Decoding NVIDIA: A Qualitative Analysis

NVIDIA is one of the biggest tech success stories in recent years.

History + Contextualization

NVIDIA was founded in 1993 by Jensen Huang, Chris Malachowsky, and Curtis Priem with the goal of bringing 3D Graphics to the Multimedia industry. Their breakthrough came in 1999 with the introduction of the high-performance Graphics Processing Unit (GPU). It was a revolution for media at the time, significantly improving 3D rendering and graphics performance on PCs. It’s really here that their dominance in the GPU market began. Multimedia graphics rendering took off to levels never before seen. Their GPUs were the best at making the high-volume, complex calculations needed for 3D rendering.

These same hardware capabilities originally intended for 3D rendering translated incredibly well to machine learning (ML). NVIDIA’s GPU was perfect for running the high-volume, data-intensive calculations required to train neural networks. Initially, this type of architecture was inaccessible to the general public. But, with the release of their CUDA platform in 2006, everything changed. Developers now had the ability to interface with powerful GPUs capable of running parallel computations to accomplish machine-learning tasks from their homes. However, the field was very much academic at the time (early 2000s). There was a sizable market, but large-scale applications of ML didn’t expand until 2010, with the rise of “deep learning.”

As public interest in Machine Learning and its applications grew, NVIDIA grew with it as the owner of the development stack that drove the industry. Deep Neural Networks expanded to include multiple layers as NVIDIA’s GPUs became more and more robust over time. Google created its advanced facial recognition image search. Siri was born using Natural Language Processing. Tesla began developing its autonomous vehicle technology. All of this interest in ML was built on one development stack: NVIDIA. However, it wasn’t until the Generative AI boom that NVIDIA became the giant it is today.

In 2022, OpenAI released ChatGPT, a large-scale LLM chatbot, to the public. It took the world by storm. And it was built, trained, and verified on on NVIDIA’s GPU and CUD. Following the Generative AI boom, we see NVIDIA shares take off. Since ChatGPT’s release in Late 2022, NVIDIA’s shares have almost 10x-ed in value.

Differentiation

For starters, there was (and is) nothing accessible on the market that could outperform an NVIDIA GPU at the same price. You have your AMD GPUs, but they lag behind NVIDIA on most ML and Gaming benchmarks. Google has its TPUs that completely smash NVIDIA on performance, but they’re all in-house on Google’s ecosystem and incredibly scarce. Google also never made the push to commercialize their TPUs, likely because it would require a considerable investment and effort to topple NVIDIA’s massive market share. Plus, if they made their hardware more accessible, there’s no incentive to utilize Google’s cloud offerings. Due to high variable costs and hyperspecialization to Google’s ecosystem, they were content to build up their existing platform and keep the TPUs internal.

The real deal breaker though, was CUDA. CUDA was brilliant because it was simple, accessible, and early. NVIDIA came in with the CUDA platform so early that by the time AMD or any competitor was ready to break into the ML space, it was too late. Every ML library was written in CUDA simply because there was no alternative. This early release of integrated software to interface with GPUs poised them perfectly for the ML revolution. They owned a fully integrated development stack that made every other alternative non-viable.

Business Model

NVIDIA started off as a simple chip designer, but it is much more than that today. They are a technology giant. Every new innovation in the ML sector is run on their hardware and CUDA platform. They’re working to develop their own huge data centers to run large-scale AI applications (they’ve partnered with Foxconn to build huge mega-datacenters and supercomputers), and are working with large tech companies to continually innovate.

Initially, you’d think that NVIDIA is your typical B2C GPU distributor. And that was true in 1999. NVIDIA’s large success is owed to their long-term contracts and shift to a B2B model. They service data centers, AI applications, professional visualization, autonomous vehicles, and anything mathematically intensive across the entire scope of technology. With their groundbreaking hardware and huge market share, they’ve captured almost every big tech company in a long-term, high-value contract to run their applications on their platform. With these long-term contracts and ownership of the full development stack, they’ve locked in high-margin, high-value, stable recurring revenue over the next few years. And with such high switching costs, there’s no incentive to not renew the contracts.

Surprisingly, NVIDIA isn’t vertically integrating manufacturing. However, when you look at their margins, it makes a lot of sense. By outsourcing their manufacturing to a specialized manufacturer like TSM, they can keep their focus on design, quality assurance, marketing, and customer support. They also avoid the substantial capital expenditures and operational risks associated with owning and operating manufacturing facilities. This partnership maintains NVIDIA’s high operating leverage and mitigates variable costs. And, as the most significant player in the GPU market, macroeconomically, they have the flexibility to set the prices for their products. This has led them to maintain a gross margin of ~73%, with an operating margin of 54%. The high operational leverage may potentially be of concern as earnings are directly tied to sales revenue, but by capturing long-term contracts with large-scale tech shops, they’re pretty much set for the next half-decade off of recurring revenue. They can (and probably will) continue to scale pretty effectively. Through and through, it’s a great business model.

Potential Concerns

Everyone loves vertical integration. Fundamentally, most of NVIDIA’s contractors don’t want to front high costs for external hardware. NVIDIA markup is substantial, but no one else on the market has the means/ability to produce high-performing GPUs at that large scale and that price point. But that doesn’t mean competitors aren’t trying. Almost every big tech player (Google, Meta, Amazon, etc) is working to develop their in-house GPUs. Google has its TPU, but, as I said earlier, they’re scarce and hard to generalize. They’re still using NVIDIA chips in parallel to TPUs to maintain their operational capabilities. The switching costs to migrate the entire development stack off of NVIDIA’s platform would be very high. The bottom line is that no company currently has the means and blueprint to produce a quality alternative to NVIDIA’s scale. NVIDIA is a specialized GPU designer, and these other companies are not. The GPU market’s lack of fragmentation and high entry barriers pose a substantial hurdle for new competitors to overcome.

NVIDIA is also actively at the center of anti-trust probes. The DoJ launched an investigation into their operating processes and issued a subpoena in September 2024. They’re investigating allegations that NVIDIA may have engaged in practices such as preferential supply and pricing and are scrutinizing the company’s acquisition of the Israeli AI Startup Run. NVIDIA has maintained that they “win on merit” and provide value to customers who are free to choose the best solutions for their needs. They’ve expressed their willingness to cooperate with regulatory authorities during the investigation. Regardless, it’s a strong consideration for investors as they continue to grow.

Breaking it down

Despite NVIDIA’s beautiful business model, I’m skeptical of the market ticker itself. The business has sound economics, but public opinion drives its valuation. The market keeps buying NVIDIA, and it raises the question of what price is too high. Based on discounted cash flows, their ticker is 50% above their intrinsic value. Most Bulls argue this is reflected by the high potential of the AI market, but, regardless, the discrepancy makes me skeptical. It feels too generous of a valuation given current projections. Fundamentally, I think the market has overprojected how much revenue the company could realistically capture in the next 5-10 years. I’d likely expect a regression towards the company’s intrinsic value over the next few months, but, again, because the public has such a high expectation for the potential applications, it’s possible the share price won’t drop to its true value unless we see the overall market turn to correct itself.

The real question is whether NVIDIA will continue to grow at the same rate. The business model is very sound. It’s a market leader with no apparent competition in the short run. However, this market position could be jeopardized with the development of competitor GPUs and other tech companies looking to vertically integrate. Their product is not driven by brand value, just performance, positioning, and a lack of viable alternatives. As soon as someone can mass manufacture a higher-performance GPU and platform (and doesn’t want to sell to NVIDIA) or one of their larger contractors switches to an internally designed chip, their growth will significantly slow down.

The antitrust investigation is also of concern. It may limit NVIDIA’s ability to operate and scale at the same rate they have been, which would obviously impact the company’s future. It’s clear that NVIDIA has a substantial market penetration (80% market share), but the question is whether the DoJ allows it to operate, considering this ecosystem is currently beneficial operationally for everyone.

All things considered, NVIDIA will be stable, at least over the next few chip generations. Nobody can topple that massive market penetration in the next few years, even if they have a viable product. There are significant entry barriers within the market. However, it’s unlikely we see a lot of explosive growth in the next few years. The company’s valued at 3 trillion dollars, so there’s not much more room to grow. If you look at it from a cashflow and comps standpoint, the ticker appears overvalued, but there’s enough market hype around the company and recurring revenue to where it will stay that way short term. These high expectations could backfire, though. If they aren’t meeting the massive market expectations for their product, their shares would likely plummet. Chip delays, leadership resignations, and other legislative issues are also of concern, but, over the next few years, the business model will continue to grow with the AI industry. My analysis indicated that this will be closer to its intrinsic projected value as attention to the company starts to dwindle and actual cash flows materialize at their projected rate.

The bottom line is that this is a great company, but there’s too much priced-in hype and lofty expectations around the market for shares to compound at the same degree.