Forecasts for data-center server growth far exceed any that occurred during the rise of the internet, so while one analyst is positive long term on the AI gold rush, he warns the risk of building too much too soon also runs high.
This past summer, graphics-processing unit giant Nvidia Corp.
reported a 141% leap in quarterly data-center sales to $10.32 billion as the chip maker began recording sales of AI-focused chipsets that were in very high demand from hyperscalers and public-cloud providers.
On Tuesday, Bernstein analyst Toni Sacconaghi reminded investors that it wasn’t too long ago when the server market was a sleepy, single-digit-growth business even as the internet changed the world. Sacconaghi said the compound annual growth rate of the server industry was 3% over the past 25 years, and had “essentially no growth between 1998 and 2014, a period that includes much of the build-out of the internet!”
“While we are constructive on AI longer term, we do worry that projected AI
infrastructure build-out may be occurring too quickly, necessitating a digestion
period, which could result in a commensurate stock pullback in AI-related names,” the Bernstein analyst said in his note. A “digestion period” is common industry euphemism for a glut.
Sacconaghi said AI could change the world even further, noting three possible outcomes, though he doesn’t “have conviction in any of them necessarily occurring.”
First, he said that while there are more consumer-facing AI applications like OpenAI’s ChatGPT in development, most “do not initially have a clear path to monetization.” Secondly, while AI could drive a massive increase in business spending, adoption is still in the early stages, and may not drive medium-term expectations. Finally, data-center growth could become “massively cannibalistic” to traditional servers and IT spending.
Read: As AI matures, Nvidia won’t be the only pick-and-shovel company to thrive, BofA analysts say
“We believe that some cannibalization will likely happen, but that IT decision-makers may struggle to justify cutting spend on existing initiatives that are already producing productivity gains,” Sacconaghi said.
The Bernstein analyst has overweight ratings on both Dell
with an $80 price target, and Hewlett-Packard Enterprise Co.
with a $20 price target.
Also read: Will AI do to Nvidia what the dot-com boom did to Sun Microsystems? Analysts compare current hype to past ones.
Recently, Bernstein analysts looked at the nascent AI market through the lens of other big tech infrastructure buildouts, and some flops. Sacconaghi said the server market took about six to seven years to double after 2014. For cloud service providers, the hyperscale server market has grown about 17% on an annual basis since 2017, “and much of that growth came from cannibalizing traditional servers.”
With estimates of compound annual growth rates of 75% or more over three years, and a doubling of the server market, Sacconaghi called those projections “a historically unprecedented level of growth.”
And it’s not just the big players like Amazon.com Inc.
and Alphabet Inc.’s
Google in the cloud. Analysts are also surprised by demand from smaller cloud and consumer customers.
Shares of Nvidia are up 200% year to date, while the PHLX Semiconductor Index
is up 37%. Dell shares are up 69%, and HP Enterprise shares are up 6%.
‘Unprecedented’ AI server buildout may be happening too quickly, this analyst warns