢ U.S. model performance, raising fears that the ⁤U.S. lead⁣ has evaporated.

China has also taken the lead in integrating AI into high-tech manufacturing. Xiaomi, for example, uses over 700 AI-guided robots in its Beijing factory to produce a new electric vehicle every 76 seconds. AI is widely used in ⁤Chinese cities for traffic ⁤management, surveillance, and law enforcement.

Moreover, U.S. export controls have proved⁤ less effective than expected. Beijing has used shell companies and stockpiled chips to circumvent controls and accelerated domestic chip development. chinese firms have also pioneered software techniques to optimize hardware and lower costs. Whether china is in the lead or not, the days of absolute American‍ AI dominance are over.

Making the Best of Second

American AI firms may still retain their global lead in building foundational AI models. OpenAI, with SoftBank and Oracle, announced a $500 billion AI‍ infrastructure project. Amazon, Meta, Microsoft, and Google‍ continue to invest billions in startups and AI labs.

Amazon AWS, Microsoft ⁣Azure, and Google Cloud make up over 60% of the global cloud market, a ⁤critical resource. But the pace of U.S. innovation could fall ⁤short ‍as advancements become more difficult,the U.S. slips in the talent⁣ competition, and ‍federal research funding is cut. The White House Office of Science and Technology has begun developing a new national ‍AI Action Plan, expected in July.

Policymakers should also plan for a world where competing AI ecosystems coexist. Washington ⁢can find alternative ⁤strategies to ⁣ensure the U.S. benefits from AI progress even without winning outright.

Frist, the U.S. should demonstrate the merits of its models to global markets. The National Institute of Standards and Technology‍ could promote new⁣ evaluation⁢ frameworks that incorporate clarity, ‍accountability,⁣ accessibility, cost, ⁢and ease of modification.

As more models emerge, users will want to avoid being locked into either the ⁢U.S. or China’s offerings. Minimizing migration costs between models will become attractive.The American AI industry ⁣can reduce the cost and complexity of transitioning by ⁢lowering ‍prices and reducing software modification. Washington could lead efforts to standardize application programming interfaces to lower costs and reduce dependence on any one country’s AI offerings.

Policymakers should also plan for a world in⁢ which competing AI ecosystems coexist.

As Chinese models grow more powerful, Washington can’t simply highlight the risks of censorship and espionage. Rather, U.S. companies must build systems that mitigate the risks⁢ of relying on any one model. Incorporating an intermediate abstraction layer can⁣ isolate downstream systems, making them more independent.

The adoption of Chinese models creates risks, including vulnerability to incorrect outputs and exposure of sensitive data. When building on foreign models, U.S. companies need to ensure their applications are resilient. U.S. firms should build adjudication systems capable of comparing responses from⁢ trusted and untrusted models,warning users about potential risks.

the U.S. should regulate the types ⁣of data shared with foreign model builders without blanket bans. There might potentially be instances where the benefits‍ of fine-tuning a ⁤Chinese model with U.S. data outweigh the risks. For example, if a Chinese AI tool makes more accurate medical diagnoses, American hospitals should use it, even at the risk of sharing⁢ patient data.

In those instances, U.S. firms can ameliorate privacy and security risks ‍by data anonymization and differential privacy.

Washington will ⁣want to⁤ standardize new evaluation metrics and develop guidelines for sharing data⁤ with allies. It will also need to provide technical and financial assistance to partners who lack the expertise to migrate between models ⁤and build adjudication systems.

Losing the Right Way

Washington⁢ faces an evolving global AI landscape where absolute dominance is no longer assured. American policymakers cannot simply rely on calls to win the AI race. A more responsible strategy would promote policies that help the U.S. thrive while preparing⁣ for the⁤ possibility of failing to achieve outright dominance.

Or else, Washington will face a superior competitor with increasing economic and ⁢military power enabled by AI, and ‍a domestic AI industry unable to keep up. Finishing second is not a death knell for American AI, but refusing to adapt would be.

What’s next

The White House Office of Science and Technology is expected to release a new national AI Action Plan in July,⁢ which ‍may provide further guidance on navigating the evolving AI⁢ landscape and ensuring U.S. competitiveness.