The two most surprising things that have happened over the past week or so are (1), the explosive arrival of Chinese AI app DeepSeek onto the world stage, and (2), the fact that President Trump has gone strangely silent on China at the same time.

Before getting into DeepSeek--a "low-cost" ChatGPT rival that is cratering shares of NVIDIA by 13% today--let's talk about the president's pivot. Instead of spending his first days and weeks going after and slapping tariffs on China, Trump instead has focused primarily on Mexico, Canada--and yesterday, Colombia.

We had a fascinating discussion with Derek Scissors of AEI about this on Friday, who thinks it's no temporary oversight.

"I think it's very, very risky," he warned. "Mexico is the biggest trade rival of China. If you put tariffs on Mexico," as the president has threatened to do on Feb. 1., "you automatically get more Chinese goods here. More dependence of the U.S. on China. And I don't think the president understands that." 

But isn't the president's signature campaign promise the 60% tariff threat on China, I asked? Scissors said that could prove more talk than reality. Trump, he added, has been under pressure by voters to be more aggressive on China--especially after he didn't retaliate strongly when Covid broke out. Hence the big threats during the campaign.

During the first administration, however, there was also a clear strategy to get tough on China, led by Robert Lighthizer. "We don't see any of that now. Trump did not emphasize China during the transition," as he did the first time around, Scissors said. "We don't have an experienced trade person at the helm. So if there's a turn toward China, it's going to take a long time, and we don't see the start of it yet." 

And now, we have the breakout success of a Chinese AI app that is upending the whole narrative of "U.S. exceptionalism" in big tech. The S&P 500 is up 24% over the past year, while the rest of world ETF (ACWX), is up just 7%, as our Fred Imbert notes. Whether or not it really cost only $5 or 6 million to make (and that was only for the final training run, and excluding all other costs, as Ben Thompson notes), it casts the prevailing narratives--we need Stargate's million GPUs to keep our edge!--into question. 

Also, China has been under strict U.S. sanctions on leading-edge chips (like NVIDIA's H100s), and was able to make a successful AI chatbot anyway. Some have cast doubt on this, suggesting theft on some level must have been at play, but Thompson thinks it's reasonable they were using the "permitted" H800 chips. Their resulting R1 model is "definitely competitive with OpenAi's 4o, and Anthropic's Sonnet 3.5, and appears to be better than Llama's biggest model," he writes. 

And yet, shares of Meta, which developed Llama, are fractionally higher today. That's because he and many others believe Meta "is the biggest winner of all" from AI, and if DeepSeek just shows the next rounds of inferencing will be dramatically less expensive than the training stage, that will benefit companies like Meta (and Apple) in the long run. 

Where does it all leave Nvidia? With some questions around near-term financials, but still as a key player in the AI world for the foreseeable future. Where does it leave the U.S. relative to China in the AI race? DeepSeek's success, Thompson warns, "will result in a further unleashing of Chinese innovation as they realize they can compete." 

And the U.S. chip ban, if anything, might have backfired and resulted in DeepSeek's innovation with inferior chips. So we can either keep ramping up the export controls and hoping to slow their lead, or we can double-down on trying to keep our competitive edge. With the president distracted, it's looking like innovation will be the best--and only--option for now. 

See you at 1 p.m!

Kelly

Twitter: @KellyCNBC

Instagram: @realkellyevans