U.S.-China AI arms race increasingly becoming national security issue, Wolfe says
Earlier this month, U.S. President Donald Trump rolled out a new 25% tariff on certain artificial intelligence chips, including Nvidia’s H200 AI processor and a rival model from Advanced Micro Devices.
The announcement, which came after a nine-month investigation into the semiconductor industry, was viewed as part of a wider effort to incentivize U.S. chipmakers to build more products domestically, rather than rely on foreign companies, particularly in Taiwan.
Officials from the White House later clarified that the tariffs will be narrowly focused, adding that they will not apply to chips and other devices imported for use in U.S. data centers — many of which house the high-end processors which power AI models. Commerce Secretary Howard Lutnick will also have broad discretion around other exemptions, according to the announcement from the Trump administration.
The move comes after Trump vowed to place levies on imports of Chinese semiconductors, although he postponed the order until June 2027. The president also said last year that he would allow Nvidia to export its H200 chips to China in exchange for a portion of the sales — but questions have surrounded whether this would be in violation of a constitutional ban on export tariffs.
Still, the White House’s posture suggests the AI race between the U.S. and China has increasingly become a national security issue, according to analysts at Wolfe Research.
In a note, the analysts including Stephanie Roth argued that this was underscored by the COVID-19 pandemic, when constraints in the flow of chips highlighted global dependence on steady supplies of these processors.
“Leadership in AI matters for technological leadership, military capability, and economic growth,” the analysts wrote.
They added that while the U.S. remains the “clear leader” in training the biggest and most capable AI models, China “has made progress largely through state-directed, capital-intensive policies.” Beijing has also sought to compete through efficiency, optimization and faster diffusion of “good-enough” models under hardware constraints, the analysts said.
