The AI Giant
If for more than a decade, the attention of tech investors has been on “Big Tech” (Microsoft (MSFT -0.24%), Google (GOOG +0.38%), Facebook (META -0.53%), etc.), the last few years have seen a marked shift toward hardware over software. The first sign was the spectacular rise of Tesla (TSLA -1.36%) from a niche cult-like stock to one of the largest companies in the world.
But there would be one company sitting at the border between software and hardware that would capture as good, if not stronger, returns: NVIDIA (NVDA +2.91%).
Now mostly viewed as an AI company with sudden success, NVIDIA has actually patiently built its unique technology and market position over 20-30 years. This might give it a strong position to stay a dominant actor in the world of tech for the years to come.
NVIDIA’s Path To Success
CPU vs GPU
For a long time, NVIDIA was a successful but niche computer hardware company specializing in producing graphic cards or graphics processing units (GPUs). At the time, GPUs were seen as an important computing hardware element but secondary to the all-important central processing unit (CPU).
CPUs are designed to perform very quick computations that require doing one after the other, making them great at complex calculations.
In contrast, GPUs are less powerful but designed to perform many parallel calculations simultaneously, making them better at handling large amounts of data.
During this period from the 1990s to the 2010s, CPU producers like Intel (INTC -0.9%) reigned over the industry, while high-quality GPUs were mostly only used by gamers and graphic designers for high-end PC.
Building A GPU Business
Early on, NVIDIA founder Jensen Huang and his co-founders reasoned that the pace of computing would outstrip CPU capacity. Jensen was instrumental in developing the first GPUs for Sun Microsystems, today Oracle (ORCL -0.58%).
He would then become one of the co-founders of NVIDIA in 1993, embracing the PC revolution in the early 1990s.
“We thought, you know, maybe 3D graphics would be the thing that’d be really cool. And for the very first time, you have a platform that could both be a computer and used for, you know, whatever you want to use it for. You could also use it to play games. And, we just need to go build a chip that makes it possible to play games.
None of us had even seen a PC before. So we had to go buy a PC. We bought a Gateway 2000. Nobody even knows how to program Windows or DOS. Nobody even seen DOS. And so we had to tear it apart, start learning about the industry.”
Jensen Huang, in an interview with Sequoia
It’s funny to think that, in retrospect, gaming was not a very “serious” market at the time compared to more lucrative and larger enterprise-focused business models. The first cards were not a commercial success. Their 2nd generation GPU was better but turned suddenly obsolete when the market turned toward Microsoft’s DirectX architecture for videogames.
Ultimately, it took NVIDIA six years and three product lines to find product-market fit, with many near-death events for the company.
Success would come with the Riva 128: in its first four months, it sold 1 million units. It would be followed by a long line of successful graphic card designs, including the GeForce series, to this day the dominant player in the market alongside AMD’s (AMD +0.63%) Radeon.
Source: UBuy
CUDA & Crypto
In 2006, now a well-established GPU leader, NVIDIA released CUDA, a general-purpose programming interface for NVIDIA’s GPUs, opening the door for other uses than gaming. This was done because some researchers were already using GPUs to perform calculations instead of the usual supercomputers.
Source: NVIDIA
“Researchers realized that by buying this gaming card called GeForce, you add it to your computer, you essentially have a personal supercomputer. Molecular dynamics, seismic processing, CT reconstruction, image processing—a whole bunch of different things.”
Jensen Huang, in an interview with Sequoia
This wider adoption of GPUs, and more specifically NVIDIA hardware, created a positive feedback loop based on network effects: the more uses, the more end users and programmers familiar with it, the more sales, the more R&D budget, the more acceleration in computing speed, the more uses, etc.
Source: NVIDIA
Today, the installed base includes hundreds of millions of CUDA GPUs.
Source: NVIDIA
Not only would this prove very useful to researchers, but a new technology would make great use of GPU parallel computing: blockchain and cryptos.
Crypto Boom
Now getting a little sidelined by AI enthusiasm, crypto was the first large-scale application of GPU beyond gaming and scientific research. Many blockchains and crypto projects require a lot of computing power. Quickly, NVIDIA GPUs became the central hardware for performing these calculations.
This created a boom in NVIDIA sales, and the company stock started to rise in unison with the forming crypto boom, with the stock price increasing more than 10x.
NVIDIA Corporation (NVDA +2.91%)
The stock price action from cryptos lost some steam in 2022 before markets realized that NVIDIA had been building up a remarkable AI strategy for many years.
AI
Neural Networks
From the early 2010s, researchers had started to deploy GPUs to study neural networks. These are a type of computing method that differs from usual programming and was awarded 2 different Nobel Prizes in 2024, in Physics and Medicine.
Neural networks are the technical basis for what is commonly referred to as “AI” today.
In 2009, one of my students at the time, Ian Goodfellow, who was my undergrad, helped me build a GPU server in his dorm room. And that server wound up being what we used for our first deep-learning experiments to train neural networks.
We started to see 10x or even 100x speedups training neural networks on GPUs because we could do a thousand or 10,000 things in parallel, rather than one step after another.
Andrew Ng – DeepLearning.AI founder & managing general partner of AI Funds, in an interview with Sequoia
This was before AlexNet, the first breakthrough in computer image recognition in 2012, and years before AlphaGo.
Pivoting NVIDIA To AI
NVIDIA realized AI’s potential early, ways before anybody, out of specialized researchers, cared about neural networks.
This was, at the time, a risky move into an unproven, barely existing sector, or as Jensen Huang put it:
“We’re investing in zero-billion dollar markets.”
In 2016 & 2017, NVIDIA released the Pascal and Volta architectures, respectively, the first GPU-based AI accelerator, while Volta introduced the Tensor Cores, which accelerated deep learning tasks by up to 12 times.
It was a wholesale pivot in this new direction. When we pivoted the ship in that direction, we sought out every single AI researcher on the planet.
And our platform being useful to them was the positive feedback that we were getting at the time. Which is the reason why I’m friends with, you know, all of the world’s great AI researchers.
They were all helpful in providing the early indications of future success along the way for me and, you gotta make a big deal out of those small wins.
Jensen Huang, in an interview with Sequoia
This would prefigure the building up of AI computing infrastructure, emerging massively into the public consciousness in 2023, with the release of popular LLMs (Large Language Models) like Chat GPT.
But this was actually built over the slow and often forgotten development of ever more powerful AI-dedicated GPUs by NVIDIA since 2016.
Source: NVIDIA
Another remarkable thing about the evolution of AI computing power is that it follows an exponential law instead of the more linear Moore’s Law for CPU. This is because not only is the GPU hardware getting better, but the required processing power has decreased over radical improvement in how neural networks are trained.
In addition, more available data makes training more efficient, giving researchers many angles to work in parallel to boost performances.
This has led to a radical decrease in energy consumed to train the same GPT model over time, 350x less in 8 years, and an even more extreme reduction in energy required to make a request to these LLMs.
Source: NVIDIA
NVIDIA Partnerships
NVIDIA has from its inception been a company deeply connected within the industry. Instead of a vertically integrated company, it seeks to establish deep ties with the best, while staying razor-focused on its own competitive advantages.
For example, NVIDIA is a so-called “fabless” hardware manufacturer, focusing on design and concepts, leaving to world-leading semiconductor “fab” like TSMC (TSM +0.32%) to produce its GPUs.
By not developing its own LLMs or AI system, NVIDIA is also a trusted partner for virtually all “Big Tech” and AI startups, which see it as an essential partner rather than a potential competitor. In turn, this gives NVIDIA the scale of sales to keep reinvesting in R&D and stay on top of the game from a technology standpoint.
This has proven to be the right choice, with NVIDIA the largest beneficiary of the most impressive capital expenditure (capex) spending spree in the history of the tech industry.
AI capex is expected to reach as much as $200B in 2025, on top of an ever-growing cumulated capex by the largest tech companies in the world since 2016.
Source: Sherwood
Financials
NVIDIA’s growth just from 2023 to 2024 has been incredible for a company that size:
- Revenues are up 126%, from $27B to $60B.
- Operating income tripled (311%) from $9B to $37.1B
- Gross Margin went up from 59.2% to 73.8%
Overall, the company is richly valued, but not even that much due to its earning growth. Still, with a P/E ratio above 60, and a dividend yield of just 0.03%, investors buying NVIDIA are assuming a lot of future growth to justify the current stock price.
Source: NVIDIA
Future Of NVIDIA
Sustainable Growth?
NVIDIA’s triple-digit growth rate has been astonishing and reflected in the company’s stock price. Of course, every good thing comes to an end one day, and investors are becoming concerned that this might be sooner rather than later.
The same concerns were already loud when NVIDIA sales were booming from crypto sales or in the early stages of the AI boom, so pessimism is not necessarily a sound investing strategy.
In an interview on the BG2Pod podcast, Huang explained that the world needs to update up to $1T worth of datacenter and computing to incorporate and adapt to AI. And that so far only $150B has been spent of that total.
So, according to him, there is still plenty of space for NVIDIA to keep growing sales, even if it is only due to existing computing needs. That’s before even more applications for AI became mainstream, such as self-driving cars.
Such concerns about total demand also ignore that, ultimately, all industries will likely deploy AI at multiple levels in one way or another, including sectors like healthcare that represent a double-digit percentage of GDP.
Source: NVIDIA
Blackwell
In March 2024, NVIDIA released the Blackwell platform, “enabling organizations everywhere to build and run real-time generative AI on trillion-parameter large language models at up to 25x less cost and energy consumption than its predecessor.”.
Source: NVIDIA
This is a very important step, as energy consumption is quickly becoming one of the main concerns of AI-focused companies, as illustrated by the recent Microsoft deal to re-open a whole nuclear power plant and use all its power output for the next 20 years at a pre-agreed price.
In-House Designs
One risk for NVIDIA is that while it is a key partner to the world’s largest companies, it is also a very expensive and profitable one (70% gross margin). So when companies with the size and skill set of Alphabet/Google are spending hundreds of billions of dollars on AI chips, they are tempted to do it in-house.
And this is not just hypothetical, with for example Tesla having developed its own hardware by hiring top designers from NVIDIA competitor AMD. Up to 2019, Tesla was using the NVIDIA Drive PX 2 AI computing platform instead. As Tesla is seemingly getting really close to actually commercializing robotaxi, this could become a massive missed sale for NVIDIA.
At the same time, Tesla’s case might be more of an exception to the rule, with Tesla and Elon Musk’s other companies, like SpaceX, notorious for always looking for more vertical integration and a stronger level of control over its hardware.
Companies less experienced in hardware or more software and/or marketing-focused, like Facebook or Microsoft, will likely be fine relying on the finest and latest NVIDIA technology.
In addition, many AI models are currently built and coded with the assumption that they will run on NVIDIA architectures, and the AI programmers are experienced with NVIDIA’s hardware, which are both valuable business moats for the company.
AI Market Risks
The AI market as a whole may be a larger risk over which NVIDIA’s excellent management has less control. It is booming for now. However, there is a growing concern that the AI applications released have failed to transform into massive new revenues as the iPhone did for Apple back in the day.
This is likely just a sign that the technology is still finding its mark and developing its market.
But would this situation persist too long, and we could be at risk of a situation like in the late 1990s, where the predictions about the importance of the PC and Internet were right, but the timing was a little too optimistic, leading to the dot-com bubble popping.
For sure, Jensen Huang signing an autograph on a woman’s breast in June 2024 is somewhat a surprising sign, and maybe a bit concerning for investors worried about a potential financial mania around AI.
Financial history is not necessarily repeating, but investors will want to properly analyze this risk for NVIDIA and look at potential parallels with telecom & Internet hardware manufacturer Sun Microsystems (Jensen Huang’s first employer) in 2000.
At 10 times revenues, to give you a 10-year payback, I have to pay you 100% of revenues for 10 straight years in dividends. That assumes I can get that by my shareholders. That assumes I have zero cost of goods sold, which is very hard for a computer company. That assumes zero expenses, which is really hard with 39,000 employees. (…)
Now, having done that, would any of you like to buy my stock at $64? Do you realize how ridiculous those basic assumptions are? You don’t need any transparency. You don’t need any footnotes. What were you thinking?
Scott McNealy – then CEO of Sun Microsystems
For reference, NVIDIA’s current P/S ratio is 35.
Source: YChart
Conclusion
NVIDIA is a company built on taking the correct calculated risks several times in a row at the right time, from PC graphic cards to CUDA release for new applications to embracing neural networks early. This has made its founder, Jensen Huang, something like a rockstar in the semiconductors and IT industry.
The company’s recent performance has stunned the market and created massive enthusiasm for the stock of the like only Tesla can claim in recent years. This creates a massive opportunity, as many early investors in Tesla know, having faced almost a decade of naysayers expecting the company and its stock to fail “any minute now.”
This also creates some risks, as the AI boom has yet to generate the sort of revenues justifying the current capex and might experience a downturn before becoming a fully established economic sector.
