“OpenAI is highly overvalued. I think we saw their business model sort of blow up over the last few days with DeepSeek basically giving away for free what they [OpenAI] wanted to charge money for,” Gary Marcus, a professor at New York University (NYU), said in an interview with CNBC on Tuesday, January 28. “We may see OpenAI’s evaluation plummet in the same way we saw with WeWork,” he said, referring to the co-working space startup that went from a peak valuation of $47 billion in 2019 to filing for bankruptcy in 2023. The AI expert’s remarks come as OpenAI and other Silicon Valley giants face a reckoning after a nondescript Chinese AI research lab, DeepSeek, developed large language models (LLMs) with “performance comparable” to US alternatives at a tiny fraction of the cost. The rise of DeepSeek’s AI chatbot through app store rankings triggered a broader sell-off in tech stocks across markets, with AI chipmaker Nvidia witnessing nearly $600 billion wiped from its market value (the single largest decline of a public company in US history). Nvidia’s stock has since rebounded 5 per cent, as per Forbes. DeepSeek’s AI models have been hailed as a research breakthrough as they demonstrate that it is possible to develop competitive, frontier AI models using less cash and fewer GPUs - as opposed to the billions of dollars spent by OpenAI, Meta, Google, Microsoft, and others to do the same. Addressing doubts about DeepSeek R1’s capabilities, Marcus said that he had read the Chinese AI startup’s technical paper and found that it “makes sense”. “If the company just made it all up, they are going to flame out pretty quickly. What they did is plausible, they worked harder for optimisation than other people have before,” he said. “The final problem is that LLMs are not reliable. We all know by now that it hallucinates, makes weird errors, and so forth. That’s a problem,” Marcus said. Talking about the geopolitical implications of DeepSeek’s success, Marcus said, “What this really shows is that AI is anybody’s gain and nobody is going to win by just playing with LLMs. LLMs are too well understood. Everyone is playing the same game so it is impossible to get a big lead. Now, even smaller countries are going to be able to play because of these results.” “If somebody is going to get an upper hand here, they will need to invent something fundamentally new, something that works differently from LLMs,” he continued. Marcus further opined that LLM advancements alone will not help companies achieve artificial general intelligence (AGI), that is, AI systems capable of thinking as well as or better than human beings. “Getting to AGI probably requires five or six more breakthroughs and the company or country that can ramp up those breakthroughs first may win,” he said, adding that the US should focus on sponsoring innovation rather than imposing export controls on AI chips.
