Equities
5 min read 30 Jan 25
Although stocks have subsequently recovered, DeepSeek’s emergence seemingly out of nowhere has led investors to question the perceived dominance of US tech stocks in the AI race, as well as their future growth and valuations.
Competition is at the heart of capitalism; dominance is rarely guaranteed as new companies are constantly emerging to displace the market leaders. As January 2025 drew to a close, an unknown Chinese company abruptly disrupted the prevailing narrative about the future of AI and the investment landscape.
The recent panic among investors in AI stocks stems from the release of a new large language model (LLM) by a Chinese start-up, DeepSeek. The model is similar to ChatGPT from OpenAI in that it generates answers in response to prompts.
DeepSeek’s model appears to perform as well as those from OpenAI and other US tech giants. The critical difference is that the Chinese firm said it cost considerably less to train and run than its US counterparts. The development of a comparable model that is apparently more efficient and available at a fraction of the cost triggered sharp falls in AI-related stocks and challenges assumptions about AI investments.
The start-up’s cutting edge product took many by surprise as it was widely assumed that US tech firms were ahead of their Chinese counterparts in the AI race. Chinese firms have been denied access to the most advanced chips from Nvidia, yet DeepSeek appears to have created a chatbot that can deliver comparable results without them.
It is reported that DeepSeek’s model cost around $5 million dollars to develop, which is significantly lower than the billions of dollars invested in existing US models and proposed capital expenditure plans from the likes of Meta and Microsoft. This figure may turn out to be too low, however, as they may have used some prior work before building their model. While there is a lack of clarity around the exact cost, in terms of performance DeepSeek appears to have been validated.
This shouldn’t come as such as shock. China has many talented computer scientists and the fact they been able to develop this product with less computational power demonstrates their capabilities. The attention has been on US firms, but the emergence of DeepSeek could be seen as validating China as a major force in AI globally.
The company that has arguably been most impacted by this development is Nvidia. The company’s chips are considered to be an essential component of the AI infrastructure. In the past couple of years, the development of new LLMs has fuelled demand for Nvidia’s chips and propelled the company’s revenues – in November the company reported record quarterly revenue of $35.1 billion, an increase of 94% on the same period a year earlier1.
In the age of AI, so far, Nvidia has been the dominant player. But DeepSeek’s emergence could be seen as an example of Economics 101: a company with super normal profits attracts competition and its moat is shown to be less secure than perceived. The current narrative around Nvidia is that if it is possible to create LLMs that are less compute intensive, there may be less demand for its chips than anticipated.
The market also focused on some of the mega-cap firms that have been spending billions on Nvidia’s chips to develop their models and applications. While DeepSeek’s low cost model has raised questions about their capital expenditure, the leaders of these companies are arguably some of the most brilliant computer scientists around. They have many smart engineers and, in our view, they're also very capital disciplined. If they need to keep building up data centres, we think they'll continue to invest. And if they don’t need to spend as much, free cashflow will likely increase. But they will be focused on getting a return on that capital and being able to monetise those investments.
When we think about the AI opportunity, we don't think it stops with generative AI. It essentially started with image recognition and predictive analytics. Generative AI is a big use case and that's maturing now. But there are other AI opportunities beyond this such as agency AI, where computers start to be able to reason like humans do. There’s also robotics – industrial robots and maybe even personal robots in the future – as well as self-driving cars. We believe the AI opportunity has the potential to be a multi-decade opportunity and could continue for some time.
One potential implication of DeepSeek’s model is that it is a catalyst for the development of AI. More companies may seek to deploy AI technology if the cost comes down. DeepSeek could actually mean the addressable market actually gets bigger for AI.
In the history of computing, whenever computers become more powerful, their new applications for computing means the addressable market increases. So in the last few years, that computational power has been driven by semiconductor manufacturers, notably Nvidia. But when the software becomes more efficient, compute becomes more powerful as well.
So the implication here is that for companies that want to incorporate AI into their products, particularly enterprise software companies, it's less capital intensive and it's moving faster. They can add these enhancements to software more quickly so the end market actually grows as well.
For these companies having more large language models available could actually be a good thing, because there's a competitive element and they can deploy one or many to their customers.
Another potentially positive aspect of DeepSeek’s model is that it could validate the concept of a small language model, which would enable enterprise software companies to deliver bespoke language models to their customers. Different companies typically have different terminology, different key words and different workflow, so this development could enable companies to have bespoke models that are adapted to their specific requirements.
At M&G Investments, we think about the AI investment opportunity in three broad categories. The first category is enablers: companies that provide foundational technology for AI such as semiconductor firms.
The second category is providers: companies, particularly in the enterprise software area, that will take the AI technology and productise it. AI can make their products easier to use, more user-friendly and they can charge extra for it.
And the last category is beneficiaries. These are companies that can use AI internally to grow their business or just become much more operationally efficient.
The investment opportunity is likely to evolve, starting with the enablers and the development of the infrastructure and LLMs. Then as the product matures, providers will be able to deploy them into products and finally the opportunity will emerge in the beneficiary category as companies deploy AI themselves.
From our own investment experience, we are seeing more signs that the technology is making progress and companies are starting to use existing models.
Although Nvidia has been the poster-child for the new era, there are lot of other great companies in the overall tech economy. We have touched on the software sector but there’s also semiconductor companies that are tied to the AI theme but not necessarily on the training side.
We think there’s a great deal of opportunity in this area and over the longer term, tech will remain a growth industry. It has been for the last 50 years, driven simply by the fact that as computing power increases, the addressable market for compute increases as well.
Some commentators have drawn comparisons between the huge amount of investment in AI and the investment in the internet and the dot-com bust in 2000. In fact, many great companies emerged from that time when the underlying infrastructure for the internet was built, including the likes of Netflix, Facebook (now called Meta Platforms), Google and the whole ecommerce industry.
It is possible that we're seeing the same thing with AI. We will have infrastructure in the form of the AI models, and then companies that can use this technology will be able to grow and we may see the creation of new companies too.
The emergence of DeepSeek has disrupted the prevailing narrative and prompted some thinking about the future of AI. When we look back we think it will be one of the important data points in the evolution of the technology. Ultimately we think it's a sign of acceleration of the development for AI. We can potentially get to new AI applications sooner rather than later. Opportunities such as agency AI and robotic AI could be nearer because the pace of innovation is actually increasing.
The value of investments will fluctuate, which will cause prices to fall as well as rise and investors may not get back the original amount they invested. Past performance is not a guide to future performance. The views expressed in this document should not be taken as a recommendation, advice or forecast. The information provided should not be considered a recommendation to purchase or sell any particular security.