Ai selloff was overdone

Monday was about discovering the Chinese version of ChatGPT – called Deep Seek, and watching the US markets get hammered by the news. Truely, it didn’t feel disruptive to me – as Deep Seek mirrors ChatGPT which already provided that wow effect two years ago. Plus, it avoids China-sensitive subjects by proposing to talk about something else. Yet the fact that the company claims to have build its model with cheaper and less advanced chips significantly shifted the cost and ROI outlook for AI investments and led to a trillion dollar selloff in the US stock markets. The selloff was certainly amplified by high valuations and the ongoing fear regarding the massive AI investments and the longer-than-expected payoff period.

As such, Nvidia tanked nearly 17% yesterday and erased more than half a trillion dollar in market cap by itself! It was apparently the biggest single stock fall of all times. But hey, in the list of the top ten biggest single stock selloffs of history, you can count the name of Nvidia about 8 times... AMD plunged more than 6%, Intel – which made a positive attempt at the open on hope that its less advanced chips could replace Nvidia’s – ended up losing more than 2.5%. ASML fell 7%. And overall, Nasdaq 100 lost nearly 3%, VanEck’s semiconductor ETF nosedived almost 10% and Global X Uranium ETF tanked more than 11%. It was a rough day. Interestingly, the names that are said to be linked to DeepSeek climbed, but the gains remained very limited. Iflytek for example gained less than 2%, while Nasdaq’s Golden Dragon index was barely up by 0.10%. Consequently, money didn’t move from US tech to Chinese tech, it moved from US tech to safe haven. The US yields fell, the franc rallied but gold couldn’t benefit from the market panic – to my surprise. The British FTSE 100 on the other hand gave no reaction to the market rout, highlighting the very small exposure of the British big caps to technology – which is disquieting but could be interesting in case of a deeper tech selloff.

I believe that yesterday’s AI selloff was OVERDONE and the futures look better this morning, with Nasdaq futures even up by 0.20% at the time of writing. There are reports praising DeepSeek’s performance, some experts say it’s impressive, others say it’s disruptive, and Nvidia itself said that the company came up with something ‘excellent’ – using a lot of its less advanced chips. But beyond the fact that the company used less advanced and cheaper Nvidia chips to build its model, there are a lot of unanswered questions about DeepSeek, including whether its model could be integrated and used by other applications and whether the company really built a model for less than $6mio whereas the price mark of the US AI models reaches several hundred million dollars. And last but not least, DeepSeek looks like it made something that already existed for a cheaper price. But it did not come up with an end product that did not exist.

But if DeepSeek successfully does what it says it does bring equal performance AI models for a cheaper price - it will clearly help the Chinese local players, and all-sized companies around the world that have limited budgets to integrate AI models into their daily lives. The latter will increase the demand for less advanced chips than Nvidia’s best performers, but it will increase demand for chips, still. In this context, we were already pointing at a growing window of opportunity for alternative chip makers – like AMD – in the process of wider adoption of AI models with cheaper chips. We now have a stronger conviction in this view. As such, yesterday’s selloff could move capital around and benefit to the makers of cheaper chips that could appeal to a larger client base than the US Big Tech.

What about Nvidia?

Nvidia sells premium chips to the world’s biggest tech companies that have all the interest in the world to use the most premium chips to protect their advance and come up with disruptive innovations. For these companies, it’s not only about increasing productivity, it’s also about innovating. Therefore,

1. All of a sudden, the high concentration of Big Tech clients in Nvidia’s client book – where big tech clients accounted for more than 50% of its revenue in the Q3 – looks like it could be an advantage to weather a potential demand shock.

2. Deep Seek will hardly disrupt the US tech companies’ dominance in AI. True tech dominance requires breakthroughs in hardware, software and infrastructure. And we don’t see a revolutionary advance from Deep Seek.

3. Big Tech’s ambition to build their own chips is a bigger threat to Nvidia’s revenue than an increased demand for cheaper chips.

To sum it up, the next big leap in AI is expected to be AI agents capable of executing a series of tasks in a specific order. The race to achieve this milestone is on, and it will be fascinating to see who gets there first. One thing is clear: progress is maximized with top-performing chips rather than average ones. After all, a Peugeot will get you from point A to point B, but a Ferrari will get you there much faster.