While the numbers that Nvidia (NVDA) shared in its latest earnings report were far from perfect, the company did signal it's seeing improving demand in two important areas -- desktop gaming GPUs, and server GPUs sold to cloud giants -- where sales had recently softened.
To recap: Nvidia, which was up 57% on the year going into earnings, is down about 2.5% today after beating its October quarter estimates and issuing below-consensus January quarter guidance that implies a slight sequential revenue decline.
The October quarter revenue beat was driven in large part by stronger-than-expected desktop and notebook gaming GPU sales. January quarter sales are expected to be pressured by sequential drops in notebook GPU and Nintendo Switch processor sales that will overshadow "strong sequential growth" for Nvidia's Data Center segment, which covers server GPUs and the company's GPU-packed DGX systems.
A Desktop GPU Rebound
During a talk with TheStreet following the company's earnings call, Nvidia CFO Colette Kress indicated that while Nvidia sees notebook GPU sales dropping sequentially this quarter -- the company insists this is due to seasonal trends for notebook builds rather than any demand issues -- it expects desktop GPU sales, many of which are tied to retail graphics card sales to gamers, to be either flat or up sequentially. They're also expected to be up strongly on an annual basis, after being depressed in the year-ago period due to a mid-range inventory crunch that followed a collapse in demand from cryptocurrency miners.
And on the call, Kress mentioned that more than two-thirds of Nvidia's desktop gaming GPU revenue now involves its RTX-branded mid-range and high-end GPUs, which feature specialized cores that enable real-time ray tracing (it enables photorealistic game scenes) and deep learning super sampling (a technique for using AI/deep learning algorithms to improve game performance).
The desktop RTX strength is important for a few reasons:
- In spite of the notebook gaming GPU market's strong growth in recent years, the desktop gaming GPU market remains larger, with more demanding gamers often still preferring to do their gaming on desktop PCs.
- The desktop RTX line carries MSRPs ranging between $349 and $999. As a result, its strength is boosting Nvidia's gaming GPU average selling price (ASP), and with it the company's gross margin.
- The pickup in demand comes amid a steady increase in the number of marquee games supporting real-time ray tracing, which Nvidia has strongly trumpeted as a selling point for RTX GPUs since the first ones were launched in September 2018. While ray-tracing hasn't been embraced as rapidly by gamers as Nvidia once hoped, it looks as if interest is starting to pick up some.
Inference Helps Fuel Stronger Cloud Demand
As desktop GPU sales pick up, so are server GPU sales to cloud giants (the proverbial hyperscalers), which were pressured in the first half of 2019 due to a broader slowdown in chip and hardware purchases among hyperscalers.
The Data Center segment's total revenue fell 8% annually last quarter to $726 million due to lower sales to enterprise clients. When asked about the enterprise decline, which comes as Cisco Systems (CSCO) and other enterprise hardware firms report macro pressures are affecting deal closings, Kress attributed the drop to a large, year-ago, DGX deal, while also stating a quarter can see some DGX sales push-outs based on when a customer is ready to receive new hardware.
But while enterprise sales fell, sales to hyperscalers rose sequentially and annually as cloud capital spending begins to more broadly pick up, and CEO Jensen Huang indicated on the call that they'd grow strongly this quarter.
Some of the growth has to do with the growing adoption of Nvidia's Tesla T4 GPUs to perform inference -- the running of trained AI models against new data and user requests. Though a lot of this work has historically been done by Intel (INTC) server CPUs, much of its is now being done by Nvidia GPUs and other accelerators such as Google's Tensor Processing Units (TPUs) and Intel and Xilinx (XLNX) FPGAs, as the demands placed by voice assistants, content recommendation engines and other AI-powered services drive a surge in the inference computing needs of cloud giants.
And while the lion's share of Nvidia's sales to hyperscalers have traditionally involved the sale of powerful GPUs used to perform the training of AI models (the company is still the dominant player in this space), its inference sales, which are benefiting from both the T4's popularity and large software investments, more than doubled annually last quarter. In addition, though it's worth noting the T4 carries a much lower ASP, inference growth led Tesla T4 shipments to exceed those of Nvidia's Tesla V100 flagship GPU for AI training systems and supercomputers (nonetheless, Nvidia says both V100 and T4 sales hit new records).
When asked about Nvidia's inference momentum, Kress noted that whereas the company's inference-related sales were "approaching zero" a little over a year ago, they're now "solidly in [the] double digits" as a percentage of total Data Center revenue. She added that inference is "moving quite well [towards being] a great piece" of Nvidia's Data Center business.
It's also worth noting that Nvidia's sales to hyperscalers are growing even though the Tesla V100 launched in mid-2017 and is starting to get long in the tooth. Nvidia's 2020 training-related sales could get a boost from the launch of a new flagship GPU based on its next-gen Ampere architecture; in October, RBC Capital published a note suggesting that such a product could be announced in the January quarter.
Overall, it's not shocking that Nvidia's guidance is sparking some profit-taking given the large gains its stock has registered since August. But while digesting the guidance, it's worth keeping in mind that the company is forecasting a good quarter for two strategically important parts of its business that not long ago had markets on edge.