DeepSeek Shakes Markets in 2025
On January 27, 2025, a relatively unknown Chinese-based artificial intelligence company shook up Wall Street. At the time, the AI buildout was in the early innings, and the majority of leading U.S-based technology stocks were at or near all-time highs. However, DeepSeek changed the narrative overnight when it launched its R1 AI model. The company claimed that, unlike leading AI companies like OpenAI and Alphabet (GOOGL), which require hundreds of billions of dollars in hardware, power, and data centers, it was able to build a rival model for a mere $5.6 million. Meanwhile, according to DeepSeek, it was able to bypass U.S. government controls on NVIDIA’s (NVDA) top chips and get similar results using fewer and older GPUs (A100s & H800s).
If DeepSeek’s claims were true, it would mean that hyperscalers like Meta Platforms (META) and Amazon (AMZN) were undercut after spending hundreds of billions of dollars on expensive AI infrastructure. In late January 2025, investors didn’t wait for an answer and instead sold first, and asked questions later. Wall Street would lose roughly $1 trillion in a single day while chip leader NVDA would plunge 17% on massive volume turnover.
Image Source: Zacks Investment Research
DeepSeek Fallout: Cooler Heads Prevailed
After the DeepSeek carnage, AI stocks slumped for a few months before fully recovering their losses (and then some). Below were the reasons for the recovery:
· “Jevons Paradox”: The main reason for the vicious recovery was that investors realized that efficiency does not equal demand. In fact, Jevons Paradox suggests exactly the opposite. That is, when costs come down, consumption explodes.
· CapEX Spending Stayed Steady: Despite the fear reverberating through Wall Street, big tech CEOs continued to commit billions to AI infrastructure, suggesting that they did not view the DeepSeek breakthrough as a reason to stop committing capital. Later, corporate earnings would suggest that the fundamentals of AI companies continued to grow despite the DeepSeek threat.
· Debunking Spending Claims: Following the knee-jerk reaction to DeepSeek, tech experts investigated and found that the $5.6 million in spending suggested by the company was misleading at best. The $5.6 million was instead an iteration on top of pre-existing hardware and multiple years of research and development.
Kimi K3: DeepSeek Déjà vu?
Currently, the Nasdaq 100 Index (QQQ) is having its worst July in over two decades. The reason for the sell-off is that a Beijing-based start-up named Moonshot AI released a flagship named Kimi K3. The open-source model is one of the most efficient AI models ever released, and its performance is said to be near or on par with top U.S AI models. Once again, investors are selling first and asking questions later amid CAPEX ROI anxiety and hardware demand concerns.
