Equity Analysis: NVDA
- Spencer
- Jun 23, 2024
- 11 min read
Remember, the purpose of these articles is not to answer questions. It is to provide you with information, and a possible template of how you can think about your own analysis.
How to read this article
All research projects at GVR are meant to be digestible by both beginners and advanced readers. In order to achieve this, no definition is given in the text body itself. Instead, all concepts and words that I believe need explaining are bolded, italicized, and subscripted[1]. At the bottom, there is a definition and concepts page. I recommend having this open on two tabs. One will be for the reading portion, and the other with the definitions open so you can quickly pan back and forth without having to scroll annoyingly.
Equity analysis article structure is split into 4 parts.
Business Focus
The purpose of the Business Focus section is to get an idea of the business model, both current and past. By understanding the trajectory of the company, we may gain insight into their strengths and weaknesses, and how they address and overcome hurdles
Financial Statement Analysis
The first portion is a cross-sectional analysis. How well are they doing when compared to industry standard and competitors? What questions should we be asking?
The second portion is time-series analysis. Where is the company trending? Are there any red flags we should be looking for?
Risk Factors
Are there any special risks associated with this industry or company? What have we come across in the annual reports that could cause us to be nervous?
Time Series Analysis
A look into their price action. How much volatility would we have to endure? Where is it looking in terms of historical price levels?
Final Discussion
What are the highlights and take aways from the article?
Business Focus
How have the stated business focus and strategies of Nvidia changed over time?
Since their first filing in 1999, Nvidia has come a long way in terms of strategy and scope. Their first filing reads the following: "We design, develop and market 3D graphics processors and related software that provide high performance interactive 3D graphics to the mainstream personal computer." At the time, they had only 4 products.
Contrast with the first paragraph of their 2024 10-k report. "NVIDIA pioneered accelerated computing to help solve the most challenging computational problems. NVIDIA is now a full-stack[1] computing infrastructure company with data-center-scale offerings that are reshaping the industry. Our full-stack includes CUDA[2]... as well as hundred of domain specific software libraries, SDKS[3], and APIS[4]... Our data center scale offerings are comprised of compute and networking solutions that can scale to tens of thousands of GPU[5] accelerated servers interconnected to function as a single giant computer."
So how did they go from point A in 1999 to point B in 2024? This section will go through the highlights and fun facts of each year. The hope is to give you an idea of the trajectory and thought process at Nvidia through the years
COMPANY SUMMARY
Nvidia reports their business in two segments
Compute and Networking
The compute and networking segment is centered around enabling large scale AI in robotics, driving, and a multitude of other implementations.
Nvidia powers 75% of the supercomputers on the global TOP500 list.
Graphics
As the name suggests, their graphics market powers graphics implementations for cinema, gaming, animation, etc.
Nvidia does not manufacture any of their products. Instead, they develop and prototype new products, and then make use of third parties to execute and manufacture their new designs. This is called a "fabless" strategy. Their supply chain is concentrated in the Asia-Pacific region. Nvidia separates their manufacturing needs into 3 segments: Wafer[6] production, memory production, and assembly, testing, and packaging. The following are their suppliers.
Wafer Production
TSMC
Headquartered: Hsinchu, Taiwan
TSMC accounted for 48 per cent of the foundry market and had 61 per cent of the world's capacity to make chips with a 16nm or more advanced process node.
Samsung
Headquartered: Suwon-si, South Korea
Memory
Micron
Headquartered: Idaho, USA
SK Hynix
Icheon-si, South Korea
Samsung
Assembly, Testing, and Packaging
Hon Hai Precision Industry
Headquartered: New Taipei City, Taiwan
Wistrong
Headquartered: New Taipei City, Taiwan
Fabrinet
Headquartered: Bangkok, Thailand
There may be others not listed in their reports as well.
Application and end users of Nvidia products are varied; however, their primary customer are not end users, for a variety of technical reasons. The main purchasers are equipment manufacturers who will either build their product around an Nvidia spec, or purchase Nvidia products to put in their products. Examples include Tesla, Microsoft, HP, Dell, as well as many popular gaming computer companies.
INDUSTRY SUMMARY
1999
Company Context
At this point, Nvidia was focused entirely on the design of 3D graphic processors for implementation into PCs, with their aim to be the leading supplier for OEM[8] and add-in board manufacturers.[9]
Even at this point, Nvidia ran a "fabless" manufacturing strategy. This means that they were purely in the research and development business. Their graphics processors were fabricated by Taiwan Semiconductor Manufacturing Company ("TSMC") and Wafertech, (who changed their name to TSMC Washington in 2023) and assembled by Amkor Technology, Siliconware Precision Industries, and ChipPAC Inc. All of the above mentioned companies are still in business today.
They did have exposure however to faulty manufacturing process. On their 10K[10], they report the following: "semiconductor companies frequently encounter difficulties in achieving acceptable product yields. When production of a new product begins, we typically pay for wafers, which may or may not have any functional product. Accordingly, we bear the financial risk until production is stabilized. Once production is stabilized we pay for functional die[11] only."
Market Context
2000
Company Context
With the release of the GeForce 256, Nvidia claims to have developed the worlds first GPU. (whether or not that's true is up for debate). Nvidia also entered into a fairly large agreement with Microsoft to develop and sell graphics chips for a "certain technology"
Market Context
2001
Company Context
In their 10k, Nvidia announced that the "certain technology" that they were developing for Microsoft, was the graphics chips to be used in the Microsoft Xbox.
Nvidia also became a key player with the leaders of the CAD industry, with cutting edge technology in the industrial design market.
Market Context
2002
Company Context
They release their GOForce Go mobile GPU family of products for use in the growing laptop market.
2003
Company Context
Make the goal to be the leading supplier of performance GPUs, for virtually all applications
They also entered into a strategic partnership with IBM to manufacture the GeForce GPUs
2004
Company Context
Begin aiding in the cell phone market and make the prediction that "future cell phones will be able to receive television programs, record digital video like a camcorder, enable video phone calls and be a portable game player. We see an exciting opportunity to leverage NVIDIA’s resources and expertise in digital media processing to offer products for the multimedia handset era."
With this in mind, they begin manufacture of WMPs. (Wireless media processors)
2006
Company Context
In April 2005, they finalized their definitive agreement with SCE to jointly develop a custom GPU to put into the PlayStation3. They make yet another prophetic statement, "We believe the synergy created by the combination of 3D graphics, HD video and the Internet will fundamentally change the way people work, learn, communicate and play. We believe that our expertise in HD graphics and system architecture positions us to help drive this transformation."
They are heavy in the fixed costs space, meaning they are a great buy when coming out of recession.
They announce their Compute Unified Device Architecture, or CUDA, which utilizes the parallel compute power of a GPU. This will become vital in the AI revolution of 2020 and beyond.
2008
Company Context
Nvidia purchases Ageia, a industry leader in gaming physics technology. Their technology was used in several games built for Playstation 3, Xbox 360, Nintendo Wii, and gaming PCs.
At this point they have 63% of the standalone desktop GPU market, As well as 63% of the standalone notebook market, according to Mercury Research.
They also launched the GeForce 9400M mGPU, which came standard in every Apple Macbook Pro and Macbook Air.
They are very bullish on handheld devices, saying:
"We believe that mobile devices like phones, music players, and portable navigation devices will increasingly become multi-function, multi-tasking, PCs. As such, we anticipate the architecture of these devices will increasingly become more consumer PC-like and be capable of delivering all the entertainment and web experiences that end users currently enjoy on a PC, but in a form-factor that fits nicely in their hands."
2009
Company Context
They enter into a lease for data center space in Santa Clara. In 2024, this will become a large part of the offering of Nvidia to power large scale DNN[15] models.
2010
Company Context
For the first time they name parallel processing and supercomputing as a key aspect and break through in their products.
Over 300 universities around the world now teach parallel programming with CUDA and many PC OEMs now offer high performance computing solutions with Tesla for use by customers around the world, including Motorola Inc., Chevron Corporation, General Electric Health Care and General Mills Inc.. Researchers use CUDA to accelerate their time-to-discovery, and popular off-the-shelf software packages are now CUDA accelerated.
2011
Company Context
State that they are investing in three major strategic areas. Visual computing, high performance computing, and mobile computing.
2012
Company Context
NVIDIA is known to millions around the world for creating the graphics chips used in personal computers, or PCs, that bring games and home movies to life. With the invention of the graphics processing unit, or GPU, we introduced the world to the power of computer graphics. Today, we reach well beyond PC graphics. Our energy-efficient processors power a broad range of products, from smart phones to supercomputers. Our mobile processors are used in cell phones, tablets and auto infotainment systems. PC gamers rely on our GPUs to enjoy visually immersive worlds. Designers use GPUs to create visual effects in movies and create everything from golf clubs to jumbo jets. Researchers utilize GPUs to push the frontiers of science with high-performance computing. NVIDIA has nearly 5,000 patents granted and pending worldwide.
Tesla has had particular success in supercomputing centers and in oil exploration; other applications include accelerating drug discovery, weather simulations and derivative price modeling.
This was also the year that AlexNet neural network, trained on Nvidia GPUs, won the ImageNet computer image recognition. According to Nvidia, this marked the "Big Bang" moment of AI.
2015
Company Context
US and Europe super computers were powered by Tesla GPU accelerators. The US Department of Energy announced that its next gen of supercomputers were based on Tesla GPU accelerators
Cars will feature a multitude of devices, driven by sophisticated software algorithms. These devices are designed to ensure our safety and the safety of those around us, enhance our comfort and enjoyment, and search and navigate. They will use the tools of deep learning to sense their environment, ultimately driving themselves.
NVIDIA has the potential to own the entire stack of technology that makes this possible, including computing vision, deep learning and natural-language processing.
2017
Company Context
Today, it also simulates human intelligence, enabling a deeper understanding of the physical world. Its parallel processing capabilities, supported by up to thousands of computing cores, are essential to running deep learning algorithms. This form of AI, in which software writes itself, enables computers to learn from data and serve as the brain of computers, robots and self-driving cars that can perceive and understand the world. GPU-powered deep learning is being rapidly adopted by thousands of enterprises to deliver services and features that would have been impossible with traditional coding.
For the first time they include extending technology leadership in AI into their business strategies
2019
Company Context
Talks about the importance of data centers, being fueled by the AI boom.
2021
Company Context
Enter talks to acquire ARM limited, valued at 40 billion dollars. This acquisition feel through, ending up with a 1.4B cost to Nvidia
Financial Statement Analysis
What can we understand about business efficiency, profitability, and solvency from their books.
Summary
Nvidia's impressive growth over the past couple years is due to their emphasis on their Compute & Networking segment, primarily driven by higher shipments of their Hopper GPU, which is built for datacenter applications and training and inference of Large Language Models, such as ChatGPT. They have also seen a 133% increase in Networking revenue due to higher shipments of InfiniBand.
Nvidia was exceptionally efficient at collecting credit sales[16] and using their assets to turn profits. During 2022-present, Nvidia were able to increase their gross profit margin 4.2x faster than their cost of revenues.
While their extreme increase in revenues the past 12 months is unusual, their profitability ratios and margins have been growing steadily over the past decade, and don't show any sign of abnormal activity.
They have no big debt obligations coming due soon that would put them in an unfavorable liquidity scenario.
Is this growth sustainable?
While this may be anecdotal, I think there may be some wisdom in it. Apple was founded in 1976. 25 years after it's inception, on Oct 23, 2001, Apple released the ipod. From that year until ~2012, it was able to double it's revenue every 2 years.
In 1990, 15 years after it's inception, Microsoft released Microsoft Office for windows. For the next decade Microsoft doubled its revenue every two years.
Nvidia was founded 30 years ago. Due to the explosion of AI, and their extreme capable Hopper GPU, could this be their "decade of growth" similar to other tech giants such as Apple and Microsoft?
Cross-Sectional
Nvidia has a multitude of products that operate in different markets. How then do we compare their performance to an industry benchmark in a meaningful way? Common comparisons are Nvidia, AMD, Huawei, and Intel. Below you'll see that only AMD is compared. Reason is that Huawei is not publicly available, and Intel has historically been a chip manufacturer. As of 2024, they announced that their foundry business will be getting it's own books, meaning that a comparison would be more meaningful.
For now however, a comparison of AMD is the only that makes sense as an entire business unit comparison. A deeper look at companies with large segments in the datacenter processing will also be informative, as is argued by the chart below.
Liquidity Ratios: Can they meet their short term obligations
Ratios | NVIDIA | AMD |
Current Ratio | 3.5x | 2.6x |
Quick ratio | 2.9x | 1.7x |
Cash Conversion | 135.8 | 142.4 |
Solvency Ratios: Can they meet their long term obligations
Ratios | NVIDIA | AMD |
|---|---|---|
Credit Rating[13] | AA- | A- |
Interest Coverage | 187x | 5.2x |
Assets-to-Liabilities | 2.76x | 5.81x |
Debt-to-Equity | 22.9% | 5.3% |
Debt-to-Capital | 18.6% | 5.1% |
Activity Ratios: How efficient are they at turning assets into revenues
Ratios | NVIDIA | AMD |
|---|---|---|
Inventory Turnover | 2.9x | 2.5x |
Receivable Turnover | 9.7x | 5.0x |
Payables Turnover | 12.6x | 4.95x |
Working Capital Turnover | 2.21x | 2.20x |
Asset Turnover | 1.3x | .3x |
Fixed Asset Turnover | 15.4x | 10.8x |
Profitability Ratios: How good are they at turning generating profits from resources
Ratios | NVIDIA | AMD |
Gross Margin | 75.28% | 50.56% |
Operating Margin | 59.84% | 2.55% |
Net Profit Margin | 53.40% | 4.89% |
Analysis
Ratios do not help us answer questions, rather they direct the asking of them. Let's focus. our attention on a few ratios that stand out.
Interest Coverage:
Why is Nvidia's Interest Coverage 36x larger than AMDs?
Receivables Turnover
Why does NVIDIA collect credit sales so much faster than AMD?
Payables Turnover
Why does NVIDIA pay their suppliers so much quicker than AMD?
Asset Turnover
How is NVIDIA generating $1.30 for every $1.00 of assets, while AMD is only getting $.30 for every $1.00?
Operating Margin
NVIDIA may be better at managing COGS, but why on earth is AMD performing so poorly, and have so many operating expenses? Is this abnormally bad on the part of AMD, or abnormally good on the part of NVIDIA?
Apart from the ratios, my curiosity is peaked as to whether or not this was simply a good year for NVIDIA, or if this is an indication of things to come. We may well answer some of these questions doing time series analysis, while others we'll have to look into the 10k reports. Lets hold off on answering them until we've done an analysis through time.
Time Series
Interest Coverage
Nvidias Interest coverage ratio appears to drop off in 2001 and 2009, which correspond to the dot com recession and the great recession. There is no surprising information given here, and it appears that this metric is fairly volatile. A look at Nvidias liability section show that volatility coming from sporadic current long term debt payments coming due, as well as varied interest expense for the same.
Receivables Turnover
Again, there may be some correlation with business cycle and overall economic health, but their ability to collect receivables is not currently out of the ordinary.
Payables Turnover
Payables of 41.3 are not out of the ordinary for Nvidia. This may merit a further look into how they get such favorable terms with their suppliers. A naive but very likely explanation is that the biggest players get the best terms.
Profitability Comparison
That spike we see on the graph is representative of 2022 to 2024 trailing twelve month period. During this time, Nvidia were able to increase their gross profit margin 4.2x faster than their cost of revenues. How were they able to achieve both increased revenues and efficiency?
On their 10K report, they say reported the following:
Data Center revenue for fiscal year 2024 was up 217%. Strong demand was driven by enterprise software and consumer internet applications, and multiple industry verticals including automotive, financial services, and healthcare. Customers across industry verticals access NVIDIA AI infrastructure both through the cloud and on-premises. Data Center compute revenue was up 244% in the fiscal year. Networking revenue was up 133% in the fiscal year.
The primary driver for the absurd 217% increase in data center revenue was their Hopper GPU, as well as their InfiniBand product. Both are used in the inference and training of large AI models.
Risk Analysis
What are the risks involved with Nvidia and it's multinational operations?
Risk 1 | Geo-Political
China
Why are tensions so high between Taiwan and China?
In 1912, the ROC (Republic of China) was founded, at which time, Taiwan was under Japanese rule. After the Japanese surrendered during WWII in 1945, the ROC took jurisdiction of Taiwan.
In 1948, a full scale civil war breaks out between the ROC and the CCP (Chinese Communist Party)
In 1949, the CCP beat the ROC, and the ROC relocate to Taiwan, bringing with them 1.2 million Chinese. The ROC is still the governing body of Taiwan.
As of writing, and according to the taiwanese government site, 95% of Taiwanese are ethnically Han Chinese.
Likely due to the number of Chinese, and the dissent by the former governing body of China, they see Taiwan as a lost child, just waiting to be re introduced into the fold.
Why was there a show of force recently by China?
an 2024, Lai Ching-te won the election to become president of Taiwan. In his inauguration speech in May, he made the following comments as reported by AP: “I hope that China will face the reality of (Taiwan)’s existence, respect the choices of the people of Taiwan, and in good faith, choose dialogue over confrontation.” Lai pledged to “neither yield nor provoke” Beijing and said he sought peace in relations with China. But he emphasized the island democracy is determined to defend itself “in the face of the many threats and attempts at infiltration from China.”
In and around that time, China began making shows of force and flying combat aircraft in Taiwan airspace.
What would an invasion of Taiwan mean for Nvidia?
For the top line of Nvidia, China represents a mid-single digit percentage of their Data Center revenue and they expect it to be the same going forward. the US government has placed embargos and regulations on selling high performance products to country groups D1, D4, and D5, which include China.
Taiwan produces the majority of semiconductor chips around the world, and are reported to make 90% of the worlds most advanced chips. An invasion would mean serious trouble, that would likely mean more for the global economy than just for chip makers. That being said, Taiwan does not own the entire supply chain of high performance semiconductor chips.
Qartr has a great article, on this which I will link here
Risk 2 | Revenue Concentration
Data
One customer, (referenced as customer A in their 10k reports) represented 13% of total revenue in the Compute and Networking Segment.
One indirect customer which primarily purchases Nvidia products through system integrators and distributors, including through customer A, is estimated to have represented approximately 19% of total revenue for fiscal year 2024. This customer did not exist in the 2023 10k report.
Revenue from sales to customers outside of the United States accounted for 56% and 69% of total revenue for fiscal years 2024 and 2023 respectively. In 2022, revenue from sales to customers outside of the United States accounted for 84% of total revenue
Interpretation of Data
Could it be that this mystery indirect customer is responsible for the large uptick in sales? Their primary driver was their Hopper series, formerly known as their Tesla line responsible for AI training and inference. There is also a large drop in foreign sales from even 2 years ago. My pure speculation is that the Nvidia could have won some sort of government contract, or a contract with a large org within the USA.
Time Series Analysis
What can we learn from the price action of their stock?
Simulation
I ran two different batches of 100,000 random simulations based on the past 4 years of historical data. The following were the rules that the 2 simulations followed as well as the results
Simulation 1
Rules
The minimum trade lasted 4 hours.
We only took long positions
The length and number of trades were randomized
no commissions or fees were calculated (essentially giving us an upside bias)
Results
Return
Median Return: 92.90%
Min Return: -24.02%
Max Return: 2298.40%
8.26% of simulations lost money
Drawdown
Median Drawdown: 20.54%
Min Drawdown: .01%
Max Drawdown: 71.98%
Simulation 2
Rules
The minimum trade lasted 4 hours
When exiting a long trade, we immediately enter a short position and vice versa
The length and number of trades were randomized
no commissions or fees were calculated (essentially giving us an upside bias)
Return
Median Return: 00.76%
Min Return: -1766.04%
Max Return: 2511.03%
49.86% of simulations lost money
Drawdown
Median Drawdown: 45.12%
Min Drawdown: 12.56%
Max Drawdown: -1766.04%
Discussion
Simulation 1: The returns were obviously heavily skewed to the right, with some fantastic returns. (the right skew is also why I chose the median as a measure of central tendency) I think what the drawdown highlights however, is the dangers of simply trading with hype. On average, what this shows is that a monkey can return an average of 92.90% over 4 years, which is awesome. The question you have to ask yourself though, is if you have the nerve to hold on even when your trade is down 21%?
Simulation 2: As expected, going short the past few years at Nvidia would have been a horrible play. I'm not sure there is too much more wisdom to glean from this, besides the fact that Nvidia stock is extremely volatile.
Overall Analysis: Whatever your strategy, even a good strategy that only buys, will undergo periods of intense drawdown. Before making an investment, it's important to define what your investment horizon is, and set appropriate stop losses. More important than all of those however, is to have a good logic based reason to be in a stock, and then exit once that reason is violated, no emotions about it.
Final Analysis
It's obvious that Nvidia has excelled throughout it's history at being innovative and accurately predicting future trends. It has undergone a period of intense growth these past couple years, largely due to a growth in their compute and networking sector.
Their revenue is semi-concentrated, and more analysis may be warranted as to why foreign sales are making up a smaller and smaller portion of Nvidia's top line, as well as who their new buyer that made up 19% of last years top line is.
Investment horizon and trade management will play a crucial in anyone investing in Nvidia in the near future. Overall, they are an exciting company with a great track record and impressive entrepreneurism.
Definitions & Explanations
Full-Stack
In software development a full-stack application refers to all of the technology that makes an application function. In the case of Nvidia, full-stack refers to the fact that they enable the entire AI workflow, from start to end.
CUDA
CUDA is a parallel computing platform and programming model that makes using a GPU for general purpose computing simple.
SDK
A collection of software tools, libraries, documentation, code samples, processes, and guides that developers use to create applications for specific platforms or systems.
API
A set of rules and protocols for building and interacting with software applications. If you can think of a user interface with buttons for humans to click and scroll around in, an API does that but for computers, not humans.
GPU
"Graphical Processing Unit." Originally designed for processing graphics with implementations in the gaming and CAD spaces. It has recently been used for the enablement of AI and deep neural network processing"
Wafer
A thin slice of semiconductor material, such as silicon, used to fabricate integrated circuits and other microdevices.
OEM
A company that produces parts and equipment that may be marketed by another manufacturer. For example, computer components made by one company but used in computers branded by another.
Add-in board
A circuit board that adds functionality to a computer system, typically inserted into a motherboard slot. Common examples include graphics cards and sound cards.
10K
A comprehensive annual report filed by a publicly traded company with the U.S. Securities and Exchange Commission (SEC). This is where the majority of the information for this report came from.
Die
A small block of semiconducting material on which a given functional circuit is fabricated. It is the basic building block of semiconductor devices, such as integrated circuits.
GigaFLOPS
A measure of a computer's performance. One GFLOP equals one billion (10^9) floating-point operations per second.
TeraFlops
A measure of computing performance equal to one trillion (10^12) floating-point operations per second. Used to quantify the power of supercomputers and GPUs.
Fp16
"Floating Point 16" A computer number format that occupies 16 bits in computer memory, used in computing and digital signal processing.
DNN
Deep Neural Network. The term describing AI models that depend on large models mimicking human neurons.
Credit Sales
Sales where the customer is allowed to pay at a later date. This
References:
Comentarios