The lead story last night on the CBS Evening News – and everywhere else for that matter – was about DeepSeek, the new Chinese Large Language Model (LLM) that performs as well (or better) than most U.S. models but was trained for a fraction of the price and costs a fraction of what ChatGPT, Google Gemini, Microsoft Copilot and other widely used models cost.
I turned on the news last night expecting that the DeepSeek story might get a mention, but was genuinely surprised when it was the lead story. AI breakthroughs that cause breathless coverage and wild predictions have become a weekly – sometimes daily – occurrence, but mostly in the echoey chambers where people like me spend our time.
This is different.
Here’s just a sampling of the headlines this morning:
Chinese AI threat triggers $1 trillion market crash
Chip Stocks Tumble After China’s DeepSeek AI Models Raise Doubts Over U.S. Tech Dominance
So what’s going on? Why is this such a big deal? I’m no DeepSeek expert (in fact, like most of the world, I had never heard of it before last weekend), but here are a few of the things at play:
U.S. AI companies are (or at least thought they were) in a domestic arms race
It was only a week ago that AI watchers were breathlessly reporting on Stargate, the new Administration’s $500BN investment in AI infrastructure and innovation. Sam Altman, CEO of OpenAI, was on hand for the White House announcement and said, “I'm thrilled we get to do this in the United States of America. I think this will be the most important project of this era.”
Altman has been competing with other tech billionaires for AI dominance, all of them representing U.S. companies. DeepSeek is the first non-U.S. AI tech to truly break through – and it just happens to come from the country’s top global competitor.
This changes the dynamic into one that suddenly has massive geopolitical implications.
Model training is crazy expensive…right?
LLMs like ChatGPT are only as good as the data they’re trained on. The more data you have, the more capable the AI is. And you need TONS of highly specialized computer chips churning away endlessly to train the data. All of which is very energy intensive (AI has been called an “energy vampire”). Hence the $500BN Stargate price tag and the rumors last year that Altman was seeking $7 TRILLION in funding for his AI vision (nuclear power, anyone?).
Nvidia has become one of the most valuable companies in the world by manufacturing many of the computer chips fueling the AI boom. And as a U.S. company, they’ve been able to limit China’s access to their chips, thus limiting China’s ability to keep up in the AI race…right?
One of the biggest reasons for the freakout of the last 24 hours is that DeepSeek is apparently trained on “inferior” computer chips at a cost of only $6M. Compare that to the $100M that Altman estimates that it costs to train one of his models and you can see why Nvidia’s stock led the plunge yesterday.
Model prices have been in an endless, upward climb
All of this leads to a high-and-growing price point for consumers. There is a free version of these tools, but to get the fancy new functionality, you need to upgrade to the paid version. ChatGPT Pro, the current OpenAI model that includes lots of impressive bells and whistles, is $20 per month per user. But OpenAI has already said it’s planning to increase the price to $44 per month within the next couple years. And if you want to get the Pro version of Sora, OpenAI’s new video tool, you’ll have to fork over $200 per month. You read that right.
So while OpenAI was founded on the principle of making AI available to and beneficial for everyone, it is quickly becoming a tool that fewer and fewer people can afford.
And OpenAI isn’t alone. Google Gemini, Perplexity, Grok and others are all charging similar prices for their advanced models. And enterprise licenses for Microsoft Copilot can be hundreds of thousands of dollars.
So if DeepSeek truly can do all or most of what these other tools can do at a fraction of the cost, why would the average person spend all that money on something that is becoming a commodity?
Which brings us to, you guessed it, data privacy
The U.S. government somehow just temporarily shut down TikTok, a Chinese-owned social media platform known for viral dance trends (among other things). We’ve heard for years that TikTok is stealing our data, fomenting discord and undermining the American way of life. Now imagine Americans using an LLM created by someone with a cozy relationship with the Chinese government. Doom-scrolling the latest TikTok videos will pale in comparison to people using a free LLM and uploading who-knows-what kind of information for our greatest geopolitical rival to see.
Summing it all up
AI is driven by an incredible amount of hype right now. It might be that six months from now we’ll have forgotten the name DeepSeek, and ChatGPT will be back on top of the App Store’s download chart. Or we could be witnessing the first real shot across the bow of the seemingly unbreakable hold that U.S. companies have on the future of artificial intelligence.