James D. Spellman is principal of Strategic Communications, a consulting firm based in Washington.
Nvidia CEO Jensen Huang warns that U.S. efforts to block China from accessing advanced AI chips and software are backfiring -- a sharp pivot for a highly visible supporter of President Donald Trump.
China is now fully capable of developing its own AI tools despite restrictions designed to prevent its technological ascent, according to Huang. Washington's "fortress" strategy has accelerated Chinese innovation while escalating a broader geopolitical contest for hegemony over critical technologies and rare-earth minerals.
"The U.S. has based its policy on the assumption that China cannot make AI chips," Huang said during an earnings call in May. "That assumption was always questionable, and now it's clearly wrong. ... China's AI moves on with or without the U.S. chips. Shielding Chinese chipmakers from the U.S. competition only strengthens them abroad and weakens America's position."
China's stake in artificial intelligence is deeply moored. Venture funds backed by the Chinese government invested $184 billion into AI startups between 2000 and 2023, according to a Morgan Stanley report last month. The sector's value plus related industries, including infrastructure and component suppliers, could reach $1.4 trillion by 2030. "Six tigers" -- Stepfun, Zhipu AI, Minimax, Moonshot AI, 01.AI and Baichuan Intelligent Technology -- dominate among some 4,300 AI companies.
The Chinese startup DeepSeek upended investors' assumptions about AI's trajectory by demonstrating in January that high-performing models can be developed at a fraction of the stratospheric costs that tech giants are paying -- $5.6 million versus hundreds of millions, if not billions, more. DeepSeek used Nvidia chips to match OpenAI's quality and operating at nearly twice the speed.
On the scorecard Chatbot Arena, DeepSeek and Alibaba's Qwen lead the best open language models. By one count, China has already secured the top position in AI-related patents, research papers and citations. Yet, by another measure, the U.S. endeavors far exceed those of China. There are 690 privately held U.S. tech companies worth more than $1 billion, representing $2.53 trillion in combined value, compared to China's 162 businesses worth $702.46 billion.
Alibaba's Qwen is one of the most high-profile open language model in China. © Reuters
But the distance between the frontier and the herd is narrowing quickly, a Stanford University assessment concludes. The U.S. lead is shrinking as competitors leapfrog forward in education, capital markets, high-precision manufacturing and technological prowess.
"They're not scared," said Jamie Dimon, JPMorgan Chase's chief executive, after a recent trip to China. "They put 100,000 engineers on it." In January, Ruth Porat, president and chief investment officer at Alphabet, forecast that the U.S. is "probably a year-plus ahead in models."
Nvidia's quarterly filing admitted for the first time that restrictions on China will hurt business. In the current quarter, Nvidia said $8 billion in planned orders for its dumbed-down H20 chip must now be scrapped after Trump tightened export-licensing in April.
The disclosure reveals deep, wide-ranging interdependencies that also entangle other American chipmaking firms -- Qualcomm, Intel and Broadcom -- with the world's second-largest economy for up to 40% of revenues. China's semiconductor market is projected to reach $204.03 billion this year and continue growing at a compound annual growth rate of 8.24%.
Trade restrictions rarely achieve their intended goals. Instead, they often generate unforeseen and adverse consequences. We've witnessed that most recently with economic penalties on Russia after invading Ukraine. Barriers encourage the emergence of parallel ecosystems, the networks of businesses, workers, investors, customers and institutions that depend on and support each other. Protectionism may buy time -- but not necessarily victory.
