As Nvidia and other semiconductor stocks reach all-time highs supplying high-end chips to power AI applications, the same group of firms are also usually first to end up in a bubble.
This is according to Wellington Management’s Brian Barbetta, a portfolio manager on the firm’s technology team.
“I would expect you’ll see a bubble in some of the hardware providers,” he told a recent investment conference in Hong Kong. “In hardware, this is where in every tech cycle that I’ve ever studied has a bubble.”
He warned that companies providing chips and hardware to the cloud service providers are often victim of overly buoyant forecasts from their customers that come back to bite during a supply glut, as has happened in past bull-cycles for the semiconductor industry.
He said: “What will end up happening is the application providers will give demand forecasts to cloud services providers, and they’ll then start to buy based on demand forecasts – and as anyone who has analysed forecasts know, they’re often far too optimistic.”
“Then there’ll be hardware shortages that will drive pricing power – that will drive tremendous revenue growth – and then it will come crashing down at some point in time and we’ll have to reset and get started again.”
He said that the semiconductor industry is undoubtedly cyclical and although supply constraints currently exist, at some point there will be excess inventory.
Shares in the world’s leading AI chipmaker Nvidia are up 220% over the past year as its customers race to order the latest AI chips.
Performance of Nvidia over 1 year
Despite the surge, some managers believe the stock price has not yet run ahead of its fundamentals.
Barbetta said one way to avoid buying into a bubble is by looking at the revenue being generated by the companies serving the end user.
He explained: “The way I try to look at this is: does revenue the business is generating sync up with the economic activity that’s being created at the end of the chain?”
“If I were to look at a hardware company today selling billions of dollars of chips that these programs are running on, I need to see that those programs are going to generate enough revenue, that the amount of money being spent on the hardware makes sense,” he said.
“I think that’s where you’ll see the initial disconnects in this market start to come to pass; that’s where we’ve seen it in the past.”
During the cryptocurrency booms of 2018 and 2020, a lot of money was spent on buying GPUs to mine cryptocurrency, which in both instances eventually didn’t “sync up” with the money being spent on hardware, Barbetta said.
He added that even during the early advancements of machine learning, hardware companies experienced a similar boom-and-bust in chip sales.
Building durable moats
Rather than investing in the most recent beneficiaries of the hype surrounding AI, such as the semiconductor companies, Barbetta instead favours the cloud service providers where he is seeing some “really interesting investment opportunities”.
The biggest cloud service providers currently in the market are Amazon Web Services, Microsoft Azure, Google Cloud and Oracle – three of which form a major part of the ‘Magnificent Seven’ stocks which drove markets in 2023.
Barbetta said: “This is where we think companies are building durable moats because of their size, their access to hardware, and their access to models that a lot of these AI programs are being built on. This is a fruitful area for investment for us today.”
Although some investors may be wary that much of the benefits these companies will garner from AI have already been priced in – Barbetta emphasised the importance of accurately forecasting their potential future growth in earnings.
He said: “The stocks that might look cheap today may in fact be quite expensive because of the fact that they’re either over-earning or they have massive disruption risks.”
“Among stocks that might look expensive, we’ll find some of them are actually downright inexpensive when we see the growth they’re able to deliver as a result of this technology.”
