Message Font: Serif | Sans-Serif
No. of Recommendations: 25
Here's some text at the front of this post to make it different enough for me to post it again after fixing some typos.

I’ve been a longtime user of NVIDIA’s graphics cards. As an investor I’ve sat on the sidelines since it always looked too expensive (mistake). During the recent cryptocurrency bubble, when seemingly everyone was buying up graphics cards and co-opting them to mine Bitcoin, the stock price looked ridiculous (not a mistake). The bursting of the crypto bubble as well as a recent pullback in data center spending has made NVIDIA’s stock about a third cheaper than it was at its height so I wanted to take a closer look.

Market Cap: $113.6 billion
Annual Revenue: $10.2 billion
PE Multiple: 42
Dividend Yield: 0.34%
Recent Price: $186.53
Fair Value Estimate: $152

The Business
NVIDIA was founded in Santa Clara, CA in April 1993 by Jen-Hsun (“Jensen”) Huang, Chris Malachowsky, and Curtis Priem. All three were engineers at other large companies at the time of the company’s founding. Huang was a microprocessor designer at Advanced Micro Devices and Malachowsky and Priem both worked at Sun Microsystems. The company started with $40,000 and no name. A $20 million investment from Sequoia Capital soon after helped fund NVIDIA’s first chip, the NV1, released in 1995 at a cost of $10 million. That initial chip was a failure and the company almost went bankrupt, but the company’s next chip, the RIVA 128 released in 1997, was a hit, up to four times faster than any other graphics processor.

When time for incorporation came a name needed to be decided upon. The founders were naming their various work files NV for “next version,” prompting them to review all words with the letters “NV” in them. That led them to “invidia,” the Latin word for envy. NVIDIA went public in 1999, the same year it released its first GeForce card. Curtis Priem has since left the company and Malachowsky is semi-retired, but Huang continues on as CEO to this day.

The company is split into two business segments: GPU and Tegra. Products from the GPU segment are used in PC’s, workstations, and data center servers for gaming, visualization, and AI. Tegra is the system-on-chip processor used in some gaming systems, car infotainment systems, and likely in the future, autonomous vehicles. Both technologies are based on the same underlying architecture. NVIDIA focuses these two business segments on four large markets: Gaming (53.3% fiscal year 2019 revenue), Professional Visualization (9.6% of revenue), Datacenter (25.0%), and Automotive (5.5%). OEM and IP revenue (ex. licensing deals with Intel, low end graphics chips packaged with PC’s, Tegra processors shipped with the few phones and tablets that use them) makes up the last 6.9%. NVIDIA doesn’t directly manufacture the semiconductors used in its products. Instead, the company uses a fabless manufacturing strategy, enlisting companies like Taiwan Semiconductor and Samsung Electronics to produce its chips.

GPU Segment
The GPU segment made up 87% of 2019 fiscal year revenue. GPU’s are very good at executing many simple sets of specific instructions in parallel (i.e. at the same time) as compared to a CPU, which is better at executing large, generalized, complicated instruction sets that may have to run serially (one after the other). A CPU will generally have between 4-16 large cores while a GPU will have hundreds or even thousands of small cores with limited instruction sets. The classical and eponymous use case for a GPU is graphics processing, which makes repeated use of simple mathematical computations like the translation and rotation of vertices into different coordinate systems in order to display a 3D-rendered scene on screen. Billions of these simple mathematical operations are performed to generate a single image, and many of the operations can be performed independently of the other.

NVIDIA’s GeForce GPU line is centered around consumer level gaming and VR. The more expensive Quadro line is geared toward professional graphics visualization (CAD, video editing, medical visualization). The two are based on the same core chipset but Quadro has better error correction code and floating point precision. Through different drivers and firmware GeForce emphasizes high frame rate performance while Quadro emphasizes fidelity of detail. Quadro also offers more comprehensive customer support than the GeForce line. NVIDIA’s Tesla cards are geared around data center AI work and general purpose computing and have four times the double precision performance of GeForce cards. The NVIDIA DGX is a line of servers and workstations with multiple GPU’s geared toward AI work. The GRID line is optimized to deliver GPU capabilities to data centers for cloud computing uses, making it possible to run graphics-intensive applications remotely on a server in the datacenter.

There are two major players in the GPU market. NVIDIA’s main competition is AMD’s Radeon line of GPU’s. It wasn’t always like this. Back in the late 1990’s there were about 70 GPU companies that existed but today it’s largely just NVIDIA and AMD. AMD, probably better known known as the smaller competitor to Intel’s CPU’s, entered the GPU market when it bought ATI in 2006. AMD and NVIDIA have comparable offerings on the low end and mid range, and at the low end both companies also have competition from the built-in graphics functionality of Intel’s and AMD’s CPU’s . On the high end of the GPU market NVIDIA has the edge, and NVIDIA is the reigning market share champion for discrete desktop GPU’s with 67.9% unit share vs. 32.1% for AMD. If you include CPU-integrated GPU’s and laptop graphics chips then AMD has the slight edge, though it’s not an apples-apples comparison since NVIDIA doesn’t make CPU’s. There are other players in the graphics card space like Asus, MSI, Gigabyte, and EVGA, but these companies use chipsets from NVIDIA and AMD to make their graphics cards, with the aim to usually make more affordable versions. High end gaming consoles like the Xbox and Playstation have sometimes used NVIDIA chips and sometimes used AMD chips. Currently the latest versions of both the Xbox and Playstation use AMD chips.

Market leadership in desktop PC graphics cards only has so much value going forward since PC unit growth is in secular decline. The interest around NVIDIA as a growth stock centers around other uses for these chips. If a computing task can be broken up into many simple tasks that can be done in parallel, it’s a good chance a GPU can help you with it, regardless whether the task has anything to actually do with graphics. API’s written for these GPU’s like NVIDIA’s CUDA (released in 2006) and OpenCL (used for similar purposes on AMD Radeon GPU’s), have allowed programmers to adapt GPU hardware to other tasks. Previous to these API’s, programming a GPU to do something else was a painful task. Newer uses for GPU’s include high performance computing that uses numerical computational approaches to solve large and complex problems, and neural networks used in deep learning and machine learning (ex. analyzing millions of images to match or find patterns). GPU usage for AI applications has led to their inclusion in large data centers, spurring NVIDIA’s growth, and now data centers owned by Amazon, Google, Baidu, and Tencent all use NVIDIA chips. When NVIDIA developed the CUDA API, Huang did it because he knew the GPU could be applied to other programming tasks, but Huang himself has said he’s surprised at the deep learning applications for GPU’s and didn’t foresee it, and now this is the GPU’s biggest growth market. Brute force cryptography used in cryptocurrency mining has also been a recent use case, and cryptocurrency mining caused a spike in GPU prices and shortage of supply during the height of the crypto bubble. That said, NVIDIA doesn’t look at crypto as a source of any meaningful amount of revenue going forward.

As new uses for GPU’s continue to develop, hardware manufacturers like NVIDIA are customizing GPU’s for some of these new tasks, such as adding neural network-specific hardware and corresponding software API’s to further accelerate AI tasks. According to NVIDIA, an NVIDIA GPU-accelerated machine learning cluster for data science is 1/8 the cost, 1/15 the space, and 1/18 the power of a traditional CPU-based cluster. NVIDIA is also expanding its data center footprint through its large $6.9 billion acquisition of data center networking company Mellanox, helping NVIDIA build more of an end-to-end GPU network system for clients.

Tegra Processor Segment
The Tegra segment made up 13% of 2019 revenue. Tegra is a system on chip (SoC) originally built for mobile devices like smart phones and PDA’s. An SoC includes the CPU, GPU, memory controller, and connectivity (Bluetooth, WiFi, LTE) all on a single chip. the aim here being to create a chip that takes up much less space and power than a traditional chipset. While Tegra was initially developed for traditional mobile devices like smart phones and tablets, it ultimately didn’t find as much success there as it did in gaming systems like the Nintendo Switch, automotive infotainment systems like Tesla’s large display screens and digital instrument cluster, and drones. Today there are very few phones and PDA’s that use the Tegra chip (Google Nexus 9 and HP Chromebook are a couple of exceptions, with Apple, Samsung, Qualcomm, and Arm Holdings supplying the bulk of mobile device SoC’s).

While Tegra chips are largely relegated to non-critical automotive systems now, car companies are looking at Tegra chips for autonomous vehicles, and NVIDIA’s DRIVE line of Tegra-based processors are focused on autonomous driving. NVIDIA DRIVE can perceive and understand in real-time what’s happening around the vehicle, locate itself on an HD map, and plan a safe path forward. NVIDIA reports that several hundred OEM’s have tested or are testing its DRIVE chips. NVIDIA competes with Qualcomm and Intel in car infotainment systems. In autonomous driving NVIDIA competes with Intel’s recent acquisition Mobileye, as well as Tesla, which makes its own custom AI chips.

NVIDIA makes most of its revenue in Asia, with only 12.9% of its 2019 revenue coming from the United States. This also helps contribute to a very low corporate tax rate, with NVIDIA paying an average 10.2% income tax rate over the last five years.

Return on capital is excellent and since 2017 has ranged between the mid 20% and low 30% range. Prior to 2017 ROIC ranged more in the mid-high teens. Higher margin data center sales boosted gross margins, and SG&A costs that grew at a slower rate than revenue boosted margins further. Additionally, R&D cost has been scaled back as a percentage of revenue from several years ago, now a little over 20% of revenue vs. the previous high 20% range. The low 20% of revenue range is about what Intel and AMD spend on R&D, though with Intel much bigger and AMD much smaller, the absolute amounts being spent are much different.

NVIDIA’s balance sheet is in excellent shape, with $8.5 billion in cash and equivalents vs. $2 billion in debt, all long term. Debt/EBITDA is just 0.75 and interest coverage is almost 40.

NVIDIA gets excellent Glassdoor marks, with 4.6 out of 5 stars. 95% of reviewers would recommend the company to a friend and 98% approve of CEO and co-founder Jen-Hsun Huang. Frequent pros cited are working with very smart people, good pay and benefits, work flexibility, and ability to work from home. Cons mentioned cited the high cost of living and traffic issues in the Silicon Valley area and that sometimes deadlines can be tight. There were actually very few cons listed. People who posted reviews largely seemed quite happy.

NVIDIA makes over half of its current revenue from PC gaming but the real growth area is in AI applications centered around hyperscale data center and automotive use. In my valuation I break out Games, Professional Visualization, Datacenter, Automotive, and OEM & IP separately, estimating revenue growth ramps for each to arrive at a revenue growth ramp for the entire company.

I estimate that Gaming grows at 7% for the next five years, dropping to 5% for years 6-10. This is much lower than the 32.8% five year historical compound annual growth rate for the segment, but this was artificially boosted by the cryptocurrency bubble. PC growth has been on a low single digit decline in recent years, though high end gaming PC’s and professional workstations are shielded from this a bit. For example, gaming PC growth has been healthier, growing at 7% a year from 2013 to 2018, so I’m basing this segment’s growth on that, assuming that NVIDIA maintains its market share. Market research groups such as IDC and Newzoo forecast the PC and console gaming markets growing in the low single digit and mid-high single digits respectively going forward over the next five years, with the overall gaming market growing at around 9% (mobile pushes this up but this isn’t where NVIDIA has a significant presence).

I use the same 7% and 5% growth rates for Professional Visualization. This segment, unaffected by the crypto craze (Quadro cards are more expensive so crypto miners buying cards en masse went with the cheaper GeForce cards), has had a five year historical CAGR of 7.5%.

I estimate that Datacenter growth ramps from 30.0% to 10.0% over the next nine years, corresponding to a 5 year CAGR of 26.0% and 10 year CAGR of 19.2%. The data center market itself is growing at about 10% a year but the hyperscale data center market has been and is forecast to grow in the low 20% range for the next few years. Hyperscale data centers are the very large data centers built by the likes of Apple, Google, Amazon, Facebook, Microsoft. Many of these are also the types of centers geared toward AI work and so have been a driver of NVIDIA’s growth. NVIDIA’s outsize 71.3% 5 year and 105% 3 year Datacenter CAGR’s are a result of the uptake of GPU’s into existing data centers on top of a rapidly expanding market. My goal here is to ramp NVIDIA’s outsize growth back down to the growth rate of the overall market it’s contained in, also assuming the overall market’s growth rate resolves to something more normal. It’s hard for me to see how NVIDIA can have a long run of outsize growth in this field with so many much better financed companies (Google, Intel, Microsoft, Samsung), furiously at work developing their own GPU chips. NVIDIA has a great head start though.

I estimate that Automotive growth ramps from 28% down to 8% over 10 years. NVIDIA’s 5 year historical CAGR in this segment is 45.3% and the three year is 26.0%. Although I see high growth here at the outset, it can’t be high growth forever since this is a component going into a very mature industry. For example, I forecast GM’s revenues to grow at 1% over the next ten years. NVIDIA will also likely see competition from Intel’s Mobileye, limiting growth prospects. Tesla might possibly license its technology out as well.

OEM & IP revenue I assume continues to shrink. This market has been steadily getting smaller over time and CEO Huang has said that this is a declining part of the company’s overall business and the margins are significantly below the corporate average. About 34% of OEM & IP revenue are royalty payments made to Intel from a deal put in place in 2011. This deal will eventually go away and it’s not certain it will be replaced. This market has shrunk by -13% compounded over the last five years and by -0.7% over the last three. I estimate that this segment going forward shrinks by 6% over the next five years and by 1% for years 6-10.

Summing the results of these separate growth categories gives me a 13.5% five year CAGR for the entire company and an 11.3% ten year growth rate.

I estimate that NVIDIA’s operating margin expands a bit, since Datacenter will be growing fastest and it has higher gross margins than the company’s average. NVIDIA doesn’t break out what exactly the margins are across products, but does say what product category contributed to margin expansion or compression. Good sales of the highest end graphics cards and Datacenter cards boosts margins, and good sales of Tegra processors weigh on margins. NVIDIA’s five year operating margin average is 25% but it’s clocked low 30% margins in 2018 and 2019 as sales have expanded. Currently the company’s operating margin is lower in the mid 20% range due to a lull in data center spending from customers as well as slack sales in graphics cards compared to a year ago due to the cryptocurrency bubble pop. I assume in my valuation that NVIDIA’s operating margin can crawl back up to 34% over the next several years as higher margin Datacenter takes a bigger piece of the pie. While Datacenter is only 25% of sales now, if my growth ramps are roughly right the segment should be about 50% of sales five years out.

I estimate that capex stays at about 3.6% of revenue, it’s five year average, then ramps down a bit to 3.3%. I keep depreciation at its five year average of 3.1%. This might sound like a tiny amount of capex for a firm like NVIDIA, but keep in mind that the company expenses a ton of R&D every quarter, an amount equal to around 20% of revenue. Also keeping capex costs low is NVIDIA’s fabless model; all chip fabrication costs are outsourced to other companies.

The combination of the majority of NVIDIA’s revenue being made in low tax geographies (even with the passage of the TCJA), R&D tax credits, and tax benefits due to stock-based compensation leads to a very low tax rate, and the five year average income tax rate has been just 10.2%, and 12.7% if I throw out the 0% 2019 tax year and look at the four year average. I’m going with a 12% tax rate in my valuation.

Discounting back at 9.5%, I get a per share value estimate of around $152. My valuation is available for download here:

In order to maintain high growth, the company needs to switch from PC gaming to AI applications as the majority source of revenue over the next several years. PC sales growth is in secular decline, while gaming and visualization PC’s/workstations are growing in the mid single digits. GPU’s are being used for AI applications even though they weren’t specifically designed for that: they’re used since there isn’t anything better and because NVIDIA has done well creating an API allowing users to customize what that GPU does. It’s possible that a well-funded competitor will come up with something better, designed from the ground up. Intel, for example, spends 5.7 times more on R&D than NVIDIA does in absolute dollars, and won’t like giving up its presence in data centers to NVIDIA. The wide-open growth prospects of supplying AI hardware has brought the interest of very well-funded competition. While NVIDIA is ahead now, the company has well-funded competition from Intel, Samsung, Google, and Amazon when it comes to AI chip research. Car manufacturer Tesla has a big head start over NVIDIA in practical application and integration of AI in vehicles.

NVIDIA is not only a high growth company, but a high growth company in a cyclical industry (semiconductors), and the company’s key growth driver is transitioning from an area that’s matured (GPU’s for gaming) to two others that are growing very fast (data center-based AI and automotive). Trying to estimate what kind of growth rate this will result in for the whole company is challenging and I don’t have high confidence I’m accurate with what I’ve come up with. I think I’ve erred on the conservative side. Maybe I’m too conservative in hyperscale data center growth and low 20% industry growth will continue on for more than just a few years, and might even accelerate. That said, my estimate for ten years of nearly 20% compounded growth for the NVIDIA’s Datacenter market is not exactly a signpost of conservatism.

China is a major growth driver and accounts for a fifth of NVDIA’s revenue. Trade relations are not good with China these days and NVIDIA’s importance in AI research adds a possible national security issue on top of general tariffs and trade restrictions. Recently 28 Chinese companies have been put on a list that prevents US companies selling to them due to alleged human rights violations, and the list includes some of China’s biggest AI growth companies. Meanwhile, well-financed Chinese domestic companies like Huawei and Alibaba are hard at work designing domestic chips, built for AI work from the ground up vs. NVIDIA’s repurposed GPU’s.

The semiconductor industry requires a lot of R&D and innovation to stay on top. NVIDIA spends 20% of revenue on R&D. Products have to be sold quickly to minimize obsolete inventory.

NVIDIA is a well-run company with an excellent balance sheet and return on capital, but I think it’s trading above fair value these days. It’s a stock I’ll keep tabs on in case the price drops enough for it to look interesting.


NVIDIA History:

NVIDIA GPU Cards vs. AMD GPU Cards:

Quadro vs. GeForce:

Graphics Card Market Share:

CPU vs. GPU:


Tesla Autonomous GPU’s:

NVIDIA OEM and IP Revenue:

Mobile Device GPU’s:

China and NVIDIA:

Data Center Growth:

Hyperscale Data Center Growth:

PC/Gaming PC Growth:

Gaming Market Current Stats and Future Growth Estimates:

Print the post  


What was Your Dumbest Investment?
Share it with us -- and learn from others' stories of flubs.
When Life Gives You Lemons
We all have had hardships and made poor decisions. The important thing is how we respond and grow. Read the story of a Fool who started from nothing, and looks to gain everything.
Contact Us
Contact Customer Service and other Fool departments here.
Work for Fools?
Winner of the Washingtonian great places to work, and Glassdoor #1 Company to Work For 2015! Have access to all of TMF's online and email products for FREE, and be paid for your contributions to TMF! Click the link and start your Fool career.