$NVDA valuation: The AI Engine Room
I will be honest with you – in my previous years my portfolio was tilted towards “value” stocks and rarely had any prominent “growth” stocks in it. Therefore, it will not be feasible for me to boast about being an early holder of some outperforming stocks from tech sector. Only in 2025 I initiated positions in companies like Adobe, Google and Nvidia, and my opportunity costs of missing out on past returns are huge. Do I regret about this? Not at all. My portfolio still slightly outperforms S&P500 over a long horizon and I definitely had a joy from stock selection within other sectors. Now, with a small position in Nvidia which was initiated in June 2025, I am curious to look into this company in more detail.
The Semiconductor Industry: From Growth to Maturity
The modern semiconductor industry gained momentum during the PC revolution of the 1980s, as personal computers shifted from niche tools to essential workplace instruments. Over the last four decades, semiconductors have become embedded in virtually every aspect of our daily life. But the journey was not smooth.
Semiconductor revenues, which were negligible in the early 1980s, surged through the PC boom and later the dot-com era, marking the industry’s high-growth phase. Between 2001 and 2020, growth decelerated into single digits, reflecting commoditization, price declines, and intense competition. While demand has re-accelerated over the past three years—driven by cloud, data centers, and AI—the industry overall has matured, with growth less about volume expansion and more about innovation and value capture.
Semiconductors remain a structurally profitable business. Gross and operating margins have consistently been healthy, with an uptick since 2010 reflecting strong economics in advanced chip segments (e.g., GPUs, foundries). However, the industry is inherently cyclical, as periodic dips highlight demand swings and inventory corrections. Importantly, reported margins understate true profitability due to accounting nuances: adjusting for R&D capitalization adds 2–4% to operating margins across time, underscoring the sector’s resilience.
Investor sentiment toward semiconductors has mirrored industry cycles. Market capitalizations soared in the 1990s, stagnated from 2000–2010, and surged six-fold between 2011–2020. Valuation multiples have likewise fluctuated—from over 7x revenues at the dot-com peak, to 2–3x in the 2000s, and spiking again in 2019–2021. The market today appears to recognize semiconductors as a mature, but indispensable sector with durable profitability and selective high-growth niches. Notably, most semiconductor firms operate with minimal leverage, with enterprise value closely tracking equity value.
The competitive landscape has also shifted markedly. Of the top 10 revenue leaders in 1990, only Intel and Texas Instruments remain on the 2024 list. Japanese dominance has faded, replaced by Korean and Taiwanese champions, with TSMC emerging as the industry leader. Intel has lost ground, while NVIDIA’s strategy of targeting high-value niches such as AI and graphics has driven market cap gains disproportionate to its revenue ranking.
End-market demand has also diversified: once PC-centric, semiconductors are now integral to smartphones, automobiles, gaming, and increasingly, data centers and AI infrastructure. Forecasts that once underestimated AI-driven demand illustrate how rapidly the industry’s growth vectors can shift.
Closer look at AI chips
So who's the star of the show? I am not an IT person, but I have a questioning mind, so I decided to conduct deeper research. Here is the first mind-blowing fact: the key component of a GPU chip is a microscopic semiconductor called a transistor. It is sized at one millionth of a centimeter and is so small that the human eye won’t be able to descry it. (pic transistor inside GPU). Impossible to imagine, but there are 1 billion transistors on a die with the size of a nail.
Let’s look at it from a different angle. How many calculations do you think your graphics card performs every second while running video games with incredibly realistic graphics? Maybe 100 million? Well, a 100 million calculations a second is what’s required to run Mario 64 from 1996. We need more power. Maybe 100 billion calculations a second? Well, then you would have a computer that could run Minecraft back in 2011. In order to run the most realistic video games such as Cyberpunk 2077 you need a graphics card that can perform around 36 trillion calculations a second. This is an unimaginably large number, so let’s take a second to try to conceptualize it. Imagine doing a long multiplication problem once every second (ex. 3 656 915 * 9 042 738 = 33 068 524 233 270). Now let’s say everyone on the planet does a similar type of calculation but with different numbers. To reach the equivalent computational power of this graphics card and its 36 trillion calculations a second we would need about 4,400 Earths filled with people, all working together and completing one calculation each every second. It's rather mind-boggling to think that a device can manage all these calculations,
This technology, by all parameters, is extremely valuable for society. A small rectangular plate does unimaginable “heavy lifting”, and that is the real reason why society keeps attaching such a high valuation to this stock. Of course there are other companies involved in AI supply chain, but Nvidia is the clear market share leader.
NVIDIA: The Opportunist!
Let's be honest, many analysts have been consistently wrong with estimating Nvidia’s intrinsic value in the past. Take Morningstar's estimates, for example. In December 2015, their valuation was $0.55 per share (converted price taking into account the subsequent stock split), which was approximately equal to the market price at that time.
It is important to emphasize that if you made an intrinsic value estimate and the market price grew by an average of 80% annually over the next 10 years, then you were catastrophically wrong with your estimate (although it was most likely not your fault).
There is a high probability that the intrinsic value I calculated will also be wrong. However, since the chances of a correct calculation in the absence of attempts are 0, I decided to increase it to at least a minimum positive number. Today, I share with you the results of my assessment.
Let's start with a little history.
Through hardships to the stars
Nvidia was founded in April 1993 by Jensen Huang, Chris Malachowsky, and Curtis Priem with the vision to bring advanced 3D graphics to the gaming and multimedia markets. Starting with $40,000 in initial capital and securing $20 million in venture funding, the company focused on creating graphics processing units (GPUs) to handle demanding video game graphics, which traditional CPUs struggled with. The name “Nvidia” was inspired by N.V. for “Next version” and “Invida”, the Latin word for envy. They hoped to speed up computing so much, everyone would be green with envy.
The first product Nvidia launched in 1995 was called the NV1. On paper, it seemed like a miracle. Once chip that could handle graphics, audio and even game controls bundled with SEGA’s gaming console. The stars were aligned. But then came disaster. MSFT had just launched DirectX, a new graphics standard that required games to be built from tiny triangles. Nvidia instead built its chip to use curves, and suddenly game developers had no reason to support NV1. Out of 250,000 units shipped, almost all came back unsold. Nvidia was left with warehouses full of useless chips, their Sega partnership collapsing, and just 30 days of cash left. Things got so bad that Jensen literally dimmed the office lights just to save on electricity. This was the moment most companies would die. But instead of giving up, Jensen made a radical move. He gathered his remaining engineers and told them they were going to design a new chip. But this time, they would do it in record time without even testing it on hardware. In the chip industry, that was suicide. One wrong connection, one logic gate misplaced, and the entire production run would fail, costing millions. But Nvidia had no choice. So they simulated everything in software, sent the design off to TSMC, and waited eight long weeks. When the chip finally came back, the team held their breath, powered it on..…..and boom! It worked. No bugs, no glitches, smooth, stunning graphics. That chip was Riva 128, launched in 1997.
Following this early success, NVDA made a significant breakthrough in 1999 with the release of the GeForce 256, the world’s first GPU with integrated transform and lighting features. I remember my first desktop PC with Intel Celeron 433 processor had that GPU installed.
That year, Nvidia also went public and later supplied graphics hardware for Microsoft's Xbox console.
As far as NVDA stock prices are concerned, it hasn't been a smooth ride.
In 1999, the price drawdown reached a mind-blowing 89%. Then followed a period of rapid revenue growth and S&P500 inclusion (in 2001). During 2002-2004, NVDA struggled heavily due to technical missteps, competition, and financial issues - but it wasn’t close to collapse.
In 2017, during the crypto boom, Nvidia’s gaming GPU sales surged, partly because miners were buying huge quantities. There is no publicly available information on the portion of company’s revenue attributable to crypto, but from the high stock price correlation with Bitcoin we can potentially say that its share is significant, if not more, versus gaming-related sales.
Decomposing past returns
Over the long term, the main drivers of value creation are always Growth and ROIC. Nvidia is a vivid example of this rule and in the chart below I am showing how these drivers contributed to the remarkable 80% average stock return of the last decade.
Nvidia’s median 10Y revenue growth exceeded 40%, while EPS growth was even higher @ 65%. Shareholders love it when there is a combination of high growth and improving ROIC (even more so for companies with already high ROIC!). R&D spending is also important, and Nvidia succeeds in this area too. All of the above contributed to 80% annual shareholders' return of the last decade.
They do have cards in their hands
Nowadays, the company is a market leader in graphics processing units, or GPUs, hardware, software, and networking tools needed to enable the exponentially growing market around artificial intelligence.
“Intelligence is no longer stored and retrieved. It’s continuously generated by AI factories. NVIDIA is building the global supply chain for this new essential resource.”
So what is the main driver behind Nvidia’s success story? It’s the way how GPUs handle data processing. Nvidia’s GPUs handle parallel processing workloads, using many cores to efficiently process data at the same time. In contrast, central processing units, or CPUs, such as Intel's processors for PCs and servers, or Apple’s processors for its Macs and iPhones, process the data sequentially, one calculation after another. Nvidia cores are smaller and typically do less than CPU cores, but are specially designed for running AI calculations. The wheelhouse of GPUs has been the gaming market, and Nvidia’s GPU graphics cards have long been considered best of breed. With the expansion of the crypto industry, and lately “AI revolution”, Nvidia’s chips outvie competitors and continue to dominate in 2025.
Valuation – key assumptions
1. Growth: NVIDIA is projected to maintain a dominant position in the AI chip market over the next decade, with varying estimates depending on the segment and region. In 2024, NVIDIA commanded about 85% of the data center AI chip market, with competitors like AMD holding a much smaller share. It is very likely that AI Tech titans will strive to diversify away from Nvidia in long term and competitors will not be shy, therefore I expect Nvidia’s share to gradually decline to 50-60% by 2035.
I am incorporating conservative explicit 25% growth for the first 5Y period of future projections, and 20% thereafter until terminal year 2035. Main value driver (Revenue) is estimated to be in a range of $900-950 bln by that time, which is achievable given Nvidia’s market share. The size of the global GPU market is expected to jump to almost $2 trillion by 2035. Assuming Nvidia loses ground in this market but manages to cling on to even a 50% share, it could reach $1 trillion in sales in 10 years. My assumptions are also conservative versus the historical 10Y CAGR of revenue at 39%. Given the opportunistic nature of this GPU juggernaut, it should not be impossible to reach those numbers.
Terminal growth in perpetuity is 3% reflecting long-term GDP growth for developed economies.
2. WACC: My WACC estimate resulted in 11%, which is in line with the number used by most analysts, although it can be perceived on a higher range for the most valuable company in the world. Despite all the recent achievements, it is difficult to argue that Nvidia operates in a cyclical business, and that cyclicality contributes to keeping the cost of capital higher than for the median company.
3. Relative valuation: Going into specific of the model, my calculation of value per share is an average between value per share based on DCF model, and value per share based on relative valuation (I use EBITDA multiple). EBITDA exit multiple of 18 is based on current multiple for semiconductor industry (24), discounted by 25% to reflect maturing industry in 2035. We don’t know if this expectation will hold true by 2035, but you can call it a reasonability check, which is giving me additional comfort over my workings. However, I present both numbers separately for the benefit of DCF model’s proponents.
4. Reinvestment: The input that drives reinvestment is the sales to capital ratio, and while I set NVIDIA's sales to capital ratio at 1.8, the company latest annual figure, it is possible that it can continue to reinvest at closer to industry average which is 1.15 (leading to more reinvestment). I am using company specific number, since historically Nvidia’s sales to invested capital ratio was much higher than the semiconductor sector average, highlighting its capital allocation discipline and operational excellence.
5. I incorporated a consistent operating margin of 58% which is in line with the most recent reported performance. At the same time, I do admit that price wars, rising costs, or discounts to cloud providers could compress margins.
After incorporating all assumptions and factors, I came up with a fair value of $191.
Under my base case valuation scenario, NVDA stock is trading at 12% discount to fair value. If adjusted to FV within 3 years, it will generate alpha ~ 4%.
On top of this, I am sharing sensitivity analysis where you can choose your own growth and WACC to find out intrinsic value:
5) Risks and uncertainties
1. Taiwan exposure
NVIDIA continues to rely heavily on TSMC for manufacturing its advanced AI GPUs - like Blackwell - as well as for critical packaging technologies such as CoWoS, which CEO Jensen Huang described as “irreplaceable” in terms of performance and integration. Accordingly, we cannot underestimate the geopolitical component. The risk of China invading Taiwan in 2026 is estimated by Polymarket at 17% .
Incidentally, the risks of such an invasion/annexation are much lower in 2025 – only 6% at the time of writing. As grand as it may sound, China is currently holding the entire AI industry hostage – by stoking the rhetoric of a potential “reunification” of the island. Nvidia, however, is not sitting still. Company has historically used Samsung for certain GPUs (e.g., some GeForce RTX 30 series chips). Recently, media mentioned that NVIDIA has considered sourcing some AI GPU production from Samsung to mitigate reliance on TSMC if capacity becomes constrained. On top of this, NVIDIA is actively expanding production and infrastructure in the U.S. Blackwell chips are already being produced at TSMC’s Arizona fab, and construction of AI supercomputer manufacturing facilities in Texas is in progress. Nvidia plans up to $500 billion in U.S.-based AI infrastructure output over the next few years, including using partners like Amkor and SPIL for packaging and testing. All these moves support geographic diversification and localization, but the total replacement of TSMC will be difficult to achieve. President Trump didn’t reveal how high the semiconductor tariffs will be or whether they’ll be in addition to the existing 20% tariffs on Taiwan’s imports. But he has a goal to push chip manufacturing back to US shores.
2. Overestimation of AI and data center potential.
AI topic can sound like a flashback to those experienced dot-com bubble of 1999. At this stage it seems that separating the wheat from the chaff is a challenging task, but we should keep in mind that probability of AI industry not delivering on expectations is real. A decade-long 20% growth is only feasible for Nvidia if AI chip demand and related sectors continue expanding rapidly. Also, if enterprises and cloud providers reach hardware saturation sooner than expected, NVIDIA may hit a ceiling on high-margin AI GPU sales.
3. Counterparty concentration risk. Two Nvidia customers made up 39% of revenue in its July quarter, the company revealed in a most recent financial filing, raising concerns about the concentration of it’s clientele. It means that Nvidia’s explosive growth is probably being driven by a handful of large cloud providers. Revenue can potentially decline if something happens with the largest customers, i.e. it scales down investments or diversifies away to other chip makers.
4. Increased competition: Major cloud providers are designing their own chips. The idea is simple: they can optimize their computing workloads for the software that runs on their cloud to get a performance edge and they don’t have to give Nvidia it’s very juicy profit margin on the sale of every chip. AMD, Microsoft, Google TPUs, Amazon Trainium, and other specialized AI chips could limit NVIDIA’s share of the AI infrastructure market. New AI chip designs from competitors could outperform NVIDIA GPUs (e.g., AI accelerators with better efficiency). For example, in August 2025 Alibaba announced the development of a new AI chip designed to challenge Nvidia's dominance in the AI hardware market. This move is partly in response to U.S. export restrictions that have limited Nvidia's ability to sell its most advanced AI chips, such as the H100 and Blackwell series, in China. While the new chip is still in the testing phase, its introduction marks a significant step in China's ambition to compete with global leaders in AI technology. Amazon and Microsoft currently use a mixture of it’s own chips and Nvidia’s chips to give customers multiple options. Eventually, for those cloud providers the question is how much of their computing workloads for AI is gonna be offered through Nvidia versus their own custom AI chips. And that’s the battle that’s playing out in corporate boardrooms all over the world.
5. Regulatory risks, such as export restrictions or global tech regulation could disrupt chip supply and sales. So far, semiconductor investors seem to have shrugged off the impact of tariffs during Trump’s Tariff Turmoil. But he did announce plans to impose a 100% tariff on imported semiconductors, including chips produced in Taiwan. However, companies that manufacture or commit to building semiconductor facilities in the United States would be exempt from this tariff. While the headline rate is 100%, some Taiwanese semiconductor products and major firms (such as TSMC) are currently exempt due to their US-based operations and investment commitments.
6. Macroeconomic risks such as recession or slow growth can lead to currency and interest rate swings and weaker demand.
7. Expectations treadmill: Maintaining shareholders returns for Nvidia will be a hard task for managers, because they are leading an already successful company, rather than a company with substantial room for improvement. The reason is that a company’s progress toward performance leadership attracted investors expecting more of the same, pushing up the share price. Managers then have to pull off herculean feats of real performance improvement to satisfy those expectations and continue maintaining returns.
Managers’ capacity to influence returns depends heavily on their business’s position in the cycle of shareholder expectations, from start-up to maturity. Considering Nvidia position, there is a good chance that at some point it will reach the stage where management will find it difficult to continue expansion. This phenomenon is called “growth ceiling”; professors call it The Law of Diminishing Returns. As a company scales, each additional investment or effort produces proportionally less growth. Management can find it difficult to manage a very large operations with the same efficiency, due to complexity of internal structures and bureaucracy. At that stage, management’s ability to reinvent or diversify will become crucial.
We obviously can not predict the future with any level of certainty, but we can use historical data and research to understand what to expect. If a company finds a formula or strategy that earns an attractive ROIC, there is a good chance it can sustain that attractive return over time and through changing economic, industry, and company conditions - especially in the case of industries that enjoy relatively long product life cycles. But high growth is very difficult to sustain—much more difficult than high ROIC.
Conclusion: Today, odes are heard everywhere about data centers and the revolutionary AI changes. But before we get swept away, it’s worth remembering that just because AI will change the way we operate doesn’t mean every AI stock is a winner. You can nail the big story and still lose money if you don’t think carefully about how it plays out in every special situation. Equally, avoiding forecasts because “it’s too uncertain” doesn’t make the uncertainty vanish. It just leaves a vacuum that gets filled with inflated AI premiums, hype, and plenty of scams. As a society, we don’t yet know if AI will make us better off—or worse. Every big technological shift comes with side effects no one saw coming. While there might be something like a dot-com bubble for the current AI hype cycle, at the end of the dot-com bubble we still had the Internet. And we might be in a similar situation with generative AI. With all of this in mind, I am not selling my NVDA shares. I was late in jumping in, but my current view that shares are fairly valued makes me feel fine about holding a small portion of my portfolio in this unreplicable company with wide moat.
Disclaimer: I hold NVDA shares. This is not financial advice. Do your own research.