Clement Inbona

Deepshift

USD 589 billion for starters – that’s the amount that was knocked off the valuation of Nvidia – the artificial intelligence (AI) stock market darling – on 27 January.And there was no significant recovery in the days that followed. So why the drop?

The Chinese AI company Deepseek, whose chatbox was known only to a close circle of insiders up until now, has shot to the top of most mobile app and media rankings. There are a number of very specific reasons for the buzz around this app: it is Chinese and open source – its source code is transparent and can be changed – but first and foremost, its performance is comparable to that of the AI stars developed by OpenAI, Meta, Google and Microsoft, whilst its cost of development and use is, in principle, far lower.

The quality of a large language model (LLM) depends on three major factors: the volume of available data, the processing capacity and the quality of the algorithm. Despite US restrictions on semi-conductor exports, Deepseek can successfully compete with the best LLM thanks to innovations in the algorithm. Even if many questions remain to be answered, particularly regarding the range of data used and respect for copyright, this event offers a number of macroeconomic and financial lessons.

Size is not everything. To date, the US giants have competed relentlessly on investment in processing power, including USD 500 billion on the enormous US Stargate Project. But China reminds us here that one of the intrinsic characteristics of its economic success since joining the WTO is its capacity for low-cost production in increasingly complex markets. This was initially the case in textiles, then in the industrial sector, through to electric cars. And today in AI?

The emergence of Deepseek will undoubtedly mark a clear break, as the structure of the AI market could previously be considered as inflationary. With rare and expensive raw materials, state-of-the-art electronic chips represent a monopoly for Nvidia, and AI models were the preserve of a few tech giants, i.e. an oligopoly. Today, the barriers to entry for AI models seem to be falling, making competition fiercer. Will this drive prices down? This would be a good thing for the democratisation of this technology – the volume effect – but undoubtedly less positive for the profitability of the players who have already invested billions.

Deepshift could thus represent a disinflationary shift, at a time when inflation remains one of the preoccupations of central bankers. It could also accelerate productivity gains, at a time when potential growth is falling in many regions of the world. Lastly, for companies, it should be a vector for value-added sharing on a broader basis, particularly for those that adopt this technology or offer products based on it.

Final version of 31 January 2025, Clément Inbona, Fund Manager, La Financière de l’Échiquier (LFDE)

The opinions cited are those of the fund manager. LFDE shall not be held liable for these opinions.