However, the stock, after a recent sharp run-up, is entering a time of trial that suggests big gains will be harder to come by. Aggressive young chip makers are finally coming to market with products that can displace Nvidia’s “graphics processing units,” or GPUs, after much delay. Until now, these vendors were a theoretical threat to Nvidia. Now, they can be a material threat.
Nvidia shares are up 27% since the beginning of September at a recent $209.42, more than three times the Nasdaq Composite Index’s return in that time. The shares had been under pressure back in the spring as the company’s outlook at the time fell short of expectations because of a continued slowdown in spending on data center technology by giant cloud computing operators such as Amazon (AMZN) - Get Report. But the outlook finally turned around in August and it’s been off to the races since then, especially with this most recent surge in stock price.
But the rise of alternative vendors of artificial intelligence silicon is something different in the way of challenges -- it is an existential threat to Nvidia. For years now, companies such as Graphcore of Bristol, England have been making the case that Nvidia’s GPUs are less than ideal for AI work. Their claims have substance, even if Nvidia has been the dominant vendor of chips for AI workloads. The company’s chips for the data center, of which an increasing amount is for AI, is a nearly $3 billion business for Nvidia, annually.
For a long time, there’s been sound and fury from Graphcore and other startups but nothing to show for it in the way of customers. Then, last month, at a conference in Denver where makers of supercomputer systems gather to show off their wares, Graphcore announced a deal to roll out its chips in Microsoft’s (MSFT) - Get Report Azure cloud computing facility. The same week, Cerebras Systems of Los Altos, California, another startup, announced a computing system it is selling with the world’s largest computer chip, and got glowing reviews from its first announced customer, the U.S. Department of Energy’s Argonne National Laboratory.
These systems from Cerebras and from Graphcore, and ones that will come from other startups, are fundamentally different from what Nvidia offers. They can process the mathematical operations at the heart of today’s “deep learning” forms of AI, operations that consist of linear algebra, much more efficiently than those operations run inside a traditional GPU. Nvidia has anticipated the arrival of these young challengers and has tweaked its GPUs in the past couple years to add better processing of linear algebra. It’s possible that the tweaks may be enough for some clients, but it’s also possible this new wave of AI chips will set a new standard next to which Nvidia’s approach seems like trying to shoehorn things into the wrong pair of cramped loafers.
At the very least, Nvidia may lose some sales initially as customers apportion some money to buy chips from Cerebras and Graphcore that are for sale for the first time. Cerebras management crowed last week that Argonne National Laboratory is not buying any more GPUs from Nvidia.
That can put some pressure on Nvidia’s outlook. Analysts are forecasting Nvidia’s revenue to rise by almost 20% next fiscal year, after a projected decline of 8% this year. Within that, the datacenter business is expected to rise by 33% after an expected drop of 3% this year. That would be a fairly sharp rebound for the datacenter. While Nvidia faces easier “comps” next year, given this year’s sluggishness, that revenue outlook sets a very high bar.
The stock is pricing in a somewhat high multiple of projected sales of 9.9 times. That could certainly come down if the outlook for revenue growth cools.
On top of the competition from Cerebras and Graphcore, Nvidia’s traditional opponents, AMD (AMD) - Get Report and Intel (INTC) - Get Report, appear finally to have gotten their act together with AI computing chips. And Nvidia will have new competition in 2020 from smartphone chip maker Qualcomm (QCOM) - Get Report, which is getting into the AI chip race for server computers.
All in all, Nvidia faces substantial amounts of competition after years of having the AI chip market mostly to itself. It may be that the company can blow past all of these challengers. But times of trial such as these usually allow for a softening of the stock multiple, at least until it’s clear which way things will go. That could mean slower returns for Nvidia investors for a while.