What does the newest AI model mean for US big tech?
There has been widespread reaction to a new, low-cost AI model called R1 created by DeepSeek, the artificial intelligence company based in Hangzhou, China. In the midst of the US tech sell off that ensued, one of the most apt and measured comments came from President Donald Trump. Reacting to the much less expensive AI coming out of China, he said “that’s good because you don’t have to spend so much money. I view that as a positive, as an asset.”
Now that the dust has settled, it is hard to argue with President Trump’s point of view on this. In fact, we would go further by saying that DeepSeek’s achievement may herald a tentative move into an era of democratisation in AI. Positivity was certainly not reflected in the market reaction, yet we think the sell-off was indicative of some investors’ faulty reasoning.
Up to now, the consensus was that, to develop an AI model worth its salt, you needed three things: a great deal of money; a huge amount of computing power; and a vast array of the fastest computer chips from NVIDIA. All this meant that AI research was only possible for the US companies with the deepest pockets.
So, some have claimed that the status quo has been turned on its head by DeepSeek, who have showed that an inexpensive AI model could be developed by a scrappy outsider and downloaded onto your mobile for free. To put it another way, DeepSeek appeared to have filled in and traversed the ‘moats’ that were assumed to surround the US tech giants.
In a recent interview, the founder of DeepSeek, Liang Wenfeng, said that he hoped other new and emerging companies would have the confidence to take on the leading US tech giants. To support those Davids against the tech Goliaths, Wenfeng has made his AI models ‘open source’ to assist in their development.
Yet questions surround quite how cheaply DeepSeek was actually able to produce this AI model. DeepSeek revealed huge cost savings against their US rivals in the same stage, spending $6m on the final training run of its V3 model. Due to US trade restrictions, DeepSeek can no longer buy the fastest computer chips from NVIDIA, however it acquired a stock of high-powered NVIDIA chips prior to the imposition of trade restrictions by the U.S. government on China. This allowed the firm to build the infrastructure necessary to develop large language models (LLM).
Yet the $6m figure does not take into account significant research, development, and implementation costs associated with producing such a model. Many believe the all-in cost to be much, much higher, with a SemiAnalysis report quoting more than $500m all told. Revenues from Wenfeng’s multi-billion dollar quantitative hedge fund High-Flyer gave DeepSeek the consistent funding it needed to develop its model successfully. This is clearly not an industry with low barriers to entry.
The truth is that despite its advantages, DeepSeek needed NVIDIA chips to get off the ground, and had a solid foundation in AI to start with. Newer companies attempting to make a name for themselves will need chips from somewhere, and the likelihood is these will come from NVIDIA. With the highest quality design and technical know-how, it has shown that even its older chips are able to support the functioning of a leading LLM. NVIDIA still has several years’ lead in development over its rivals in China, and few could claim that peak AI been reached on the back of NVIDIA’s more dated chips. There will still be demand for more sophisticated and efficient chips as AI advances, and those, at the moment, are more likely to be produced by NVIDIA than anyone else. The weight of that intellectual property will not be shifted easily.
Not only this, but the US government, unlike in most other industries, will be happy to accept a monopoly in the chip-designing space so long as they stay ahead of their rivals in the East. The timing of DeepSeek’s announcement, so close to Trump’s inauguration, certainly raises eyebrows. Was this a destabilising effort from the Chinese government to challenge US superiority at a time of vulnerability for the nation? Wenfeng’s business certainly has close ties with the Chinese government; he was invited to Beijing to meet Chinese Premier Li Qiang, and as a ‘National High-tech Enterprise’ his hedge fund was given preferential tax treatment, as well as state subsidies to carry out research and development. These are just the publicised examples of Chinese state involvement in DeepSeek.
Politics and DeepSeek are inseparable – if you ask DeepSeek what happened to the Uyghur Muslims in China it apologises, claiming the question to be beyond its ‘current scope’.
Whatever the significance of DeepSeek’s innovation may be, there is likely to be some level of democratisation in the AI space from here. If something is more efficient and accessible, its demand could shoot up. As Satya Nadella at Microsoft says, “Jevons paradox strikes again,” meaning the demand for all things AI could skyrocket as it becomes cheaper, eclipsing any short-term doubts about stock prices. If demand does shoot up, DeepSeek has shown that more can be achieved with less, bringing AI’s costs down further and bringing forward the exciting productivity gains that it might one day introduce. We believe the US tech incumbents, with their lead in development and their ability to absorb the still significant costs associated with producing this technology, will lead the charge. NVIDIA itself welcomed the “excellent AI advancement”, anticipating the spike in demand to come. Either way, the AI story, of which we have read so much in recent years, is just beginning.
Please be aware that the value of investments may go up or down and you may receive back less than you invested originally. Past performance is not a guide to the future.
This document contains general information only and does not provide any advice or guidance specific to your personal circumstances.
Callanish Capital is authorised and regulated by the Financial Conduct Authority to manage investments but does not provide Financial Planning or Investment Advice. Please refer any such queries to your Financial Advisor.
This email and any accompanying attachment is issued by Callanish Capital Ltd, a company registered in England and Wales under company number 13182424. Registered Office: 45 Pont Street, London, SW1X 0BD.
Callanish Capital Ltd is a Discretionary Fund Manager, directly authorised and regulated by the Financial Conduct Authority (“FCA”) under registration number 955992.