AI leader NVIDIA stock took a massive tumble yesterday, dropping 17% in trading that resulted in a loss of market cap of close to 600$, the biggest ever drop for a US company in a single day. The drop in the chipmaker’s stock price was NVIDIA’s worst day since March 16, 2020, when their stocks tanked during the early days of the pandemic.
The biggest culprit for the drop? Chinese AI company DeepSeek who unveiled its new language model recently. DeepSeek’s new language model seemingly trades blows with competing language models from OpenAI and Google at a fraction of the cost. The Chinese company claims that its language model needed only 2,000 specialized chips (with reduced capabilities to adhere to US rules) to train, far less than the 16,000 or more required to train leading models from its competitors.
The fact that Chinese engineers made do with far, far less has sparked doubts in AI company’s hardware-heavy strategies which utilizes billions of dollars worth of NVIDIA’s high-end, AI-focused hardware.
The United States government has put limits on the hardware that Chinese companies can obtain and use for AI development, but it seems that the restrictions they’ve set in place may just turn out to become hidden boons for Chinese researchers. Faced with doing more with less, DeepSeek’s engineers were forced to be creative in creating their language model, which resulted in tech that only needed about $6 million in raw computing power to build their system. In contrast, companies like tech giant Meta is spending about 10 times more.
If DeepSeek’s claims on the costs it incurred to build their language model is true, then it’ll be good news for smaller companies and bad news for NVIDIA.
“The number of companies who have $6 million to spend is vastly greater than the number of companies who have $100 million or $1 billion to spend,” said Chris V. Nicholson to the New York Times. Nicholson is an investor with the venture capital firm Page One Ventures, which focuses on A.I. technologies.
