Revolutionizing AI: The Impact of DeepSeek's Advances

Our Chief Fixed Income Strategist Vishy Tirupattur thinks that efficiency gains from Chinese AI startup DeepSeek may drive incremental demand for AI.

----- Transcript -----

Welcome to Thoughts on the Market. I’m Vishy Tirupattur, Morgan Stanley’s Chief Fixed Income Strategist. Today I’ll be talking about the macro implications of the DeepSeek development.

It's Friday February 7th at 9 am, and I’m on the road in Riyadh, Saudi Arabia.

Recently we learned that DeepSeek, a Chinese AI startup, has developed two open-source large language models – LLMs – that can perform at levels comparable to models from American counterparts at a substantially lower cost. This news set off shockwaves in the equity markets that wiped out nearly a trillion dollars in the market cap of listed US technology companies on January 27. While the market has recouped some of these losses, their magnitude raises questions for investors about AI. My equity research colleagues have addressed a range of stock-specific issues in their work. Today we step back and consider the broader implications for the economy in terms of productivity growth and investment spending on AI infrastructure.

First thing. While this is an important milestone and a significant development in the evolution of LLMs, it doesn’t come entirely as a shock. The history of computing is replete with examples of dramatic efficiency gains. The DeepSeek development is precisely that – a dramatic efficiency improvement which, in our view, drives incremental demand for AI. Rapid declines in the cost of computing during the 1990s provide a useful parallel to what we are seeing now. As Michael Gapen, our US chief economist, has noted, the investment boom during the 1990s was really driven by the pace at which firms replaced depreciated capital and a sharp and persistent decline in the price of computing capital relative to the price of output. If efficiency gains from DeepSeek reflect a similar phenomenon, we may be seeing early signs [that] the cost of AI capital is coming down – and coming down rapidly. In turn, that should support the outlook for business spending pertaining to AI.

In the last few weeks, we have heard a lot of reference to the Jevons paradox – which really dates from 1865 – and it states that as technological advancements reduce the cost of using a resource, the overall demand for the resource increases, causing the total resource consumption to rise. In other words, cheaper and more ubiquitous technology will increase its consumption. This enables AI to transition from innovators to more generalized adoption and opens the door for faster LLM-enabled product innovation. That means wider and faster consumer and enterprise adoption. Over time, this should result in greater increases in productivity and faster realization of AI’s transformational promise.

From a micro perspective, our equity research colleagues, who are experts in covering stocks in these sectors, come to a very similar conclusion. They think it’s unlikely that the DeepSeek development will meaningfully reduce CapEx related to AI infrastructure. From a macroeconomic perspective, there is a good case to be made for higher business spending related to AI, as well as productivity growth from AI.

Obviously, it is still early days, and we will see leaders and laggards at the stock level. But the economy as a whole we think will emerge as a winner. DeepSeek illustrates the potential for efficiency gains, which in turn foster greater competition and drive wider adoption of AI. With that premise, we remain constructive on AI’s transformational promise.

Thanks for listening. If you enjoy the podcast, leave us a review wherever you listen and share Thoughts on the Market with a friend or colleague today.

DISCLAIMER

In the last few weeks… (Laughs) It’s almost like the birds are waiting for me to start speaking.