Firstly, below is a great chart of this week’s craziness. Deepseek is regarded as negative for most AI stocks due to disruption fears (lower cost models).
Investors rushed to sell pick ‘n’ shovel companies (which provide the services/products needed to create a product), while AI enablers outperformed
DeepSeek is a Chinese artificial intelligence company founded in 2023, headquartered in Hangzhou, China. It develops open-source AI models, including large language models and visual language models, designed to be efficient and cost-effective. The company aims to handle multiple tasks simultaneously without slowing down, providing low-cost AI solutions. DeepSeek’s model apparently performed at similar levels to OpenAI’s o1 reasoning model on some benchmarks (math, coding) but was trained at only a fraction of the cost, and its mobile app surged to the top of Apple’s download charts in the US. Stocks with exposure to Artificial Intelligence sold off at the start of this week as DeepSeek prices their models anywhere from 20x to 40x cheaper than equivalent models from OpenAI.
Many questions about DeepSeek.
These include, among others, that the company is accused of theft by OpenAI, accused of breaking US export restrictions when they bought Nvidia chips (probably via Singapore), and most experts doubt the low costs Deepfake cited. In addition to that, Deepfake is censored by the Chinese government, has insufficient privacy policies, and all data is stored in China. In some recent tests, Deepfake results were incorrect in 83% of all cases, and to make things worse, it got hacked and more than 1m data points of users have been exposed in the internet.
Jevons Paradox
In the discussion about Deepseek, an important theory was cited very often: Jevons Paradox.
William Stanley Jevons observed this phenomenon in the 19th century with the use of coal in England. As steam engine technology improved, engines became more efficient at using coal. One might expect that this would lead to a decrease in coal consumption. However, the opposite happened. The increased efficiency made coal a more cost-effective energy source, which led to its wider use in various industries. As a result, the overall consumption of coal increased significantly. This example is what led Jevons to formulate his paradox, highlighting how improvements in efficiency can sometimes lead to increased resource use.
So, even if we assume that Deepseek is NOT Deepfake, it may be a net positive for the whole AI space, as lower costs will accelerate AI adoption everywhere. Currently, AI is a very small part of the overall semiconductor market, only a limited number of very expensive, high performance chips are being bought by a few hyperscalers like Microsoft. As the cost for these chips is high, it is economically not viable to put them into devices like smartphones or PCs. If costs fall dramatically, there will be billions of ADDITIONAL chips for NEW applications in smartphones, PCs, cars, robotics etc. Basically everywhere. More chips mean more chip equipment will be needed, which is positive for the likes of ASML. And this will then mean more data, which will require more datacenters, and more usage of power.
And last but not least, lower costs for AI, and on the back of that a faster adoption of AI, is very disinflationary! BNP Exane for example expects AI to increase global productivity by 1% per year. And this will benefit most companies, globally. And whatever you may think of famous tech investor Cathy Wood, she made an interesting comment at a conference in Germany yesterday. She thinks that thanks to AI, global growth could grow from 3% to 7% thanks to productivity increases!
Ralf Schmidgall, Fund manager MW Actions Europe