Nvidia is making headlines again with the Mellanox acquisition, albeit more positive headlines than the previous quarter. On one hand, the Mellanox acquisition could negatively impact Nvidia’s financials as Mellanox is posting a very small profit margin of 2%. The acquisition also required all of Nvidia’s cash reserves, and the company could potentially have to take on debt moving forward. The prospects for growth and how Nvidia will use Mellanox is not straight forward beyond acquiring interconnectivity and network solutions to bolster data centers, which is why this acquisition requires further discussion.
In this analysis, I speculate how Mellanox can add to Nvidia’s growth potential both strategically and tactically, and why there is still lots of runway. This analysis is an update to Nvidia: Two Impenetrable Moats which I wrote at the height of the crypto bust, proving my analysis is not fair weathered or a reaction to the markets. I defended why I was keeping my investment in Nvidia when the market was against me, which are the same reasons today, and will be the same reasons 5 years from now. Nvidia has an unshakeable developer ecosystem and an unshakeable future with the GPU-powered cloud, with the later recently strengthened with the Mellanox acquisition. Crypto was never Nvidia’s story, but it did provide an excellent entry for anyone who missed Nvidia the first time around. Keep in mind, Nvidia will likely trade sideways for at least another quarter before the excess inventory is absorbed.
X-Factor: Putting the Pieces Together
James Huang had a great quote recently, which describes Nvidia’s ongoing cooperation with CPUs as the necessary backbone to GPUs, and why his focus has been elsewhere in the stack. This quote helps to provide a glimpse into his future strategy:
“These two types of processing are going to be here to stay. With accelerated computing, we don’t suffer from Amdahl’s Law – we obey it, and the thing that you don’t accelerate becomes the critical path. We believe in fast CPUs, and that is why we work with all of the world’s fastest CPU makers – IBM, Intel, AMD, Arm.”
He went onto say that he is focused on the X factor, or what will accelerate the path forward at the highest percentages possible. Rather than compete with many other players on what is the least expensive piece in the stack where many companies compete (CPUs), Huang is focused on how to become the top and only player on the most expensive pieces of the data center stack (parallel computing and acceleration).
To illustrate how Mellanox accelerates the performance of GPUs, Nvidia and Mellanox support more than 250 of the world’s Top 500 super computers, including the world’s two fastest supercomputers, Sierra and Summit, operated by the U.S. Department of Energy. In this situation, the 10x better power consumption and 4x better general-purpose compute from FPGAs is not missed. That should have competitors Intel and Xilinx concerned as these two products are now under one company.
Here is a hint that Nvidia will use Mellanox to narrow the performance and latency gap between GPUs and FPGAs, or any other competing chips, while continuing to be the easiest and most universal platform for developers.
“If we want to create something extraordinary for the future, we have to align on architecture and vision early,” James Huang on the Mellanox acquisition
In other words, I do not believe Nvidia acquired Mellanox to simply own InfiniBand and ethernet, but rather to boost GPUs as the best data center option available on the market today and into the future. As stated when discussing FPGAs, the challenge GPUs potentially face is competing chips that test 10x better in power consumption and 4x better for general purpose compute. Nvidia is clearly aligning its architecture with speed, which may be the highly competitive combination as development on Nvidia’a architecture is more universal than FPGAs.
How Does Mellanox Fit In? Technical Specifications.
Mellanox offers networking and interconnectivity for throughput, which is the amount of data moved successfully and measured in terms of megabits per second (like your home internet) or gigabits per second (data centers).
Data centers need a lot of throughput for lower latency to deliver data faster and unlock system performance. Mellanox is the supplier for this, including network and multicore processors, network adaptors, switches, cables, software and silicon (you get the idea). Mellanox’s products are used to accelerate application runtime and maximize results for high performance computing and enterprise cloud data centers.
Mellanox supports Virtual Protocol Interconnnect (VPI), and allows both Ethernet and InfiniBand to coexist on one platform. This combines the strengths of two rival network solutions. Mellanox has done an excellent job of eating market share from Ethernet incumbents such as Cisco, Arista Networks, Juniper Networks, Hewlett Packard, Dell and also Intel. Some of this is due to InfiniBand, and now a hybrid of the two, to provide the bandwidth for as cheap as possible through the ubiquitous Ethernet, yet provide higher throughput and fewer bottlenecks during high-loads through InfiniBand.
Ethernet is the most used internal system in the Top 500 report released at ISC in 2018 and InfiniBand the second most-used at 247 to 140, respectively. However, InfiniBand connects the most high-powered computers where the presence of ethernet is nearly non-existent.
Note: Mellanox is the dark blue InfiniBand and Intel is the dark red OmniPath
Intel’s OmniPath Architecture (OPA) connects 48 of the top 500 HPCs. This helps Nvidia become more of a data center ecosystem company by providing both the compute and network. InfiniBand continues to lead, and recently, was first to ship 200 Gbps HDR adapters and switches while Intel’s OPA200 will ship later in 2019.
Most importantly, Mellanox is diversified and does well with both ethernet and InfiniBand, as ethernet revenue grew 81 percent year-over-year due to strong traction for 25 Gbps and above solutions. Back in 2011, Mellanox released SwitchX multi-protocol ASICs, which is a single chip that can change protocols on demand by putting a single InfiniBand ASIC next to an Ethernet ASIC. Mellanox is the number two supplier of Ethernet network adapters (NICs) behind Intel.
Takeaway to Mellanox’s specifications: Mellanox is used in the top 3 fastest high-performance computing (HPC) systems and has the capability of pushing intelligence into the network to reduce the load on CPUs. Nvidia’s strategy is to double down on the data center, and with Mellanox’s speed, James Huang’s goal will be to reduce the attractiveness of competitors, namely FGPAs, while reducing the requirement for CPUs from companies like Intel or AMD.
In regards to the Mellanox acquisition, James Huang stated his plans are to create the future. This does not seem like a static, fixed acquisition to me where Nvidia will add Mellanox as-is Certainly, Mellanox’s financials and the price of the acquisition suggest something more strategic.
Nvidia is attempting to close the competitive gap in performance between FPGAs and their own GPUs through Mellanox’s InfiniBand and ethernet. This will offer the preferred developer platform, which Nvidia already has, and a superior performance, which the current acquisition is leading up to. This should have primary competitors concerned – Xilinx and Intel – as these two products are now under one company. If Nvidia can pull this off, which I believe James Huang most certainly can, they will be the go-to combination for artificial intelligence and machine learning computations.
I believe Nvidia will leverage Mellanox to speed up GPUs and close the gap on latency performance with FPGAs (Xilinx/Intel). Meanwhile, as stated in a previous analysis I wrote a few months ago published on FATRADER, Nvidia has one of the very best competitive moats in technology – developer adoption. Mellanox strengthens the second moat I previously covered, which is the GPU-powered cloud. Despite the blood bath we saw over crypto, Nvidia’s story remained intact, and I believe we could see on order of magnitude to its market cap once AI and ML startups that use GPUs reach maturation and go public.
For disclosure purposes, I’ve been invested in Nvidia for some time and bought more during the crypto bust. I am at $160 per share on Nvidia and am comfortable here for the long term. Many of my readers used my contrarian Two Moat analysis to build a position in Nvidia at $130. Please be aware, I expect Nvidia to trade sideways through the next earnings season as guidance is flat earnings.
I am a tech analyst, not a financial advisor. These are my opinions.