DeepSeek threw a monkey wrench into the AI community.
TL;DR: Some of the biggest American companies, and some of the richest men in the world got pantsed by a 200 employee Chinese company.
When the Soviets launched Sputnik in 1957, it led to a crisis of confidence within the American populace. Large parts bubblegum and bailing wire, Sputnik would crash back into earth in three short months, but it shook the US, a country that considered itself number one. Hubris was the first casualty, and the space race was off and running.
DeepSeek has a similar potential. We have no idea how lasting their technology will be, but they did pop the balloon on the reputation of the great American techbros. Maybe they weren’t the smartest in the room after all.
A quick summary. Here’s where AI stood the day before DeepSeek was released:
- OpenAI, mainly by being first to market, is considered the market leader. But every major software company has their version. A good tell is the companies / people that anted $1 million dollars to Donald Trump’s inauguration fund: Amazon, Meta, Google, OpenAI, Microsoft, and Apple, among others.¹
- Nvidia is by far and away the leading chip maker for AI technology². However, Nvidia doesn’t make chips, it designs them. 90% of AI chips are made by one company, Taiwan Semiconductor Manufacturing Company. Yes, Taiwan, as in the small self-governed island that China claims as part of its territory.
- Donald Trump is picking winners and losers in the AI race, publicly touting the Stargate initiative that may eventually lead to a $500 billion dollar investment in AI technology³. This declaration was to Elon Musk’s chagrin as he is not involved in this joint venture.
- These AI technologies all use similar if not identical algorithms and require massive amounts of electrical power. For instance, a ChatGPT query uses nearly 10 times the electricity of a traditional Google search. This has led to aggressive participation in nuclear energy by the competing companies, including Google’s plan to own private nuclear reactors known as Small Modular Reactors (or SMRs).⁴
Then, shortly after Trump’s inauguration, DeepSeek was released:
- Developed by a company of 200 employees at an estimated cost of $6 million. It is also understood that none of these employees were educated in the US. ⁵
- DeepSeek used more modest, slightly older versions of Nvidia chips (probably due to US policy limiting microchip sales to China). This caused Nvidia to shed $600 million in value on the day of the DeepSeek R1 rollout.⁶
- DeepSeek punctured the concept of the United States dominance in AI technology. Who will eventually emerge has yet to be determined. ⁷
- It is believed that DeepSeek uses less energy in its learning modules, but it is uncertain if it is more energy efficient answering queries. It does produce equivalent or better results than the much bulkier LLM models. ⁸
Winners and Losers?
Winner: Necessity is the mother of invention. While the American companies just raced to the finish line, DeepSeek developed a radical new approach.
Losers: Our way or the highway. Some of the richest men in the world just assumed AI was a closed community. This is no longer true.
To Be Determined: Security is mostly a superstition.⁹ DeepSeek will have to overcome TicTok-like security concerns to truly succeed. Oddly enough, ChatGPT uses proprietary software while DeepSeek is open source. This may not allay security concerns, but it does upend any pricing models.
Marc Andreessen sees DeepSeek as an inflection point in AI 10. But maybe we could take a step back and view it more as an existentialist point. Regardless of the staying power of DeepSeek:
- Any new technology is a competition between be first and be right. DeepSeek exposed these companies as only interested in being first.
- Software tends to become bloated over time (Microsoft Office being a textbook example), to the point where consumers use a product despite itself rather than because of itself. Haven’t Facebook, Google Search, X and Amazon Prime reached this point? Will adding AI make it worse or better?
- Good software defines a problem and then develops a solution. What specific problem are AI chatbots solving? Or is it a solution looking for a problem?
² Motley Fool: Nvidia dominates the AI Chip Market.
³ CNN: Trump announces $500 billion infrastructure investment.
⁴ AP: Amazon and Google make dueling nuclear investments.
⁶ Yahoo Finance: Nvidia loses $600 billion in value.
⁷ CNN: DeepSeek could mark a turning point for Silicon Valley on AI.
⁸ Reuters: DeepSeek raises data center demand doubts.
10 Forbes: Marc Andreessen warns of AI’s Sputnik moment.
