So it won’t be much of a surprise to any market participant that there’s been some pretty big moves in the past week. So we saw the biggest market cap loss ever in the history of the stock market, with Nvidia losing US$560 billion worth of market cap (value) in a single day.

The reason for this was because of the emergence of a Chinese AI company that has built a large-language model, cheaper, that can compete with the big existing ones from OpenAI and Meta.

Now, it is clearly far too early to know exactly what the entire ramifications of DeepSeek’s model will be on the AI space. However, there are three things that we are thinking about.

The first one is DeepSeek’s model is open source, so therefore it means that it can be leveraged by others. So we probably will see a proliferation of other models such as DeepSeek come along.

Secondly, it uses a process called ‘distillation’, so it is essentially leveraging bigger models such as OpenAI’s ChatGPT or Meta’s Llama to train itself. So we think the future here is much more of an open-source model for AI generally. The [first] big conclusion is that DeepSeek have claimed that it only cost them US$5.6 million to train this model.

Now given it looks like it was incredibly efficient in terms of training and inference (that’s when you ask it a question), could it be that DeepSeek is going to democratise AI? Ie. given its efficiency, given the cost – will we see a proliferation of different use cases across the space. In a world of technology, any cost improvement, usually increases the number of use cases. So we are quite excited about what it means for the industry as a whole.

The third important point is actually that infrastructure will still be a defining differentiator for the AI players. So just this week, Meta had its results, where it confirms that it still intends to spend between US$60-65 billion on capital expenditure, mainly to build out its data centre infrastructure and compute power. Because they believe, still, that compute [power] is going to be a differentiator going forward.

So where does that leave us here, in European equities? Well, the way in which we have positioned our funds is agnostic as to which large-language model will win, or if there is a proliferation of them. Because the European semiconductor capital equipment space and electrical cap goods space plays into all of these. Because it’s the enabler. It’s the picks and shovels that helps the AI revolution to happen.

The second broad conclusion that we can take from this week is the dangers of concentration. So you will have seen the big market cap falls that happened in the US across the big tech space. And clearly when the S&P500 has 32 per cent sector exposure to tech, there is that danger. In Europe, the equivalent figure is only 8 per cent and the valuation starting point is much lower.

Glossary:

Agnostic: Not advocating for one investment style (or stock) over another.

Capital expenditure: Money used by a company to purchase, upgrade, maintain or develop its assets, be that equipment, technology, or infrastructure, including buildings.

Large-language model (LLM): A machine learning model designed to understand and generate responses to human language via the processing of large amounts of text data.

Open source: Software for which the original source code is freely available to be shared, distributed or modified.