Nvidia has announced its Q2 2021 revenues: $3.87 billion, outperforming analyst expectations amid a global pandemic and showing signs of a successful (and ongoing) reorientation of its business strategy. Compared to Q2 2020, earnings were up 50 percent; compared to Q1 2021, up 26 percent. This is another major win for Nvidia, which recently saw itself eclipse Intel as the most valuable U.S.-based chipmaker and now holds a price-to-earnings ratio of roughly 90.
The earnings show clear signs of a shift in the winds for Nvidia, which has identified itself as a gaming-oriented company for virtually its entire history. For the first time ever, Nvidia’s gaming sector revenue (1.65 billion, up 24 percent year-over-year) was outpaced by another earnings in another sector: datacenters ($1.75 billion, up a staggering 167 percent year-over-year). Much of this is attributable to the recent acquisition of networking leader Mellanox Technologies, a deal that was completed just four months ago. The other two notable revenue flows for Nvidia – professional virtualization and automotive – were down 30 percent and 47 percent year-over-year, respectively.
The positive impacts of the Mellanox acquisition are glad tidings for Nvidia, which is said to be closing in on another massive acquisition: microprocessor juggernaut Arm. With Mellanox and Arm in-hand, Nvidia would be able to lean hard into a vertical integration strategy, repositioning itself as a system provider that also provides components rather than solely a component provider. Nvidia has already, of course, dipped its toes into these waters with its DGX system line, HGX baseboards and SuperPOD reference architecture. In emphasizing its system-level solutions, Nvidia has also made it more difficult to obtain certain enterprise-oriented components: its new A100 SXM NVLink GPUs, for instance, are currently only available to customers purchasing DGX systems or HGX baseboards.
The new heading couldn’t come at a better time for Nvidia, which is facing looming competition at the component level in the gaming and datacenter sectors. There were, of course, the now-traditional system wins: Nvidia components were included in eight of the top ten systems (and two-thirds of the total systems) on the Top500 list of the world’s most powerfully publicly ranked supercomputers; Selene – a SuperPOD system – ranked seventh on the list; and the next major supercomputer for the National Energy Research Scientific Computing Center (NERSC), Perlmutter, will contain more than 6,000 A100 GPUs. Nvidia’s A100s are also proving popular in the cloud sphere, with Google Cloud launching A100-based instances and Microsoft Azure adding its own A100-based instances that promise “supercomputer-class AI.”
“It would be easy to say Nvidia has gotten lucky – right-place-right-time – with the surge of AI, but that would do a disservice to the intentionality Nvidia has brought to this market,” said Addison Snell, CEO of Intersect360 Research. “Nvidia has been investing heavily in CUDA and GPU computing for fifteen years; it is one of the few datacenter companies that has the fortitude to stick with a long-term campaign to build a market. Nvidia doesn’t just benefit from the trend; Nvidia helped create the trend.”
Still, more than ever, system builders have competitive options, and in many cases, they’re taking them. The new headliner of the Top500 is Fugaku, an Arm-based system – displacing Summit and its Nvidia Tesla GPUs. Perhaps more worryingly, Nvidia is apparently absent from exascale system planning, with non-Nvidia GPUs slated to power all three of the exascale systems announced so far in the United States: Aurora (Intel GPUs), El Capitan (AMD GPUs) and Frontier (AMD GPUs). For two of those systems – Frontier and El Capitan, both being built by Cray – AMD is also providing the CPUs, underscoring the case for Nvidia’s new focus on vertical integration and, more specifically, its acquisition of Arm.
“Nvidia has a lot going for it and many options, including the potential acquisition of Arm Holdings or another firm with processor IP,” said Steve Conway, senior advisor for HPC market dynamics at Hyperion Research. “The company’s biggest challenge may be moving up from the component level to the system level, which one noted HPC architect said is the difference between designing a lobby and designing a skyscraper. Although Nvidia has been partly self-propelled and partly carried into the datacenter by system vendors, this datacenter experience has given Nvidia an important advantage over some of the newer chip designers.”
Nvidia’s share price is now holding steady after a slight dip from its all-time peak following several months of strong growth.