USA vs. China: A Big Artificial Intelligence Showdown, With One Possible Newfangled Solution

Received just as I was starting to write this post, in Anna’s Daybreak News on August 20th:  “China is closing the AI gap with the U.S., and in some areas leads on usage and cost.  Chinese models now account for 41.4% of generative-model downloads on Hugging Face and over 60% of global market share on OpenRouter…  U.S. firms still hold a computing edge” (credited to Bloomberg).

Fine, but what does all that really mean?  And what’s been in the press since early May?

A viewpoint from Jacob Dreyer, in the May 9^th New York Times: “America’s A.I. is Futuristic.  China Is Just Making It Work.”  The author, living in Shanghai, said that ‘In China, surveillance technology and A.I. surround our everyday life,” from obtaining groceries from seeing pictures of his 3-year-old “at school,” in which “an artificial intelligence facial recognition feature puts a red square around his face.”  So are the Chinese superior at AI?  Not quite.  “Americans want to create the most powerful technology humans have ever known,” but “the Chinese want to make A.I. more practical and embedded in society,” resulting in “improving the country’s domestic quality of life” and “exporting Chinese influence.”  For better or worse, it is being used in ways that could not be implemented in America.  That means there are two races going on, each with its own leader.

Soon afterwards, we read that “China Seeks A.I. Independence, Weakening Trump’s Leverage” (Meaghan Tobin, The New York Times, May 12^th).  This was about Chinese company DeepSeek announcing that “its new model had been optimized to run on chips made by the Chinese giant Huawei,” removing American ability to control their processor supply.  The chips could also be made elsewhere, as “DeepSeek is calling out into the void to… other companies, ‘Please make these changes so we can get better performance out of your chips.’”

On technical quality, “Chinese A.I. Models Close the Gap With Anthropic and OpenAI” (Cade Metz, Karen Weise and Meaghan Tobin, The New York Times, June 25^th).  That’s a consensus of intuitive estimates, as though this piece printed out to ten pages it offered no substantive or quantitative justifications.  Not so different in “American A.I. Companies Say Chinese Copycats Are Quickly Catching Up (Cade Metz, July 6^th, also in the Times).  Both pieces name a presumably estimated six-month AI development difference and mention “distillation” in which “researchers… collect data from a particularly powerful system and use that data to build a system that can run on less expensive hardware” - the second one concedes that “it is… not completely clear what Chinese companies are doing.”

After that, two of these authors told us that “China’s Latest A.I. Breakthrough Threatens America’s Lead" (Meagan Tobin and Cade Metz, July 17^th, still the Times).  They perceived that “the Chinese startup Moonshot AI released a new A.I. model that appeared to narrow the lead held by well-funded American competitors.”  But here, “according to benchmarks run by Vals AI, an independent company that evaluates the performance of A.I. models, Kimi K3 performs just below Anthropic’s Fable 5 model while outperforming OpenAI’s flagship GPT-5.6 Sol model.”  The authors do not blame distillation, but, per “experts,” “years of operating under export controls have forced Chinese companies to innovate within tight computing constraints.”  In “Chinese A.I. Start-Up Shows the World What It Has Built” (July 27^th, again in the Times), Tobin related that Kimi K3 has now been internally revealed, and “the model’s release has also inflamed a fierce debate in Beijing, Silicon Valley and Washington about whether governments should take steps to limit access to Chinese open-source models.”  There is irony there, if not a true paradox.

If all this is as significant as these “experts” say, it was valuable for Jason Hsu to tell us, in the New York Times on July 23^rd, “This Is How America Trounces China in the A.I. Race.”  He said our government issuing “a ban of its own,” in response to proposed Chinese ones, would be “a mistake,” but that “American and its allies should build and deploy their own so-called open-weight models, which users can download, run and modify on their machines.”  That refers to “the billions of individual settings… that help determine how a model responds to prompts,” which most companies including OpenAI keep proprietary.  So, yes, if there is a way to stay ahead of China, which seems now a worthy goal even if our objectives are different, it’s even worth rolling out a new-to-me buzzword.  Fair is fair.