An Nvidia commissioned economic impact report for DGX-1 was completed by Forrester earlier this year. I just stumbled upon it and it had some insights I wanted to share including a previously unknown (to me and from what I can tell, most of the investing community) revenue stream for Nvidia.
A quick rehash on the Nvidia DGX-1. Last year Nvidia announced this product that is essentially a supercomputer server featuring 8 Tesla V100 GPUs and all the supporting hardware and software equipment including NVlink interconnects and CPUs. All in a box. Because of the infrastructure it acts like one big really fast gpu (1PetaFlops or 1,000 TFLOPs) that can plug in and get a company up and running AI data science with ease. It can also scale in hyper scale data centers. It is the building block for super computers coming online around the world including Nvidia’s own Saturn V. This year Nvidia launched the DGX-2 with 16 V100s and updated NVSwitch with greater bandwidth among other upgrades. And also the HGX-1 and 2 which is the basic infrastructure for the GPUs and a roadmap for a hyper scale companies and OEMs to make their own DGX style box. Nvidia in their recent ER said that DGX is selling in the “hundreds of million in annual sales”.
https://www.nvidia.com/en-us/data-center/dgx-1/
This is competitive advantage in a box for all data forward companies, which in today’s world is pretty much everybody. The Forrester report makes this apparent if you want to skim through it.
https://www.e4company.com/imgs/dwnld/20/Forrester%20Consulti…
Through their interviews with companies that have deployed DGX-1, Forrester estimates an ROI of 294%.
Time to implement a new DGX system is reduced by 2 months vs legacy.
Time to complete model training is reduced by 3 days on average.
Data scientists have more time to experiment and develop new data products and don’t have to spend time on mainaning and updating the infrastructure. NVDAs software means they don’t need to spend time programming to make models work on the equipment. They already work!
Here in lies the hidden revenue stream that this report reveals.
The company would purchase a support contract from NVIDIA for $67,050 per year commencing with the purchase of the DGX-1 units.
A DGX-1 cost about $149,000. But the companies buy a 3 year service contract worth $67,050/yr. That’s an additional $201,050 per DGX-1. That is contractual recurring revenue that if the sales of DGX are in the hundreds of million annually would mean this is a revenue stream worth potentially hundreds of millions annually. According to Forrester the companies are very happy with Nvidia’s service and it saves money compared to the way their legacy support system worked. What is it for DGX-2? Do they have other major products that have similar service contracts? Could this be a major contributor to Data Center or other segment revenue going forward?
This has not been discussed much with Nvidia, either by the company or in analysis. To me, this is a huge revelation. I’d like to find out more.
Darth