In his first month, President Donald Trump let loose a barrage of U.S. policy pivots. But I dare say Trump was trumped by China’s DeepSeek, a free AI-powered chatbot that began modestly as a side project inside a $8 billion hedge fund. On Jan. 27, DeepSeek appeared like a meteor over the Chinese New Year, shaking billions in AI investments around the world. The AI world hasn’t stopped quaking.
Three questions to think about:
Deeper Meaning of DeepSeek?
DeepSeek signals a change from raw spending to smart design, says Ashu Garg, a venture capitalist with Silicon Valley’s Foundation Capital. Officially, the Chinese startup says it trained a cutting-edge AI model for just $6 million in GPU costs. Garg calculates an additional $500 million in previous hardware investment. Be a skeptic. Round the total investment to $1 billion if you like. DeepSeek is still crazily more efficient. American rival OpenAI says it needs to raise $40 billion to get to the next level.
DeepSeek’s cost disruption is just the beginning. AI’s future will be defined by fast, low-cost ways to solve complex problems rather than just automate tasks. Garg says this mirrors past shifts in technology, where smaller, decentralized solutions disrupted dominant players—like PCs overtaking mainframes.
Where Does AI Go Now?
DeepSeek represents two major pivots in AI. One is the end of massive R&D budgets as a differentiator. When PCs pushed aside “big iron” computers in the 1990s, IBM’s high-cost labs and Nobel scientists no longer mattered. For all its riches, IBM could not keep pace with the outside world. Not when scruffy rebels had access to Intel’s chips and Microsoft’s software. IBM’s Big Blue brand, a sponsor of Super Bowl commercials and Masters golf tournaments, gave way to scrappy Dell.
The second pivot is the rise of user innovation. Quick history: Google did not invent the internet (that was the U.S. Defense Department) or the World Wide Web (CERN in Switzerland). Nor did Google invent commercial web browsers (Netscape, via the University of Illinois). Google, for gosh sakes, didn’t even invent the internet search engine. At least five other search engines preceded Google—all were initially better funded. So what did Google invent? It had one powerful idea that cost very little to try and scale: a user-friendly, page-ranking algorithm. Google was a clever assembler of other people’s concepts. The heavy lifting had already been done. Google managed it on the cheap, which is how the founders and early investors made so much money.
Apple borrowed ideas from Xerox. Samsung from Sony. Nike from Asics. Singapore from London and Zurich. The best ideas never stay contained. With DeepSeek, AI has breached the high walls of large investment.
Who Will Capture AI Value?
As 2024 turned into 2025, investors were certain they knew who would win in the AI era. Nvidia would continue to win. Big tech would win. Huge data centers and energy budgets would win. But now comes DeepSeek, a wrinkle in the AI universe. It challenges us to think again.
Some of history’s most salutary technologies—telephones, automobiles, antibiotics, international jet travel, the personal computer—bent society’s productivity and wealth curves upward. Yet the value was captured diffusely, and overwhelmingly by users. Warren Buffett points out that most airlines have lost money almost every year since the former U.S. carrier Pan Am put the 707 jet into service in 1958. Accenture, which advises corporations how to use large computer systems, has a higher market value than IBM, which sells them.
Who will capture AI’s value? Well, that’s suddenly a new question in 2025. Likely the cleverest, not the richest.
