On 20 January, Chinese AI company Deepseek released its latest large-language model (LLM). Known as Deepseek-R1, the program has become the most downloaded app on the iPhone app store.
Deepseek-R1 is open-source. It was also reportedly trained with less of the most advanced Nvidia A100 chips and at a cost of just USD $5.6 million.
The founder of Netscape, Marc Andreessen, praised Deepseek on Twitter.
Deepseek the company is actually an off-shoot of High-Flyer, a Chinese hedge fund. It initially aimed for AI-like trading capabilities, and its success funding its AGI (i.e. general intelligence) research efforts. So its purportedly low development cost as following an unknown preceding spend by High-Flyer.
But in terms of its effect on the narrative, the supposedly low-investment release marked a paradigm shift after a week in which Donald Trump, Open AI, Oracle and Softbank were touting a plan to invest USD $500 billion in AI infrastructure. And this in combination with its unexpectedness and the fact of it being open-source sent tech stocks plunging.
Nvidia, which manufactures the powerful chips used in training LLM models, was particularly hard hit, falling 15% by 12pm Monday in New York. ASML fell 6.5% in Monday’s trading in Amsterdam.
Deepseek may also have geopolitical consequences. The Biden administration was attempting to keep high-performance computer chips from China, partly motivated by seeing AI as a new geopolitical contest akin to the Space Race.
But one player that won’t be displeased is Meta. Since July last year, Mark Zuckerberg has been advocating for AI developments to be open-source. Zuckerberg says Facebook was constrained by the iPhone ecosystem and he doesn’t want a repeat of the scenario with closed-source, proprietary AI systems.
“One of my formative experiences has been building our services constrained by what Apple will let us build on their platforms,” Zuckerberg said in a statement, following release of a new version of the open-source Llama AI. “Between the way they tax developers, the arbitrary rules they apply, and all the product innovations they block from shipping, it’s clear that Meta and many other companies would be freed up to build much better services.”
But it’s also possible that US tech will respond with something bigger and better. The innovations Deepseek used to make its LLM at a fraction of the cost could potentially be applied in tandem the greater resources available in the US.
Sign Up To Our Free Newsletter