Ben Buchanan frames the AI competition as a “bitter struggle” with three linked challenges: ensuring democracies shape AI’s development, turning technological advantage into economic and national-security strength, and pursuing both in line with democratic values.

He argues that AI presents a new policy challenge because it has emerged primarily from private companies rather than government-funded research, leaving the government with less direct capacity to understand, shape or secure its development.

A January 2023 White House demonstration of GPT-4 marked a turning point for policymakers, Ben Buchanan says, showing AI’s potential impact on national security, foreign policy and domestic affairs; the demonstration used image analysis to examine a Vladimir Putin press conference and identify questions for intelligence analysts.

Buchanan’s “bitter lesson” is that broadly capable algorithms can improve with more computing resources; GPT-3’s advances in 2020 convinced him that AI progress could become an industrial-scale equation involving computing power, chips and energy.

As AI increasingly assists AI research, computing capacity could become even more valuable and compound existing advantages, though evidence of faster progress remains uncertain: public benchmarks have not clearly accelerated, and humans still provide much of the research judgment.

Buchanan favors U.S. and democratic preeminence over a close AI arms race with China, arguing that a substantial lead could give democracies more room to regulate companies and prioritize safety without intense geopolitical pressure.

That strategy must contend with concerns that competition could weaken safety work, including calls from Anthropic CEO Dario Amodei and Sam Altman to slow capability gains and warnings from AI researchers about catastrophic risk; the article argues a larger U.S. lead could give policymakers more room to slow down safely.

He supports export controls that limit China’s access to advanced computing, describing them as a way to slow China rather than push the United States into reckless acceleration; Chinese AI firms have identified limited computing access as a constraint, he says.

Despite repeated evasion of U.S. controls, China remains far behind in access to AI computing: Epoch AI estimates that 290,000 to 1.6 million H100-equivalent chips were smuggled into China through 2025, while its tracker estimates the United States has received about 72 percent of global AI compute since 2022, compared with roughly 10 percent for China.

China could still compete through strengths in model efficiency, adoption, robotics, manufacturing and electricity, so the U.S. advantage will compound only if the underlying inputs remain uneven.

Buchanan rejects selling advanced chips to China as a way to keep U.S. firms ahead, while acknowledging Nvidia’s argument that sales generate research revenue and preserve Chinese dependence on American technology; he says strong AI demand has undercut claims that excluding China would prevent U.S. companies from prospering.

Building U.S. AI capacity will require expanding data centers while addressing local opposition, which the article says is increasingly constraining development alongside power and permitting limits.