Issue No. 1: NVIDIA

“It’s hard to do something about a problem you don’t understand.”

That’s from Jensen Huang, NVIDIA’s CEO.

I picked that line because NVIDIA isn’t a chip company, it isn’t an AI company, it’s a company that excels at understanding problems.

In this piece, I’m going to unpack how they’ve done that. In the process, I want to give you a new perspective on how to think about problems your own business can solve – a key ingredient to your internal and external narrative.

You already know, NVIDIA as the company providing the lion’s share of GPUs used in AI today. Its GPUs are the computational backbone of companies like OpenAI. The business is currently valued at over $2T. Great.

But to understand NVIDIA, we need to go back to its origins.

As we go through its history, I think you’ll see that its philosophy of “understanding problems well” was baked into its DNA very early on. It was this attitude that set it up to create – and take advantage of – the boom in AI decades later.

I broke down NVIDIA’s history into three chapters - each a good example of the power of understanding problems better than anyone else.

NVIDIA Part 1: The Problem is the CPU

NVIDIA is an old company, as far as tech companies go. And it has its origins in gaming. In 1993, I was busy playing games like MYST and Mortal Kombat. My middle school buddies and I were blown away by the graphics.

They were getting better every year because, behind the scenes, dozens of tech companies were competing in a brutal arms race to see who could come up with the latest and greatest graphics cards. You can read the history if you want (but I don’t recommend it, it’s a headache).

But graphics cards weren’t just about speed and raw power. Sure, faster chips were better, but that was just part of it. What really mattered was the entire system. That’s why graphics card manufacturers were also exploring different architectures, APIs, standards, connection ports, and so on. If they made bets on the wrong design (especially if created difficulties for software developers) it could upend the business. But for the businesses that figured out the right approach, it paid off handsomely (as you’ll see in a minute).

That’s where NVIDIA comes in.

At that time, graphics cards relied on a computer’s CPU (central processing unit) to handle some of the critical work. But CPUs weren’t designed for the unique kinds of calculations required for graphics. Soon, the CPU became the bottleneck.

While most of its competitors worked within the constraints of the CPU, NVIDIA saw things differently.

They saw the CPU as the problem itself.

NVIDIA realized that if it removed this bottleneck, it would create a step change in graphics performance. And that’s exactly what they did. NVIDIA GeForce 256 introduced this new architecture and became the first “graphics processing unit” or GPU.

That insight into the problem is a big part of why NVIDIA emerged as a winner in a very crowded and competitive space. In fact, in the decade that followed, the market for graphics cards was dominated by only two companies, NVIDIA and ATI – both of whom saw and developed this new architecture early on.

Here’s the first takeaway for you:

Winning a competitive space doesn’t mean doing the same thing as everyone else, but better. It comes from understanding the real problem and telling the right story to your team, your partners, and customers about why it needs to be solved.

NVIDIA Chapter 2: Solving Problems for Developers

NVIDIA could have stayed the course as a company focused on making better and better graphics cards. But instead, they went for something much bigger.

In the 2000s, fields like weather modeling, genomics, drug discovery, computational fluid dynamics, and machine learning were taking off. All required high-performance computing.

GPUs were already showing a lot of promise because their ability to process multiple tasks simultaneously was distinct from CPUs and offered some real advantages.

However, there was a disconnect.

GPUs might have been powerful, but developers didn’t have a good way to make use of them for applications outside of graphics.

That’s where NVIDIA’s CUDA platform came in.

CUDA (short for Compute Unified Device Architecture) is a platform and programming model specifically designed to make it easy for developers to tap into the power of NVIDIA’s GPUs. It was even based on C and C++, two common programming languages, to make adoption easy.

When NVIDIA introduced CUDA in 2006, it took GPUs from “graphics computing” to “general purpose computing” – and unlocked a huge shift in performance along the way. To see what this means (without getting into the technical details) check out the Mythbuster’s video below.

Even though CUDA is almost 20 years old, it’s still a massive driver for NVIDIA’s business. I could probably do a whole piece just on CUDA because there’s a lot I’m leaving out.

But I wanted to share this chapter because it’s a good example of looking at problems from first principles. NVIDIA didn’t just see the problem for what it was on the surface (developers need faster hardware), they saw it for what it really was (developers need a way to use fast hardware).

It’s that perspective that would play a role in NVIDIA’s next chapter.

NVIDIA Chapter 3: AI and the Problems of the Future

NVIDIA’s dominance in AI was more than a decade in the making.

In the late 2000s, NVIDIA saw that deep learning had massive potential. While they couldn’t forecast the future, they “reasoned” (Jensen’s term) their way into a perspective about what other technologies would be necessary for deep learning and AI to take off.

Here’s what Jensen and his team concluded: for AI to scale, the world wouldn’t need better computers. Instead, the world would need an entirely different type of computer. Everything from the chips to the architecture, operating systems, I/O interface, applications, and networking would have to be redesigned. Even data centers would have to be re-tooled.

That conviction led the company on an intentional path to build such a computer.

They began working on the first one in 2012, which took five years to develop. It had 35,000 parts, weighed 70 pounds, consumed 10,000 amps of power, and cost $250,000. When I listened to Jensen’s interviews, I never heard him describe their products as GPUs. Instead, he calls them “AI supercomputers” and it’s easy to see why.

NVIDIA’s bet that the world would need a new type of computer was correct.

When OpenAI needed computing power to develop the LLMs that fed ChatGPT, NVIDIA had the goods. Now, NVIDIA GPUs are essentially the only viable solution to building modern AI applications.

Remember, Jensen and his team didn’t get here because they thought about the problems of today. They got there because they thought about the problems of tomorrow – long before anyone. When he says, “We are the only company [creating GPUs for AI] because we are the only company doing it,” that’s what he’s reminding us.

Wrapping Up

NVIDIA has some of the brightest people in the world on its team, and it’s made some amazing technical achievements.

But that’s not the company’s only superpower.

NVIDIA also excels at getting clarity and conviction about the problems that need to be solved and at charging its team with finding the solution.

3 Questions You Can Discuss With Your Team

Whether your goal is to build a $2M company or a $2T one like NVIDIA, here are three questions you can use to drive discussion with your team.

Do we have conviction on what the future will be like?

NVIDIA was thinking about how the future might play out long before competitors were. It didn’t react to the market. It was proactive about creating what would be needed in the future. You don’t need to forecast every detail, but you do need to be “directionally right” as Jensen put it. Has your team gone through this exercise?

Do we deeply understand the problem we’re solving?

The ROI on understanding problems deeply is always high. That’s true whether you’re pursuing highly technical ones or problems of another sort. Understanding the problem puts you on a track to develop solutions long before anyone else, and greatly improves your odds of creating the right solution.

Can we convince others about why this problem needs to be solved?

Having conviction that a problem needs to be solved does not matter if you can’t get others to see things from your perspective. That means you must “sell” your employees, customers, investors, and partners on what this problem is, why it matters, and why you’re equipped to solved it. Without that narrative, your plans may never reach their potential.

One Thought to Leave You With

Thanks for reading.

If there’s a brand you'd like me to cover in a future issue of The Narrative Field Guide, then just reply to this email with your suggestions.

Cheers,

John Rougeux