DeepSeek, Black Swans & AI

There was a typo in my last alert, Blast-Off should have read Blast-Down. The Chinese have invented an LLM (large language model) which performs like ChatGBT or better but runs on data centres found in Christmas crackers.

This astonishing news led to an incredible day in Monday’s US stock market where Nvidia’s loss of nearly $600bn in market value was the biggest in US stock market history.

I am not qualified to pontificate on the significance of DeepSeek. Still, I find it hard to believe that replacing the staggering computing power of the latest data centres and the exponentially increasing masses of data on which they feed can be that easy.

Barrons magazine, on Tuesday, provided what seemed to me a balanced assessment of the situation.

On Monday, Nvidia shares closed 17pc lower, marking their biggest daily percentage decline since March 2020. The catalyst was Chinese startup DeepSeek’s AI models, which appear to rival their leading Western counterparts — at a fraction of the training cost.

Nvidia saw its market value fall by $593bn, the biggest one-day drop for any listed company on record.

“DeepSeek is an excellent AI advancement and a perfect example of Test Time Scaling,” a Nvidia spokesperson said in an emailed statement. “DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely-available models and compute that is fully export control compliant.”

Test-time scaling refers to the ability of new AI models to think about queries for longer time periods to come up with better answers.

DeepSeek’s advances led many investors to question whether this is the beginning of an end to the rapid rise in AI spending but some analysts say this worry is overblown.

“Longer term, a faster flywheel of innovation means a bigger AI [total addressable market],” wrote Mizuho Securities analyst Vijay Rakesh, who kept an Outperform rating and $175 target price on Nvidia stock.

The immediate test for Nvidia will come when Microsoft and Meta Platforms report their earnings on Wednesday — and whether they suggest the next generations of their own AI models will require increased hardware spending.

“Given DeepSeek is based on the distillation technique that leverages other AI models, we would note cloud service providers are still required which is evidence of continued strong growth in spending in AI,” Citi analysts led by Christopher Danely said in a Tuesday note.

While Nvidia has had an incredible run in the past few years, more than doubling in value in both 2023 and 2024, it has also suffered some massive falls. Out of the 10 largest market cap one-day drops in history, Nvidia accounted for eight of them. Meta and Amazon have also seen their market value drop by over $200bn in one day, both in 2022.

“We believe that AI graphics-processing unit demand still exceeds supply, so while slimmer models may enable greater development for the same number of chips, we still think tech firms will continue to buy all the GPUs they can as part of this AI “gold rush,” wrote Morningstar strategist Brian Colello.

Barrons, 28 January 2025

I have been reading stuff about the impact of DeepSeek. One idea is that it will act as a massive accelerant on the global adoption of AI with exciting consequences for the world economy. Another suggestion is that there is bottomless demand for computational power so all that money spent on Nvidia’s chips should not be wasted.

The news from DeepSeek hit the market with stunning impact making it a Black Swan event. All important stock market moves start with a Black Swan event, whether for the stock market as a whole, for a sector or an individual share.

The original launch of ChatGBT triggered the initial AI excitement which received a further boost when Nvidia began to massively upgrade its forecasts for sales and earnings on the back of huge demand for its GPUs to be used in a new generation of super-powerful data centres.

The latest Nvidia chart shows the moving averages bunching together ahead of a significant move. Jensen Huang will come out with all guns blazing when the Q4 2025 results are announced on 26 February. All eyes will be on the forecast for the year to end January 2026. Technically the shares look a hold awaiting the next buy signal. You could say that is always the case for companies of the quality of Nvidia.

Great companies do fall from grace. IBM, Cisco Systems and Intel are shadows of what they were in their glory days but it seems premature to think that is about to happen to Nvidia when the AI boom in which it plays such an important role is still so early in what will be a breathtaking journey.

What does seem likely is that this news is going to take some time to absorb. It raises questions about the hugely capital-intensive approach that Nvidia’s customers are on and over what is happening in China and how they have progressed so rapidly.

Strategy – Dealing With A Shock

Nvidia is trading at the same price it was eight months ago so it is building a large consolidation. The monthly moving average has rolled over and is falling. You could sell some and await the next clear buy signal. There are other exciting shares in which to invest. If you do decide to keep the faith you will be in good company. Jensen Huang is an extraordinary man on a mission. It is like Tesla and Elon Musk. You write these guys off at your peril. Plus it is very early days in estimating the true significance of DeepSeek. It is another landmark in the global march of AI.