As the DeepSeek dust settles – what have we learned about investing in AI?

Investors Paddy Flood and Ankur Dubey give their views on the long-term impact of DeepSeek for investors.

The Chinese AI startup, Deep Seek made headlines in late January when it claimed new advances in the field of AI, saying it has trained a generative AI chatbot at a fraction of the cost required by the market leaders. Off the back of this announcement, stock markets suffered a significant sell-off with the Nasdaq index falling 3% and shares of some AI-related companies plummeting even further.

Portfolio Manager and Global Tech Sector Specialist/Disruption Investor at global investment manager Schroders, Paddy Flood, explains that DeepSeek had essentially invented a new way of making the large language models that underpin things like ChatGPT. But according to Flood, it is the way they did that that is so interesting. “They did it in a very time and cost-efficient way versus the incumbent methodologies. This led to questions about what this means for spending on AI infrastructure, which has been a big driver of the markets over the last couple of years,” says Flood.

Competition for Silicon Valley, cost efficiencies, and the potential for a new world order in tech and AI, roiled markets. But, a month on, and with the dust beginning to settle, what has changed and what can investors learn from the experience about investing in AI?

DeepSeek innovation will accelerate cost reductions

“Generally, in the industry, costs to deliver AI have been coming down through time. Deep Seek's innovations, which everyone can now access will accelerate that cost-down curve. It's likely to mean that the cost of providing these services, people using Open AI, for example, will continue to fall,” says Flood.

To put this into perspective, DeepSeek claims its engineers needed around $6 million in raw computing power to build their app. Roughly one-tenth of what Meta spent in building its latest AI technology.

And while falling costs may lead you to think that demand will plummet too, Ankur Dubey, investment director in private markets at Schroders, explains that, in fact the opposite may occur. “We’re likely to see the Jevons paradox manifest. What the theory says is that when there are efficiency gains for any resource, the natural instinct is to think that that resource will not be as much in demand anymore. But what ends up happening is that because of the marginal cost of using that resource decreases, more and more people start to use it.”

If Deep Seek's claims are true, Dubey and Flood believe it could open the door to smaller competitors to participate in a market that's to date been dominated by the Mag 7 in the US.

“Computing power has been a big cost hurdle for small companies looking to build AI applications. Now, as efficiencies improve, as demonstrated by Deep Seek, and much of that is out of necessity because of the restrictions imposed on China from US. As those efficiency gains happen, more and more AI application producers will start building on AI advances resulting in a race to make better and cheaper large language models. As the number of applications goes up, effectively democratising AI, the price per unit of compute will go down, and the overall volume is likely to go up. Therefore, for the AI infrastructure space overall, the addressable market is expected to increase,” says Dubey.

Magnificent Seven may no longer be the default proxy for retail exposure to AI

Today, most investors, certainly retail investors, will have looked to the so-called Magnificent Seven (Mag7) in the US to gain exposure to the theme of AI. “The Mag7 have been a public market proxy for exposure to AI because they are the ones with the means to do all sorts of innovations, right from compute to applications,” says Dubey.

So big had the Mag7 become, that by the end of 2024, they accounted for more than a third of the S&P 500 and nearly a quarter of the MSCI world. But a focus on Mag7 would be to ignore the many other companies that operate in the AI space – in computing, foundation, infrastructure, or applications. And most of that investment is happening away from the public space and in private markets.

“In the private market space, there's a lot more opportunity to invest because a lot of innovation is happening at every layer,” says Dubey.

Where has all the capex gone?

Critically, Dubey believes that the DeepSeek innovation brings into question the high capital intensive nature of the large language model itself. “There have been questions raised over the last 18 months to two years around the sustainability of the capex-intensive model of large language models, and where it’s going.” 

Where AI is going remains a big question that needs answering, especially given the amount of money that is being invested into the space. The longer that question goes unanswered, the more the risk becomes attached to those investments.

Has DeepSeek changed the risk profile of investing in AI?

Investing in AI as a theme still has risks. “While DeepSeek has improved AI efficiencies and will drive adoption, the investment risk still remains,” says Flood. Critically, he believes these centre around the big unknown with AI – adoption. “The risk is whether this technology will be used by businesses and people all over the world to do things that help them improve their productivity or whatever it may be. That was a question before DeepSeek, and that is still the question after DeepSeek, and one that I think will determine the success or otherwise of investing in AI.”

Flood believes that investors will need to see a payoff in terms of people using this technology relatively soon. “There has been a lot of investment going into the AI infrastructure, and the companies who are trying to monetise it haven't yet done so in a significant way. We will need to see more use cases coming out over the next 12 months to justify the spend.”

Given the risks, how should investors approach investing in AI?

In general, Dubey explains that the market still hasn't seen a host of products that have been adopted en masse by enterprise or consumer use cases. As a result, he says that faith in the potential returns on each dollar spent on AI development is very low. “What this means is that any shock news that the market sees will result in an extreme market reaction – which we saw with the DeepSeek announcement.”

Because of this, Dubey emphasises the importance of keeping the long term, bigger picture in mind so that short- term volatilities are not of meaningful impact. “But that being said, disruption is always meaningful from an innovation perspective.”

So don't fear disruption. It can be a good thing. But do be aware of the short-term market ructions disruptors can cause.

And stick to the fundamentals. Flood concludes that it's always good to take a step back and think of the businesses you believe have very high defendable barriers to entry in this space – those that are really driving value to it and have long growth trajectories that aren't priced in by markets. “It's a careful balance between assessing the fundamentals of the business versus the valuation. You want to make sure that you're not paying exuberant prices for a company's destiny that might not be within its control or for a destiny that is going to be increasingly competitive. That's how we tend to think about it.”