
If you're active online, particularly in tech spaces, you might have heard about a company called DeepSeek. If not, here’s a quick intro: DeepSeek is a Chinese artificial intelligence (AI) company founded in 2023 as a spinout from Zhejiang University.
DeepSeek's models have caught the attention of Western companies because they are claimed to have been developed at a fraction of the typical cost and use significantly less computing power. To give you a quick comparison with a pioneer like ChatGPT, take a look at this table:
Aspect | DeepSeek V3 | OpenAI GPT-4 |
---|---|---|
Training Cost | Approximately $5.57 million | Estimated around $100 million |
Computational Resources | Utilized about 2,000 NVIDIA H800 GPUs | Employed approximately 25,000 NVIDIA A100 GPUs |
Training Duration | Completed in less than two months | Took between 90 to 100 days |
You get the idea. Western markets are freaking out, especially since DeepSeek has risen to the top of the App Store charts in recent days. With so many people trying to access it, there are now confirmed reports of outages and performance problems reminiscent of ChatGPT's early days. Remember when ChatGPT would regularly log people out, glitch out, and users had to report bugs to OpenAI on Discord?
This comes after Biden's last-minute rule (before leaving office) that will place further restrictions on AI chip exports to countries like China, a rule that Nvidia fiercely opposed, condemning Biden and praising Trump.
Speaking of Nvidia, Bloomberg reports that the company's stock market value has taken a historic plunge, with over $400 billion erased from its market cap. According to Bloomberg, this 13% drop is the biggest in US stock market history, surpassing Nvidia's own record from September, when the company lost almost $280 billion in value.
Investors are concerned that the rise of competitors like DeepSeek could change the way modern LLMs are trained and used. So, you might be wondering how OpenAI and its employees are responding to all this. Sam Altman, in an X post last month, said:
it is (relatively) easy to copy something that you know works.
— Sam Altman (@sama) December 27, 2024
it is extremely hard to do something new, risky, and difficult when you don't know if it will work.
individual researchers rightly get a lot of glory for that when they do it! it's the coolest thing in the world.
While DeepSeek wasn't mentioned explicitly, the comment is generally interpreted as being aimed at models like DeepSeek, which are replicating ChatGPT at a cheaper cost.
Altman tweeted this a day after DeepSeek V3 launched (December 27, 2024), as the initial hype took off on X.
Update, 28 January 2025, 3:00 pm PST: DeepSeek panic has now erased $2 trillion worth of market cap in US stocks. Some of the biggest losers include Nvidia, Alphabet (Google's parent company), Broadcom, Oracle, ASML (Dutch chip equipment maker), and Microsoft. Nvidia and OpenAI CEO Sam Altman have since responded.
Image via Depositphotos.com
50 Comments - Add comment