After the price more than doubled to $123 per share, I issued a SELL rating – see my article Nvidia Likely To Lose Its Dominant Position – because I believed that the stock had become significantly overvalued even assuming it would deliver on exuberant growth, margin, and market share assumptions
The stock has traded flat since that call made last June, but some things have changed. In particular, we now have much better visibility into what AI could actually do for the world and how much computing power it will take.
Before jumping into my forecast for Nvidia, I want to spend some time on the AI vs dot-com bubble comparison.
There is an argument to be made that the impact of a new technology is often overestimated at first. It often takes many years for that technology to actually translate into profits. As a result, new technologies are inherently prone to bubbles. Take the internet, for example. Everyone had heard of it by 2000, but it wasn't until 10–15 years later that it really changed the world. And in case of the steam engine, monetization of the technology took well over a century.
A good way of illustrating this is looking at the decoupling of the stock price and forward earnings of Cisco Systems (CSCO) – a stock which was very much believed to be a major beneficiary of the Internet in 2000. This decoupling from fundamentals and more specifically the failure of the new technology to translate into earnings is, in my opinion, one of the main reasons the bubble popped.
In the case of AI, it's fair to say that it hasn't translated into substantial earnings growth for the vast majority of large corporations yet. Moreover, adoption remains low across most sectors, suggesting that monetization of AI across the board is relatively far away.
The one exception to this are, of course, chip manufactures that stand at the very start of the AI supply chain. As a result, they don't need the technology to translate into profits. All they need is enough hype so that companies (mainly the hyper-scalers) feel the FOMO (fear of missing out) and pre-order their chips in order to not be left out. This works quite well in the short to medium-term and has caused Nvidia's earnings to skyrocket. As a result, the price, despite its steep rise, has continued to trade in line with fundamentals – a major green flag compared to what was happening during the dot-com bubble.
If it does, Nvidia is worth a look at the right price, but if it doesn't Nvidia's earnings are bound to drop substantially as its customers realize that AI is not making them money, they stop ordering new GPUs, and Nvidia revenues inevitably drop.
Therefore, the ability to monetize AI, in a reasonably short amount of time, presents the single biggest risk to buying Nvidia stock today.
One such expert, Daron Acemoglu – a professor at MIT quoted in a Goldman Sachs study, is not very optimistic. Here's a short piece from that study:
He estimates that only a quarter of AI exposed tasks will be cost-effective to automate within the next 10 years, implying that AI will impact less than 5% of all tasks... he forecasts AI will increase US productivity by only 0.5% and GDP growth by only 0.9% cumulatively over the next decade.
Goldman Sachs (GS) published another study, very much in line with this view, expecting AI hardware investment in the U.S. to peak in 2026 at around 1.6% of U.S. GDP or roughly $430 Billion.
Later on, Goldman expects spending to level off to 1% of GDP by 2028 and eventually 0.5% by 2032. Assuming 2% GDP growth over time, this would correspond to AI spend of $280 Billion in 2028 and $155 Billion in 2032.
This is an important forecast because it effectively forms the ceiling of what Nvidia can expect to get in revenues over the long term. The company currently has around a 90% market share in the AI-chip market today, but in a competitive market, it's likely that its market share will come under pressure over time.
In light of the above, the current consensus for revenues seems off in the long term, as it has Nvidia at expected sales of $377 Billion in 2032, 55% above the expected total worldwide AI spend that year.