Welcome to MT Capital Research! If you’re new, join a group of 5600+ Informed Investors who receive frequent reports on the World’s Most Interesting Companies by subscribing below:
Today I will be completing an in-depth analysis on Nvidia NVDA . Within, I will break down the plethora of opportunities the company has in front of it, and attempt to provide a concise overview of the technological capabilities they possess. In addition, I will break down each of the company’s business segments, dig into the company’s financials, and discuss some positive and risk-based factors alike that I deemed important when trying to perform a scenario analysis on the company as a whole. The format of this research, as per usual, can be seen outlined below. Please feel free to skip ahead to the sections of this research that you feel will add the most value to your research process:
1.0 Sponsor
2.0 Introduction
3.0 Opportunity
4.0 Technology
5.0 Financials
6.0 Discussion
7.0 Conclusion
Disclaimer: The information and research contained herein is all my own opinion and should not be used as a substitute for proper due diligence. Please consult your financial advisor and evaluate your financial circumstances before making any investment.
If you would like to see more frequent investing-related content, give me a follow on Twitter below:
With all of that out of the way, let’s get started.
1.0 Sponsor
Today's newsletter is brought to you by The Weekly ChartStorm. The Weekly ChartStorm has quickly grown into one of my favourite newsletters and provides succinct and useful datapoints that I can leverage within my own macro framework.
The Weekly ChartStorm brings you a selection of charts each weekend covering the S&P 500, US equities in general (including sectors/styles, industries, etc) and even global equities. The charts cover a broad range of issues such as technicals, flows, sentiment, macro, earnings, valuations, fundamentals, and more.
It features a good mix of short-term tactical insights as well as longer-term perspectives to ponder.
And what is somewhat rare these days is that it also takes a very balanced view. No politics, no permabear or permabull BS, just pragmatic insights with clear explanations and commentary around the charts to help keep you on top of the market outlook.
Come and check it out for yourself and see why it has attracted such a strong and loyal following over the years.
2.0 Introduction
At the time of writing, NVIDIA is the 15th largest company in the world, with a market capitalization of nearly $400B. The company has ridden on a host of tailwinds that are extremely hard to rival, ranging from the rise of parallel computing, and by association graphical processing units (GPUs), to an acceleration in the adoption of gaming that puts the digitization of other entertainment forms to shame. The result? A nearly 20,000%+ return all-time, if you were to have bought into the company’s story in its 1999 IPO, a journey that has followed quite the exponential trajectory, as outlined below:
When one starts to dig a little deeper however, they start to notice a pattern. Put plainly, the company’s journey to the megastar it is now has been far from easy. Throughout their two decade plus history, NVIDIA has had to reinvent itself on numerous occasions, actions spawned out of necessity, with the below figure outlining the sheer volume of instances that the company has had drawdowns that would likely put even the best operators out of commission:
Instances of savvy business maneuvers include commissioning the first GPU and intertwining these pieces of technology within the fabric of modernity, being able to fend off Intel’s anti-competitive behaviour with PCIs (Intel would allow for different types of PCI chips to plug into their motherboards but over time would include these additions as a stock feature of the motherboards themselves, oftentimes crushing competition), making noticeable gaming advances with programmable shaders and powering the first Xbox, positioning themselves and their parallel computing capabilities at the forefront of the scientific computing market that would later take the world by storm, and making further advances in the video game space with ray tracing technology. All of these moves were made despite drawdowns of epic proportions caused by a tech-bubble collapse in the early 2000s, the Great Financial Crisis that followed suit a few years later, a series of earnings misses etc. The point I’m trying to convey here is the company’s adaptability. As the Game of Thrones euphemism goes, you cannot kill what is already dead, and looking at NVDA’s ability to navigate turmoil and come out more resilient, the company may be an undead giant that continues to defy expectations. That will be the focus of the deep-dive today. My hopes is that this research will provide you enough of a knowledge base to start to analyze whether or not the current drawdown in NVDA’s stock price will be looked back at in a similar light to all of the now-minute speed bumps in the company’s past.
3.0 Opportunity
Paying any attention to the tech ecosystem over the course of the last few months would’ve resulted in one coming across Nvidia’s most recent investor presentation. Contained within were lofty projections and claims, values that are typically expected from the likes of pre-revenue startups, rather than one of the most valuable companies on planet earth. This may sound a little harsh, but i’ll let you make your own classifications after absorbing the slide below:
Nvidia estimates that they service an estimated $100T worth of Industries, and further concludes that a 1% total market grab will result in the obtainment of a TAM of approximately $1T, distributed between the Gaming, AI, Chips and Systems, Omniverse and Automotive business segments. With respect to each of the above segments, I will attempt to provide more reasonable quantifications in order to gain a sense of how exactly Nvidia fits within each of them, respectively.
Gaming
Although the exact value up for grabs in this business segment is debatable, the sheer size of gaming as a whole is tremendous. As a media cohort, Video Games have experienced insane growth over the course of the last few decades, a phenomenon that we can see corroborated by the data below:
In 2021 as an example, Total Gaming Expenditures on both console and PC, as estimated by a newzoo report, totalled approximately $85B, with DFC Intelligence estimating that video game hardware purchases, including graphics cards, gaming keyboards etc. were approximately $82B. Although these numbers are most definitely tainted from the lingering effects of pandemic-induced tailwinds, phenomena that may have lead to the most ideal circumstances for the gaming industry in its entire history, I think we can conclude with reasonable soundness that the space is large. This bodes well for Nvidia, existing as a staple of a space that will likely continue to experience a positive trajectory given its intertwinement with modern society.
Enterprise Computing
Enterprise Computing, in my eyes, is a segment that encapsulates Nvidia’s presence within both AI and the “Omniverse”, as well their provision of the hardware (such as GPUs, DPUs (Data Processing Units), etc.) & software that powers these computationally intensive workloads and beyond. Starting off with AI, the developments in this area are arguably some of the most exciting trends in computing underway at the moment, with projections largely aligning with that assertion in terms of magnitude. Fortune Business anticipates that the market value of Global AI will reach approximately $270B by 2027 and PwC Global estimates that the total contribution of AI to the global economy will be approximately $16T by 2030. Although these, and the projections made in Nvidia’s recent investor day are very ARKK-esque, there is no question that this trend has unfathomable applications within almost every industry known to man, of which will likely accelerate the productivity and output of most things it is applied to. With a host of investing spend being directed to this space, and the fact that Nvidia possesses a full-stack AI offering directed at customers, it is exciting to think of what the impact this segment will have on the company’s business overall on a longer time horizon, but more on that later. Outside of AI, the company’s hardware & software is arguably powering most, if not all, of the remaining large computing trends. With largely unbreakable bonds to the cloud, data centres etc., Nvidia stands to benefit from the continued digitization of our world.
Moving on, the Omniverse can be defined as a real-time graphical collaboration platform created by Nvidia, commonly used for both visual effects and the creation of “digital twins”. The remarkable capabilities of this platform are nicely encapsulated by the following video:
Digital twins on the other hand are a relatively new development, of which can be defined a real-time digital counterpart of a physical object or process. Nvidia offers a suite of tools for companies to both create and leverage the power of digital twins. Although these applications appear eerily similar to the euphoric touting of the Metaverse we’ve all seen of late, they have real applications, with the likes of Amazon, Pepsico, large auto manufacturers etc. all using digital twins to unlock more operational output and productivity.
Automotive
Autonomous driving is one of the most complicated engineering problems known to man and there is no shortage of companies investing in trying to both solve this problem and be at the forefront of this space. CB Insights estimates that there are nearly 40+ big-name corporations directing capital at autonomous vehicles, ranging from the likes of big-name auto manufacturers, to the largest tech companies in the world, such as Apple, Amazon, Intel, etc. Although this is not a material component of the company’s business today, it may very well develop into a tangible value-driver.
With this brief overview of the opportunity the company has in front of it in mind, let’s delve deeper into the technology behind the beast, so to speak.
4.0 Technology
To say that Nvidia’s technological offerings are diverse would be an understatement. For the sake of simplicity and ensuring that this report stays under the length of a novel, I will briefly describe the company’s offerings contained within the Gaming, Data Center, Professional Visualization and Automotive Business segments. Descriptions will be by no means comprehensive and should help facilitate further research action on your part if you feel compelled to do so.
Gaming
The company’s primary offerings in the gaming space are their GPUs. Contained within this product segment are the GTX and RTX graphics cards, which at the time of writing, can be seen summarized below:
RTX 30 Series
RTX 3060-3090 TI / 3060-3090
RTX 3050
RTX 20 Series
GTX 16 Series
Although the GTX hardware is undoubtedly powerful, top of the line gaming performance is primarily supported by RTX offerings, with individuals being able to leverage the functionality of Ray Tracing, Deep Learning Super Sampling and Tensor Cores.
Ray Tracing, prior to the last few years, was primarily a technique within the TV and film industries, used as a means of producing realistic lighting effects. Prior to Nvidia’s hardware being able to support the computational requirements of Ray Tracing as a whole, a technique called rasterization was commonplace. In essence, the game code directed the GPU to create a 3D scene using polygons, commonly triangles, in order to makeup the visual elements displayed on the screen. After the scene is drawn using said polygons, it is then rasterized, which essentially means the individual pixels are processed by shaders, adding features such as colours, textures and lighting effects to create a frame of the game, and is repeated many times per second. This process, as you may have guessed, had a difficult time tracking how exactly light portrayed in a scene would interact with individual pixels, causing in-game effects to suffer. Ray Tracing on the other hand doesn’t encounter these same downfalls. Acting opposite to how real world light interacts with objects, i.e. photons bouncing around a room after being emitted from a light and eventually reaching your eye, ray tracing traces individual rays from the camera's point of view, tracking how each ray in the scene intersects with the objects therein, tracing it back to the original source. As a result, multiple effects of absorption, reflection, refraction etc. can then be created. These two processes are neatly encapsulated and compared by the following images:
DLSS on the other hand is a type of video rendering technique that is aimed at reducing the pressure put on GPUs, rendering frames at a lower resolution and then using AI to upscale the frame to a higher resolution. As an example, if you wanted to get more performance running a game at 4K, DLSS could run the frames at 1080p and then use AI to make the frames themselves appear to be 4K, boosting FPS and associated performance as a result. An example of this tech in action for Cyberpunk 2077 can be seen outlined below:
At the core of this functionality are Tensor cores, i.e. components of newer hardware that leverages Ampere, Volta and Turing architecture, of which facilitate huge performance boosts in both performance and precision. Before moving on to other gaming offerings it is worth touching on the noticeable change in the performance of the company’s GPUs themselves, being that they are such a fundamental component of the company’s offerings as a whole. I delved into the data of top-performing cards across various architectures, as outlined below:
Kepler GTX 680
Maxwell GTX 980
Pascal GTX 1080 TI
Turing RTX 2090 TI
Ampere 3090 TI (2022 version)
Looking at metrics including the number of Transistors, Cores and the Node Size, we can clearly see that the computational power of the hardware itself (as gauged by the number of cores and transistors) is continuing to increase, all while the node size is decreasing, a phenomenon that should continue to increase the number of transistors and thus cores can be included within a hardware unit.
In addition, these GPU’s memory bandwidth has increased almost 5-fold from 192 GB/s within the Kepler Architecture to approximately 1000 GB/s within the Ampere, phenomenal to say the least. With Nvidia’s new lovelace hardware rumored to be built on TSMC’s 5nm node size process, and have approximately 18,400 cores, processing power is likely to continue to baffle.
Moving on to the next offering, Nvidia’s GeForce Now is a cloud gaming service, allowing for customers to play a variety of games in a streaming esque-manner. In essence, Nvidia leverages more than 20 different data centres situated across the world to allow players to stream an expansive array of games (more than 1000 available at the time of writing), all at the best quality since said data centres are utilizing top-notch RTX technology.
Lastly, Nvidia offers consumers access to the Nvidia Shield TV, an android TV box that one can use to stream shows and play games using the GeForce Now capabilities, all while providing access to features such as 4K HDR entertainment and AI upscaling to increase video resolution, if desired.
Data Center
Nvidia’s Data Center business segment is a full-stack offering, consisting of both hardware and software, and is aimed at consumers that are looking to perform either AI or high performance computing (HPC) workloads. Starting from the top, metaphorically speaking, Nvidia offers customers access to a host of data-center GPUs including the H100 and H100 CNX, A100, A2, A10, A16, A30, A40, V100, RTX 6000, RTX 8000 and Test Drive, each with their own unique capabilities. Similar to their consumer GPU counterparts, i.e. the models outlined within the gaming section of this report, the hardware has improved leaps and bounds as we have transitioned from architecture to architecture over the course of the last few years.
Memory performance has also improved substantially, with the Memory Size increasing 6 fold from the Pascal Architecture to the Hopper Architecture and the associated bandwidth approximately tripling during the same time, using the 80GB Memory Size and >2000GB/s bandwidth of the H100 CNX and the 16GB memory size and 732 GB/s bandwidth of the P100 as points of comparison, respectively. In short, the ability for demanding computational tasks, model training etc. to be completed expediently is only growing with time, with these powerful pieces of technology, substantially outperforming even the company’s best consumer-facing graphics cards.
Outside of allowing for customers to build their own computing systems using the above hardware, the company also sells DGX Station 100, DGX A100, DGX H100, DGX Pod and DGX SuperPod solutions that are essentially a line of pre-built servers and workstations that run AI and deep-learning workloads for interested clients. Outside of creating these powerful hardware systems comprised of GPUs, they also come with an associated containerized software architecture that allows for deployments to be easily setup, run and managed.
Nvidia offers customers access to both HGX and EGX systems within their certified systems business units as well. In essence, these systems bring together both Nvidia’s GPU hardware and the company’s high-speed networking capabilities in order to provide customers access to a functionality for pre-built use-cases/applications such as data centres, edge computing, workstations etc.
On the more powerful end of the spectrum, Nvidia offers customers the ability to bring both efficiency and optimization into their enterprise data centre processes that were once only available to cloud service providers. Acting as the world’s first data centre infrastructure on a chip, companies can now transform traditional IT infrastructure into fully programmable data centres, offloading networking, storage and security tasks from CPUs onto the DPU, all while becoming fully AI-enabled.
A recent announcement worth paying attention to within this business segment is the company’s Grace CPU. Created leveraging Arm architecture, this processor is estimated to deliver nearly 10X the performance of the world’s fastest servers and is geared at being able to cater towards the most difficult and computationally intensive AI and HPC workloads in the world. When coupled with the company’s GPU and DPU offerings, this CPU makes Nvidia officially a 3-chip company, a prospect that their CEO Jensen Huang was excited about to say the least:
“Leading-edge AI and data science are pushing today’s computer architecture beyond its limits – processing unthinkable amounts of data. Using licensed Arm IP, NVIDIA has designed Grace as a CPU specifically for giant-scale AI and HPC. Coupled with the GPU and DPU, Grace gives us the third foundational technology for computing, and the ability to re-architect the data center to advance AI. NVIDIA is now a three-chip company.”
Nvidia also bolsters a host of different software offerings, each of which can be seen briefly outlined below:
AI Enterprise Suite: an end-to-end, cloud native suite of AI and data analytics and software tools that can enable any organization to utilize AI.
Base Command: Leverages Nvidia accelerated infrastructure, such as DGX foundry, to provide centralized control over AI training projects and enable data scientists to accelerate their AI development.
Bright Cluster Management: This piece of software offers access to rapid deployment capabilities and the ability for HPC and AI server clusters to be deployed in a host of environments ranging from the edge, to the cloud and data centres.
DGX Foundry: Created using Nvidia’s DGX SuperPod architecture, Foundry is an infrastructure solution that allows for companies to create a premium-esque AI development experience without having to create or manage the associated environment themselves.
CUDA-X: Staying surface level for a moment, CUDA is a programming language that was developed by Nvidia in order to allow for engineers to complete general purpose computing on its GPUs, a truly revolutionary happening that I encourage you to dig deeper into. CUDA-X AI and HPC libraries work in conjunction with the newest tensor core GPUs in order to accelerate both the development and deployment of applications across various domains.
Fleet Command: is a managed cloud service for AI applications that allows for one to deploy, manage and scale applications across various distributed edge systems.
Magnum IO: is a software offering that is aimed at accelerating data centre IO performance, maximizing, storage, network, multi-node and multi-GPU communications in order to create a responsive and intelligent & parallel data centre IO.
Nvidia NGC: hosts a catalog of various GPU-optimized software that allows for scientists and engineers to develop their AI solutions and includes access to things such as the aforementioned Base Command, Fleet Command etc.
Nvidia DOCA: an offering that allows for developers to program the data centre infrastructure (associated with the Bluefield DPU hardware) in order to create software-defined, cloud-native, DPU-accelerated services.
Virtualization: installed within a cloud or enterprise data server, Nvidia vGPU software creates virtual GPUs that can be shared across various virtual machines & accessed by any device.
Professional Visualization and Automotive
After spending the majority of this section outlining the main value drivers of Nvidia’s current business, I decided to encompass the two other, much smaller business segments into one discussion area. Professional Visualization and Automotive, though ambitious goals for the future, may very well be significant drivers of future success and as such are worth having a brief understanding of before moving forwards.
Starting off with the former, Professional Visualization encompasses a host of functionality that could very well have their own report, but for the sake of simplicity we will define these solutions as 3D rendering technologies that work in conjunction with the company’s GPU architecture and associated computing languages. If you are so inclined you may want to look into solutions such as Iray, Mental Ray, MDL and DesignWorks, but in short, these technologies allow for customers to create custom rendering applications within a host of different end use-cases such as architecture, entertainment, 3D design etc. Nvidia’s Omniverse activities tied to this space are arguably the most interesting, and out of all associated use-cases, are the ones I feel have the most practical utility. With the ability to create fully functional digital twins, the sky's the limit for the amount of industries that will be able to leverage this tech in order to complete processes ranging from factory-floor optimization to advance manufacturing simulations. As an example, Lockheed Martin and the US Department of Agriculture Forest Service to simulate wildfires within the Omniverse, creating digital twins of these happenings and using AI to then determine different approaches on how to best control them, a pretty cool example if you ask me.
On the Automotive side of things, Nvidia is directing efforts to gain significant exposure to this space through a series of hardware, software and infrastructure offerings. Starting off with hardware, the Nvidia DRIVE supercomputing platform allows for data from camera, radar and lidar sensors to execute autonomous driving as well as in-cabin and driver monitoring functionality. The main offerings within can be summarized as follows:
DRIVE Hyperion: production-ready platform for autonomous vehicles. This reference architecture will work in conjunction with hardware, as well as with various components of the DRIVE software stack, such as AV for autonomous driving and IX for monitoring and visualization, to conduct a host of deployment, testing and validation processes with associated vehicles
DRIVE Orin: a 254 TOPS system on a chip (SoC) that acts as the central computer for “intelligent vehicles”
DRIVE AGX Pegasus: combines two Nvidia SoCs and two Nvidia Turing GPUs to offer a piece of hardware with super-compute functionality that can be intertwined with autonomous vehicle systems such as robotaxis, if desired
DRIVE AGX Xavier: 30 TOP SoC that is capable of supporting Level 2+ (partial automation) and Level 3 a (conditional automation) autonomous driving
Drive ORIN: offering the equivalent of a data-centre on wheels, this high performance offering is geared at customers that desire 2X the abilities of the ORIN hardware.
The Nvidia software is what does the majority of the heavy lifting, turning noise into signal, so to speak, and allowing for the vehicle’s hardware to be leveraged in a way that leads to intelligent autonomous driving, supporting applications ranging from perception and localization to mapping, natural language processing and planning/control. The main software offerings can be described as follows:
DRIVE OS: Foundation of the DRIVE software stack. Acts as the operating system for on-vehicle accelerated computing.
Driveworks: the middleware functions that act on top of the drive OS to perform sensor abstraction, data recording, vehicle I/O support, deep neural network processes etc.
DRIVE AV: part of the software stack that contains the perception, mapping and planning capacities/layers, as well as access to neural networks that have been trained using real-world data.
DRIVE Chauffeur: AI-assisted platform that can handle highway and urban traffic with high degrees of safety.
DRIVE IX: open software platform that delivers interior sensing capabilities for vehicle cockpit AI applications.
DRIVE Concierge: real-time conversational AI that gives drivers access to intelligent services
MAP: software that works in conjunction with hyperion sensors to identify intersection details, traffic lights, and other road landmarks.
Lastly, the company provides access to data centre and Launchpad solutions that allow for customers to complete AI and accelerated computing processes related to the development of their autonomous vehicles. In addition, customers can leverage simulation capabilities through their OVX server and DRIVE constellation to conduct a host of different autonomous vehicle simulations. More specifically, virtual vehicle fleets can be simulated driving millions of miles in a host of different scenarios in order test the capabilities of a fleet in a manner much more expedient than the real world.
To conclude these ramblings, to say that Nvidia’s business offerings are diverse would be an understatement to say the least. Through the provision of various hardware and software solutions, the company has its hand in a plethora of industries, some with real potential to be enormous in the coming decades.
With all of this in mind, let’s delve into the company’s financials to see how their business is performing under the hood.
5.0 Financials
Within this section we will be delving into NVDA’s financials, both on an annual basis and for the company’s most recent quarterly earnings release. All dollar values mentioned below are in millions of USD.
Income Statement
Revenue
Nvidia’s top-line has grown tremendously over the course of the last few fiscal years. On an annual basis, FY’22 Revenue came in at approximately $27B, representing a 61% YoY increase and 31% CAGR from FY’17. On a quarterly basis, similar phenomena can observed, with the company’s Q1 FY’23, coming in at approximately $8.3B, the company has not exhibited signs of deceleration just yet, with top-line increasing 46% YoY and approximately 170% Yo2Y, respectively. Heading into an environment where demand destruction is likely to occur for a host of macroeconomic factors, it’ll be interesting to observe how these metrics hold up, but more on that later.
Before moving on, I think it is worth stressing just how crazy Nvidia’s growth has been at such scale, very rarely do we see a company so giant putting up YoY growth numbers like they have been:
Delving deeper, as of FY’19, the company segments their revenue into two distinct components, Graphics, which mainly consists of applications related to the company’s consumer GPU hardware (gaming, cryptocurrency etc.) and Compute & Networking, which can be thought of the company’s efforts supporting commercial and academic applications such as cloud computing, artificial intelligence etc. As you can see, the company is still evidently driven by the Graphics segment, but we are starting to see noticeable impact of Compute & Networking on business growth, both on an annual and quarterly basis:
More granular insights can be observed when the top-line is split into similar constituents as what we outlined within the technology section earlier, i.e. Gaming, Data Center, Professional Visualization, Automotive, OEM & Other.
Quite evidently, both on an annual and quarterly basis, Nvidia derives the majority of their top-line from Gaming and Data Center business segments, phenomena that I do not see changing anytime soon. Noticeably, the Data Center has been growing quite substantially and in Q1 FY’23 actually overtook gaming as the primary driver of Revenue. Given how intertwined this segment is with the biggest tech trends, ranging from cloud computing to artificial intelligence, I would wager that this continues to maintain a lead over gaming heading into the next few years.
Lastly, something to keep in mind is how much of Nvidia’s Revenue is generated from customers outside of the United States. Put plainly, the company is subject to the policy decisions of these foreign governments the geopolitical happenings cantering these regions as a whole, a bundle of tail risks that investors should continue to monitor.
Cost of Revenue
Nvidia’s Cost of Revenue consists primarily of costs associated with semiconductors, ranging from processes such as wafer fabrication to testing and assembly manufacturing support etc. In addition, this metric is comprised of costs related to license and service agreements, IP maintenance fees etc. For FY’22, COR came in at approximately $9.4B, representing a 54% increase YoY and a 27% CAGR since FY’17, exhibiting signs of some cost efficiencies relative to top-line. In Q1 FY’23, we saw a continuation of similar dynamics, with COR coming in at approximately $2.9B, representing a 41% and 166% YoY and Yo2Y increase, a slightly lower rate than quarterly revenue.
Despite having substantial presence on the hardware front, the company evidently exudes pricing power, a phenomenon attributable to their technological prowess and observable within their gross profit and associated margins:
A hardware-centric company demonstrating ever increasing GP margins in the 65%+ range is ludicrous to say the least. Given the lead the company has with respect to peers they should be able to maintain these metrics at the 60% levels, however, i would not be surprised to see slight downward pressure on the metric given macroeconomic and associated demand dynamics prevalent in the market at the moment, but only time will tell.
Lastly, it is pertinent to keep an eye on the company’s inventory provisions, a cost associated with maintaining excess quantities of products and components. With supply gluts running rampant, and outside of gauging demand deterioration through top-line growth, this metric will likely demonstrate how bad inventory buildups are on Nvidia’s side of the equation in the coming quarters. For reference, these values have been around 1% of top-line on both an annual and quarterly basis in the favourable backdrop seen over the last few years:
Operating Expenses
Nvidia’s OPEX profile has remained relatively consistent over the years. On an annual basis, Total OPEX came in at approximately $7.4B, representing a 27% YoY increase and a 27% CAGR from FY’17 to FY’22, maintaining itself in a relatively narrow range at approximately 30% of Revenue for the last 5+ fiscal years. On a quarterly basis, this value came in at approximately $3.56B, representing a 113% and 246% YoY and Yo2Y increase. The substantial increase was a result of an acquisition termination cost of approximately $1.35B occurring in February. This cost was incurred due to the process by where Nvidia would have acquired Arm from Softbank, occurring as a result of the supposed regulatory challenges surrounding the acquisition.
Delving deeper, OPEX consists of two primary constituents, R&D costs and S&M costs, each being self-explanatory in nature. The profiles of these expenses relative to top-line has remained consistent on both an annual and quarterly basis, as outlined below:
The general wellbeing of Nvidia’s future trajectory is the continual engagement in long-form (multi-year) R&D projects. In the company’s past this is what drove sustained success, with examples of Nvidia spurring technological innovation including the 1999 invention of the GPU and the introduction of the CUDA programming language in 2006, and in order to continue serving the forefront of gaming, research, cloud service and consumer internet providers, AI etc., they will need to continually direct 20%+ of top-line at new projects in order to reap success at this large of a scale.
Profitability and Margins
Despite operating and growing at a ridiculously large scale, the company has still found a way to maintain an ever improving degree of profitability, on both an Operating, Net Income and EBITDA basis. For FY’22, the aforementioned metrics were the best they’ve been in the last five fiscal years, as outlined below. On a quarterly basis for Q1 FY’23, we saw the effect of the aforementioned one-off expense, but no real indicators of significant business stressors as a result of the macro environment, a phenomenon that should be monitored closely heading into the upcoming earnings season, The company did mention in their most recent earnings call that they expected sequential increases in operating expenses to taper off after Q2, being that the front-end of the year was weighted with expenses related to the bring-up of multiple new products, so this is something to keep an eye on as well:
Balance Sheet
The structure of Nvidia’s balance sheet, as of Q1 FY’23 can be seen outlined below:
The company has a strong asset position, with approximately $30B of their $45B in total assets being relatively liquid in the current variety. Within, these primarily consist of Marketable Securities, Accounts Receivable, Cash and Equivalents and Inventories in order of descending magnitude. As previously mentioned, tracking inventories will likely be a nice gauge to see how the company is dealing with supply dynamics. Long Term Assets are primarily comprised of Goodwill, Other, PPE, Intangible and Deferred Income Tax assets in order of descending magnitude. The company’s liabilities are mainly non-liquid, with 29% of Total Liabilities being of the current variety, consisting mainly of accrued liabilities and accounts payable. Long term Liabilities on the other hand are comprised mainly of long-term debt of approximately $11B. The firm’s debt levels are completely manageable, as gauged by EBIT/Interest Expense on a FY basis, a refreshing change from 90% of the nasdaq that is in a vulnerable state with this metric less than 0, as mentioned in my last writeup.
Cash Flow Statement
The structure of Nvidia’s cash flow statements, for both FY’22 and Q1 FY’23 can be seen outlined below:
Honing in on Q1 FY’23, CFO came in at approximately $1.7B representing a 7.6% YoY decrease and a 90.4% Yo2Y increase, primarily as a result of advanced payments made on supply agreements. CFI came in as an inflow of approximately $2.8B as a result of inflows of approximately $6B related to the proceeds from the maturities of marketable securities and $1B from the sales of marketable securities, offset by an outflow of approximately $4B attributable to the purchase of marketable securities. CFF came in as an outflow of approximately $2.4B, attributable almost entirely to the repurchase of common stock. Lastly, FCF came in at $1.4B, representing a FCF margin of approximately 17% and FCF less SBC margin of approximately 7%, a refreshing happening given the fact that the company is not egregiously abusing SBC to inflate these metrics.
With this quick look under the hood, let’s now jump over to a discussion of some positive and risk-based factors alike that i’m pondering with regards to Nvidia’s story.
6.0 Discussion
One of the biggest points at the forefront of everyone’s minds is how the semiconductor industry is faring. I feel this point is worth covering here, and despite not being an expert in this field by any means, it is worth providing a high level overview of some key recent happenings. Staying surface level for a moment here, Nvidia’s place within the expansive semi universe can be seen outlined below:
During the pandemic, there was without a doubt a huge acceleration in device-related demand for semiconductors, a phenomenon that was exacerbated by supply chain disruptions, of which led to the industry-wide shortages that we are all too familiar with at this point. Although there are still backlogs that need to be worked through, with the auto industry as a prime example of this, we are starting to see this trend reverse, with the supply demand dynamics moving back towards equilibrium as gluts have started to become rampant. Let’s use one of the biggest foundries, TSMC as an example. Despite reporting the fact that their profit surged over 76% to record levels in the April-to-June quarter and revising full-year guidance upwards, they have also warned of excessive inventory buildups. An article released by Digitimes earlier this month corroborated the fact that TSMC’s 3 biggest customers have revised their orders, as outlined below:
With record-high inflation inflation crushing consumer sentiment and associated expenditures as a result, as well as the Fed looking to hike their way into an inevitable recession, I have a hard time being bullish on the outlook for semi demand at least in the short term. Delving into the specific situation surrounding Nvidia, there have been numerous stories circulating regarding the buildup of the supply of Nvidia’s GPUs. Nvidia’s GeForce RTX 40-series (Lovelace architecture) delay rumours are rampant, with the excerpt from TSMC highlighting the fact that the company attempted to cut Lovelace orders, potentially due to the RTX 3000 card buildup across a variety of inventory channels. Demand destruction in this area is a muli-faceted phenomenon, but may be related to the fact that gaming-related business avenues “over-earned” in the pandemic environment, with behaviours now mean-reverting. In addition, crypto-mining related activities with GPUs have tanked as a result of poor price action and major crypto assets such as Ethereum moving away from a proof-of-work validation approach to a proof-of-stake, which doesn’t require GPUs to be used. Looking at Ethereum Mining Revenue as an example, July has seen quite the decline, a phenomenon that may continue to occur heading into the remainder of the year.
Nvidia’s CFO stated the following regarding the effects of cryptocurrency mining on the Gaming business segment.
“The extent in which cryptocurrency mining contributed to Gaming demand is difficult for us to quantify with any reasonable degree of precision.”
I believe these factors viewed in conjunction with poor consumer sentiment and high inflation effecting the sizing of discretionary budgets, is likely to result in poor short-term outlook, despite the long-term trajectory and competitive positioning of Nvidia in gaming looking quite positive, with Nvidia displaying a 80%+ market share in the discrete GPU market at the end of the 2021.
Staying on the more positive side of things for a second, it is also worth pondering how Nvidia is competitively positioned within their other business segments outside of gaming. Clearly, the company is stationing itself at the forefront of AI. Using HPC as an example, as of June of last year, Nvidia powers more than 60% of the Top 500 list, i.e. the 500 most powerful non-distributed computer systems in the world. Automotive is also developing well, with a March 2022 article outlining the fact that the company’s DRIVE pipeline has grown to approximately $11B, testament to the fact that the company has been working with the world’s largest auto manufacturers and associated technology companies to bring these hopeful engineering ambitions into fruition. In addition, the company’s data centre business segment has clearly grown to a behemoth at a rate not many were anticipating. When viewed alongside the fact that this tech is essential for most modern cloud and consumer internet companies to function, I don’t believe that recessionary happenings will slow down the activity in these areas by a substantial margin.
Jumping over to valuation, the company like most of its big-tech constituents, has seen multiple compression galore in the first half of 2022 and beyond, as of close on friday, some of the company’s forward multiples can be read as follows:
Completing a bit of oversimplified napkin math, we can get a sense, in an imprecise manner, of how expensive the company still is. Using consensus top-line and EBIT estimates alongside the assumptions that the company’s OCF as a % of Revenue will normalize at 45%, and CAPEX will remain approximately 6% of top-line, we arrive at the following FY’25E Multiples.
Although this is an imperfect exercise, I am of the opinion the company is not expensive by any means. Unlike the big-name CPU players such as AMD and Intel, the company benefits from much higher margins and should be able to maintain a fair degree of pricing power despite supply dynamics going forwards given their technological capabilities and competitive positioning. In addition, the company benefits from the fact that they operate in one of the most valuable industries on earth, all while not having to maintain the CAPEX of other companies in the space such as TSMC, being that they outsource the manufacturing of their hardware, a phenomenon that puts them in a much less vulnerable position in my opinion. At this size however, I am concerned with the company’s ability to maintain a degree of high-paced growth going into the future. Yes they have the call options in automotive, AI and professional visualization business segments, however, their is significant tail risk in my opinion that the company simply just evolves into a slower-growing giant in the coming decade, and as such I will not be allocating to Nvidia anytime soon.
Before parting ways, if you have been watching Nvida for a while now, it is worth pondering this interesting data point that Gavin Baker shared a few weeks back regarding semiconductors viewed alongside EPS cycles. Semi companies tend to peak long before the first cut and bottom before the last cut, often being up nearly 50-100% by the time said cuts end. As such, it is worth monitoring how the company is performing price-action wise in conjunction with EPS trends going forwards.
7.0 Conclusion
Aside from being the most complex I have ever written about, Nvidia offers investors a fascinating story. Existing at the forefront of modern computing, the company is likely to continue to innovate for years to come and I will continue to monitor and update this story out of sheer intellectual curiosity. Thank you for reading my research.
If you enjoyed this research piece, please give it a like, subscribe and share!
If you want to see more frequent investing-related content, give me a follow on twitter @MT_Capital1. If you have any suggestions for how I can improve my work or a company you would like to see analyzed next let me know in the comments below!
Very nicely done! Comprehensive, well written and easy to digest. I would have enjoyed a bit more analysis around your conclusion that they risk becoming a slowing growth giant in the coming years given that they are at the forefront of some very strong secular trends . I don't necessarily disagree, but I didn't quite get the full impact of how you think that may play out.
Again, great work, keep it up!