As the Chinese AI assistant DeepSeek began to go viral this weekend amid reports that its advanced reasoning large language model was rivaling the performance of ChatGPT 4o, Claude 3.5 and Llama 3.1 at a fraction of the cost, a frenzy began to build on the social media platform X as to what this means for the acceleration of AI development.
To date, the narrative driving the trillion dollar valuations of tech giants is that greater processing power is going to be needed for the coming era of AGI, as well as massive investment to fund it.
In October, ChatGPT-maker OpenAI announced the largest venture round in history, having raised $6.6 billion at a $157 billion valuation to develop AI agents like Operator to autonomously transact on behalf of users. Last week the company announced a $500 billion infrastructure venture called Stargate, backed by Softbank, Oracle, Nvidia and Microsoft, to build a half million square feet of data centers that will employ 100,000 people to power it.
DeepSeek, which was founded just two years ago by Chinese hedge fund investor Liang Wenfeng, said it trained its latest large language model to perform on par with leading AI models for a mere $5.6 million. By contrast, OpenAI’s training costs were rumored to be $3 billion last year, with $4 billion going to processing power, as reported by The Information.
Investors excited by the prospect of a less expensive, more sustainable alternative, sent Nvidia stock tanking 17% this morning as the DeepSeek app soared to the top of the charts, racking up more than 1.6 million downloads on the Apple App Store for the US, UK, Australia, Canada, Singapore and China, according to Bloomberg.
AI leaders split
Among those quick to congratulate DeepSeek on getting to #1, was Aravind Srinivas, CEO of Perplexity, an AI search assistant built on OpenAI and Anthropic tools. He posted his full support on X, “For awhile, it wasn’t clear who would beat ChatGPT for the first time. The best we could manage was #8, a year ago. Look forward to using all their models for search, assistant, and agents this year.”
Box CEO Aaron Levie said this was an important wake up call. “The reason you don’t want overly aggressive AI regulation right now is the DeepSeek problem. One day there will be a breakthrough somewhere else, and having your arms tied behind your back trying to accelerate past it will destroy your chances of doing so,” he posted.
But just how scrappy DeepSeek is, is being hotly debated, particularly regarding the claim that its latest model only had 2,000 older generation Nvidia H800 chips to work with. Scale AI CEO Alexandr Wang told CNBC that Chinese labs have more high powered chips than people think.
“My understanding is that DeepSeek has about 50,000 H100s which they can’t talk about obviously because it is against the export controls that the US has put in place. I think it is true, I think that they have more chips than other people expect, but also on a going forward basis they are going to be limited by the chip controls and export controls that we have in place,” Wang said.
OpenAI investor Thrive Capital CEO Josh Kushner stated that he believed DeepSeek wasn’t being fully transparent with the true cost of development. In a post on X, he described DeepSeek as a “Chinese model trained off of leading frontier models, with chips that likely violate export controls, and—according to their own terms of service—take US customer data back to China.”
AI security remains an issue
Even ChatGPT and Google Gemini were a bit skittish about their ascending rival when I asked them whether it was safe to download DeepSeek, particularly considering concerns that led to the US TikTok ban. Both AI assistants advised to proceed with caution.
“Given DeepSeek’s origins in China and the current geopolitical climate, there are legitimate concerns about potential security and privacy risks associated with downloading and using their AI model on a personal device,” said Gemini. “Hidden malware or backdoors that could be used for surveillance or other malicious activities cannot be entirely ruled out.”
ChatGPT responded similarly, explaining that DeepSeek’s track record with security vulnerabilities and data transparency made it a less safe choice, citing a security flaw found in December that could have allowed for the hijacking of accounts through prompt injection before being patched.
By midday today, security fears came front and center when DeepSeek announced its AI assistant was under a “large-scale, malicious attack.” The notice was still on its website, chat.deepseek.com, by end of the day.
