Is DeepSeek redefining the AI landscape?
Why DeepSeek’s achievements might represent the green shoots of an exciting new phase in AI’s adoption.
In January, DeepSeek, a Chinese artificial intelligence (AI) startup founded in 2023, released the latest version of its AI chatbot, powered by its distilled model R1. The company claimed that not only could its AI assistant do what OpenAI’s ChatGPT could do - reasoning in a similar way to how human beings work through problems - but off a base model, v3, that was trained at a cost of $5.6m; a fraction of the more than $100m price tag of GPT-41 .
Investors initially panicked. The supposedly vast infrastructure investment needed to get the technology off the ground was suddenly being called into question. They pondered if recent AI-related gains had created a US stock market bubble that was now in transit to popping. Markets fell sharply, wiping a record-breaking $600 billion off Nvidia’s market value in a single day2.
Yet, the rout proved to be short lived as investors began rationalising that DeepSeek’s progress could in fact be a boon for the technology via an economic quirk known as the ‘Jevons Paradox’.
Here, we speak to Alliance Witan stock picker Jennison to assess the potential ramifications of DeepSeek’s cheaper models, and ask where it may spur investment opportunities next.
Attention: GPT in progress
In a recent survey by Capital Economics, 80% of respondents agreed that AI would have a “transformative effect on the global economy”3 . How transformative is perhaps open to debate. But we do appear to be in the midst of adopting a new general purpose technology or GPT (not to be confused with AI GPT) – broad innovations that can profoundly impact economies and the efficiency with which they operate.
Example of past GPTs include the steam engine in industrial Britain in the 18 and 19th centuries, electricity in the US in the early 20th century, and more recently in the late 20th century, the ICT digital revolution.
Alliance Witan stock picker Jennison describes how the adoption of and investment opportunities associated with GPTs tend to come in three phases. The first is the “picks and shovels” phase, where adoption is limited, and investment remains concentrated on providers of the core “backbone” infrastructure of the technology.
Today’s AI landscape appears to be in this phase. Investment has been mostly focused on the hyperscalers - providers of enormous data centres used in training AI’s large language models, including Microsoft Azure, Amazon AWS, and Google Cloud – as well as the chip providers such as Nvidia.
From phase one, as the technology improves and costs fall, we move into phase two. At this stage, adoption becomes widespread, and investors begin shifting their focus further downstream, from the makers of the technology to the beneficiaries.
Within the AI theme, Jennison believes phase two will involve investing beyond the cloud and chip providers to sectors such as software, enterprise AI, automation, healthcare and consumer technologies, where AI can drive efficiencies and create new sources of revenues.
Below, Jennison depicts its view of changing investor focus as we move up the adoption curve – the ‘S’ curve.
Source: Jennison, as of 17/03/25
All told, DeepSeek’s progress has raised an interesting question for Jennison: might we be moving into phase two already?
Too early to tell?
As DeepSeek’s chatbot worked its way up the popular app charts in January, so too did the term ‘Jevons Paradox’ in Google Trends.
This economics concept was first described by English economist William Stanley Jevons during the Industrial Revolution in 1865. It argues for highly elastic pricing, with demand that dramatically increases as input costs reduce. Jevons could see that the production of more efficient steam engines led an increase in coal consumption, rather than a decrease as many economists at the time thought it would. It means as input costs drop, the total market grows, allowing for a host of new applications and much wider adoption of the technology.
DeepSeek claims that it has delivered AI at dramatically lower costs than other models, even with sanctions and export controls that have hobbled its access to the latest chips. But sceptics caution that its claims are not fully validated yet, with OpenAI accusing DeepSeek of copying its models.
They also point to the historically lengthy and messy lag in the adoption of GPTs. Yet, if we comb through GPTs of the past few hundred years, we see that that the adoption lag has actually been reducing over time.
Source: Comin and Mestieri, 2018
This raises the prospect that across an increasingly efficient and interconnected global economy, the adoption of AI may prove to be much faster than any GPT gone before it. But though feasible, it still isn’t clear at this stage, and Jennison notes that they will remain cautious about shifting investing exposure downstream too soon:
“The key to success is recognizing when a theme has played out—in this case when the early infrastructure winners have matured—and shifting focus to the next stage of the cycle, where the new technology’s real-world impact and monetization accelerate. We could be closer to this point now after DeepSeek’s emergence, but being too early is only slightly less bad than being too late to transition a portfolio.”
With robust tech sector valuations undoubtedly making investors a little edgy, it is perhaps unsurprising that markets reacted so strongly to DeepSeek’s sudden and seeming challenge to the status quo. Nonetheless, as AI innovation booms and costs fall, the next phase of new investment opportunities may be just around the corner.
This information is for informational purposes only and should not be considered investment advice. Past performance is not a reliable indicator of future returns. The views expressed are the opinion of Towers Watson Investment Management (TWIM), the authorised Alternative Investment Fund Manager of Alliance Witan PLC, and are not intended as a forecast, a guarantee of future results, investment recommendations or an offer to buy or sell any securities. The views expressed were current as at January 2025 and are subject to change. Past performance is not indicative of future results. A company’s fundamentals or earnings growth is no guarantee that its share price will increase. You should not assume that any investment is or will be profitable. Information contained herein has been obtained from sources believed to be reliable, but not guaranteed.
TWIM is authorised and regulated by the Financial Conduct Authority. Alliance Witan PLC is listed on the London Stock Exchange and is registered in Scotland No SC1731. Registered office: River Court, 5 West Victoria Dock Road, Dundee DD1 3JT. Alliance Witan PLC is not authorised and regulated by the Financial Conduct Authority and gives no financial or investment advice.