DeepSeek, this week’s hit AI app out of China, won’t replace teachers in the classroom anytime soon. As with any of its American rivals, the hallucination problem is still strong with this one.
But make no mistake, it is a teacher. With its significantly cheaper price, and a model that is comparable on benchmark tests to its smartest rivals, DeepSeek is already taking Silicon Valley giants to school.
No matter where Nvidia stock ends up after its current rollercoaster ride, the AI landscape beneath it has been permanently altered by DeepSeek. The wisest tech firms are the ones who will make adjustments while they still can.
Here are five main lessons that the rise of DeepSeek has already taught us.
American exceptionalism isn’t a thing …
Here was perhaps the biggest blind spot in Silicon Valley: Just because American companies have traditionally been at the head of the class when it comes to tech innovation, it was generally assumed that things would always stay that way.
China was a manufacturing powerhouse, and software engineers are increasingly in high demand there, but America has the design know-how and the entrepreneurial spirit that will win out every time, so no major changes are needed. Right?
Maybe someone in the U.S. AI industry should have asked someone in the U.S. auto industry how this complacent attitude to nimble overseas competitors worked out for them.
The U.S. may still have the world’s largest GDP, but China’s is still growing faster, and has been since 2020. The U.S. may be the investment capital of the world, but it seems Chinese hedge funds — like the one that bankrolled DeepSeek — can be just as disruptive.
DeepSeek “certainly isn’t the end of the story of U.S. exceptionalism,” one analyst told Yahoo Finance’s Morning Brief on Monday. Perhaps not, but it could well be the beginning of the end. No American tech giant that hopes to survive the 21st century can afford to act like America is the greatest country on Earth just because it’s America.
American protectionism shouldn’t be a thing either.
The second lesson is also the result of misguided patriotism. DeepSeek is the kind of foreign leap forward that the U.S. government has been trying to avoid for years, no matter which president was in charge. The CHIPS act pushed by the Biden administration sought to keep powerful processor technology from companies like Nvidia out of Chinese hands; Chinese companies had built a massive stockpile of them already.
Here’s the thing about protectionism: most economists and tech experts have been telling us for years that it just doesn’t work. If anything, the protectionist attitude of CHIPS seems to have woken a sleeping giant in China’s homegrown AI infrastructure development; DeepSeek needed to work on fewer and less elaborate chipsets. Necessity is the mother of invention, no matter the country. Chinese companies have been looking for every edge possible — including making AI models open source, a key advantage to DeepSeek.
American companies can’t even control how their own products are used abroad. Early versions of DeepSeek were reportedly built using Llama, the open-source Large Language Model created by Mark Zuckerberg’s Meta. The internet, and all the information sharing it represents, keep on flattening the playing field for business. American tech giants don’t exactly get an exemption for having built out the internet in the first place.
Mashable Light Speed
Building bigger isn’t better.
Why did the sudden appearance of DeepSeek cause the stock market to hammer Nvidia? Because Nvidia chips are the centerpiece of a data center surge planned, at great environmental cost, by just about all the major tech companies (except Apple).
The main reason for this surge was to slake the thirst of increasingly popular AI models for processing power (not to mention literal thirst for a still unknown amount of water).
But the big surprise of DeepSeek is that it is, as they say in Silicon Valley, hella efficient. This suggest that there are still major innovations to be found on the software side, not the data center hardware side. LLMs can learn to work smarter, not harder.
This increases the likelihood that in the future, every company might get away with using its own homebrew AI model. Or even that open-source AI models might eventually get small enough to fit on a smartphone. (In which case: Nice one, Apple.)
In other words, U.S. AI companies assumed the answer was to build the tech equivalent of SUVs. China just built an electric motorbike that can go as fast as an SUV. And on these metaphorical roads, the lightest vehicle wins.
AI marketing means next to nothing.
Poor OpenAI. With much of its tech leadership quitting, the company doubled down on marketing; witness the slick “12 days of OpenAI” production, where Sora and a lot of incremental product tweaks were given a friendly, fun, multi-day online keynote sheen by Sam Altman and friends. At the holidays! Who doesn’t love holidays? Who doesn’t want their product in the marketplace for the holidays?
And which brainbox would be dumb enough to launch their hot new product at the dead end of January, with absolutely zilch in the way of marketing beyond a weird-looking dolphin logo? DeepSeek, that’s who.
The app arrived with little fanfare during the inauguration; days later came its surprise benchmark results as reported by researchers. This made for eye-opening headlines, which in turn sent curious users to their respective app stores, which drove the stock market wild, which created more stories. That’s the mysterious force called buzz, and DeepSeek has it like no product we’ve seen since ChatGPT in 2022.
At time of writing, DeepSeek leads both the Google Play chart and the App Store, as it has all week. In the latter, it has pushed ChatGPT and Google Gemini into silver and bronze medal positions respectively. The 1,400 reviews on the App Store are led by one where the user has given DeepSeek four stars out of five, calling the model “mind-blowing” but docking it one star because “it would be really helpful if it had a landscape mode.”
In other words, we’re already at the “this is so great I’m annoyed it isn’t perfect in its first iteration” stage of DeepSeek fandom. There is no such thing as brand loyalty in AI consumer world. The level playing field of app stores reward the app that is the least hassle (including the hassle of paying for it) while performing at a level comparable to the industry leader.
We might distrust DeepSeek for its terms and conditions, but it’s not like we trusted ChatGPT in the first place. In both cases, we’re mesmerized by the way the thing itself can answer most of our questions. The wrapper matters little. And if that’s the AI landscape going forward, what use is American-style marketing?
Superstar CEOs are getting in the way.
You might also ask: What use is treating Silicon Valley CEOs like rock stars, or vengeful gods, if DeepSeek was brought into existence by one of the most obscure, self-effacing guys on the face of the planet?
In the U.S., tech leaders have to work really hard not to get high on their own supply of hype. They see themselves on magazine covers constantly; they get really respectfully interviewed on amped-up business news shows. Readers and viewers tend to assume they have all the answers. Biographers encourage this view.
The myth of the Steve Jobs-like tech celebrity genius is everywhere; just don’t look too closely at the details.
But DeepSeek just shone a spotlight on this myth. While it was quietly gathering steam in the past week, what was U.S. AI leadership doing? Making empty announcements and bickering like children.
Trump and OpenAI’s Project Stargate is still unclear about what it is, exactly, and where its supposed $500 million funding is coming from. The administration needed someone who looked like they knew what they were doing with AI, so they called in Sam Altman. But that upset fellow mercurial celebrity Elon Musk, who started feuding with Altman on social media.
Great job counteracting DeepSeek, guys, no notes. But seriously, if the little dolphin-logo app can help end the era where Silicon Valley followed the pronouncements of the latest cults of personality, that may be its hardest and best lesson of all.
