Nvidia: $0 + $0 = $2 Trillion???

Despite numerous failures, Nvidia stands on top of the Magnificent 7

A couple of weeks ago, we released a video about Jensen Huang, founder of Nvidia. Today, it’s easy to get carried away by the glam and hype of fancy tech jargon like semiconductors, GPUs, and AI. Heck, a 15% quarterly return for Nvidia is weak for some investors. The video covers the personal rise of Jensen along with NVIDIA’s journey from a Denny’s to the $2.2 trillion giant it is today. But before any company can be successful, it has to figure out what its secret sauce is.

In interviews, Jensen often mentions his “First Principles” approach. When facing an issue, always default to your foundational assumptions about something. In this spirit, let's transition to a time that Jensen would likely claim as the moment that would define the “First Principles” of NVIDIA itself.

After a painstaking initial 2 years, the Denny’s trio launched their first product: the NV1. Think of this as an early graphics card you could integrate into your PC setup to enable rendering of 3D environments, advanced video editing, 3D modeling, and other advanced graphical programs. The execution seemed superb! For starters, this was in line with their founding goal: making a chip that accelerates the ability of a PC to support advanced computing for gaming and 3D rendering. Secondly, months after the NV1’s launch in May 1995, support of Sega would be announced with the launch of the NV1 Diamond Edge 3D. So where did they go wrong?

To keep it short, NVIDIA simply made two bets that did not pan out. The NV1 used quadratic texture mapping, a method using quadratic functions to map texture pixels (texels) in a texture space. This was not the industry standard at the time, but NVIDIA believed it was best in representing curved surfaces best. Additionally, NVIDIA had to develop their own programming standard for individuals to use since they were unable to develop compatibility with the standards at the time, OpenGL and Glide by 3dfx Interactive.

NVIDIA was simply outmatched and spread too thinly. You see, it's important to understand the task at hand here. Not only did NVIDIA develop a revolutionary processor that used unique quadratic texture mapping, the team also had to maintain a quality API to easily enable developers to interact and create what they envisioned using this chip. Unfortunately, those who did use NV1 faced a slew of errors with their APIs and regularly complained about the difficulty of using quadrilaterals to map textures - it's just mathematically easier to deal with triangles.

One last blow was left, courtesy of Microsoft. Bill and his posse enviously watched the graphics party start and decided it was time for disruption. Soon after, they developed and launched DirectX, which became the new standard API for Windows games that conveniently used triangle primitives (over the quadratic ones that Nvidia developed). However, Microsoft’s own graphics platform push is a story of its own, so let’s get back to Nvidia.

So yes, the NV1 was a failure at large, beaten down further with the success of DirectX. Fortunately, Jensen was able to raise funding to keep on going. This inflection point eventually led to the NV3, better known as RIVA 128 - a smashing hit for NVIDIA. They went on to IPO, launch the GeForce series, and establish a path for the years to come. Now, I surely do not know what went on in Jensen’s mind at the time but if I had to take a guess - he was not done competing. Seeing Microsoft’s success, he must have quickly realized the value of controlling the platform that NVIDIA’s hardware relied on - but it was greater than that. The platform had to be as free as OpenGL but require enough effort by developers that the switching costs would be massive.

Building a Moat

Forwarding to the late 2000s, Jensen would bet the house once again. As we theorized, Jensen knew that to build a moat around NVIDIA, he’d have to control the environment developers used to interact with NVIDIA’s GPUs. Not only own the platform, but ensure that it's the best platform out there. That’s where CUDA (Compute Unified Device Architecture) came into play. After already owning and creating the hardware - the physical GPUs - CUDA was developed as an API and programming platform to build a seamless and excellent developer environment for individuals to take the GPU to the next level.

But as the song might imply, the dog days were not over. In fact, they had not even started. CUDA was a massive shift away from their core business (the gaming market) at a disastrously impeccable timing. NVIDIA announced CUDA in 2006 and heavily invested in R&D here at the expense of the business, and the other shoe dropped when the recession hit. Following their 2008 quarterly earnings miss, the stock dropped 80%. Looking back at the time, Jensen would recall being embarrassed to face the public, face his employees, and finding it hard to even get out of bed in the morning or leave the house. In a recent interview, Jensen notes that his “reaction during that time is the same as his reaction today”. In a tumultuous period for NVIDIA, the solution for Jensen was to stay consistent - wake up at the same time, carry out the same tasks, and stay consistent with what you believe.

So how’d they turn the tide? It was as simple as sticking to the “First Principles” of the business. Jensen notes that time and time again, NVIDIA succeeded when it was able to see a market and predict its value over the long term. As he says, ML and AI at the time were a “$0 Billion Dollar Market”, and it was NVIDIA’s job to create the product and create the market for it. Jensen saw anecdotal evidence of researchers and enthusiasts using CUDA for genome mapping, deep learning and early machine learning and was convinced that there will be a need for the environment and the hardware moving into the future.

The AI Boom

Nvidia is no longer a simple hardware company. Today, the recent boom in AI, proliferated by ChatGPT’s launch in late 2022 has largely been possible using Nvidia as the backbone. Google, Microsoft, Amazon, and Meta lead this charge, driving 40% of Nvidia’s revenue. Indirectly or directly, today’s AI startups (ChatGPT wrappers, GPU cloud services, and OpenAI competitors) all rely on Nvidia.

As folks flocked after AI, the landscape became quite reminiscent of the gold rush. People mindlessly searching for gold, while the real winners are those who sold the equipment, like shovels and jeans. These industries - while successful - used to be seen as unilateral or unscalable. How could a jeans company ever be a market leader? Look at TSMC today, as they are akin to Levi’s during the gold rush. As a pure play foundry, they control 61% of the entire chip-making industry. Yet they’re worth less than $650 billion. Less than a third of Nvidia’s value today.

Simply put, Nvidia is a platform, not just a product. Developers use CUDA to develop games, run machine learning models, visual effects and more. Companies run their services on H100s, A100s, and soon GB200s. Price insensitive gamers snatch up $1,000+ GeForce graphics cards. You get the point. In case you’re wondering what’s caused the atmospheric rise in Nvidia’s stock over the past 16 months, hopefully this puts it into perspective.

“We’re always 30 days away from going out of business”, as Jensen says. I hope Jensen revises that estimate soon, since NVIDIA ain’t going anywhere any time soon. That’s all for this one folks! As we develop the newsletter, I’ll try to expand the diversity of topics and angles we cover. If you enjoyed reading, please like and feel free to share with your friends. Reach out to us at logicallyanswered@substack.com if you have any insider opinions or leave a comment on our Substack page!