DeepSeek’s New R1 Model Has Stunned AI Researchers, Tech Giants, and a Complacent Market
Just as tech heavyweights had gotten comfortable writing multi-billion dollar checks for data centers and AI chips, China found a way to do AI faster and cheaper — so on brand. Last week, open-source AI research firm DeepSeek caught the world by surprise with its new R1 model. Yesterday, the markets reeled as they processed the disruptive potential.
AI, Race to the Bottom: Despite targeted sanctions on China, DeepSeek achieved a striking level of quality with its new R1 reasoning model, trained with just 2K chips and $6M in compute. That’s a jarring reduction in cost when compared to the 16K chips and $100M+ spent by AI pioneers like OpenAI to train their latest models. Even more stunning, US companies have the benefit of next-gen AI chips — while DeepSeek is operating with last-gen hardware.
- R1 ranked second on Artificial Analysis’ Quality Index, narrowly behind OpenAI’s new o1 reasoning model — and it’s more affordable, too.
- If validated, it could represent a reset for the AI industry, with a16z’s Marc Andreessen calling it “AI’s Sputnik moment” on X.
The AI Shot Heard Around the World
If DeepSeek fundamentally proves anything, it’s that you can do more with less, snapping one of the industry’s largest conceptions in half — bad news for the industry’s first-movers. That revelation shocked investors who had grown too comfortable with ultra-rich valuations, diminished reward from holding stocks, and rising risk. On the news, the Nasdaq Composite fell more than 3.3% — and across the market, tech, semiconductor, and energy stocks led the losses.
- The VanEck Semiconductor ETF fell nearly 10%, dragged lower by Nvidia, Taiwan Semiconductor, and Broadcom — they fell 17%, 14%, and 17%, respectively.
- Energy names that had become beneficiaries of an “electricity boom” from AI — Vistra, GE Vernova, and Constellation Energy, among others — fell 20%+.
Diving deeper: This revelation comes after a whirlwind 2024 for data center spending and upgrades. And in 2025, Big Tech planned to go even bigger — with Microsoft and Meta committing $80B and $65B respectively, on data centers, while OpenAI, Oracle, and SoftBank plotted $500B in spending. But just like that, investors are once again questioning if the “AI trade” can continue business as usual. It might depend on how quickly tech giants can respond or if those companies can still keep up with the lofty expectations imposed upon them.
