NVIDIA's $3T Valuation: Absurd Or Not?
Want to quickly understand NVDA's valuation? Read this. (15 min reading time)
Much ado has been made about Nvidia’s valuation recently surpassing Apple’s, standing at a gobsmacking $3 trillion. For reference, at one count per second, it would take you nearly 100,000 years to count to 3 trillion.
But where there’s sky there’s limit, and NVDA is no exception. So the question begs: has NVDA’s valuation flown too close to the sun?
I estimated NVDA’s valuation using a Margin of Safety approach. That simply means that I took NVDA’s valuation and worked backwards to its present-day earnings to assess if the gap between the two was realistic. It’s basically just a screening-level exercise, and is far from exhaustive.
However, in doing so we can test if there is even a smidgen of a chance that the stock is undervalued. NVDA’s 100x trailing PE makes it a suitable candidate for applying such a valuation methodology.
Let’s start with some basic info about them. I’m not going to regurgitate everything as most readers will already be familiar with its recent skyrocketing growth:
The AI sector has recently been doing very well. Demand for their AI GPUs is practically infinite in the medium-term.
Fortunately for NVDA, they currently happen to be the only company which can produce these special AI GPUs at scale in the entire world. For at least another 3-5 years.
All the Big Tech AI firms like OpenAI and Google have no choice but to buy these special GPUs from NVDA (since there are so many of them), allowing the latter to raise prices to ludicrous levels.
This has boosted their recent profits into the stratosphere, and attracted lots of speculators to its stock. As a result, their valuation has recently increased to 100x PE, even surpassing Apple’s market cap of $3T.
Before I start, I would just like to mention some caveats:
I only spent two hours analyzing NVDA’s financials. It’s a screening-level exercise at best.
This is meant to be a lighthearted post based on cursory analysis. Don’t take it too seriously.
Feel free to jot down any mistakes or omissions in the comments.
All the data here is from TIKR.com. Some of it could potentially be incorrect.
This is not investment advice. In fact, it’s probably the furthest thing from investment advice. Please do your own homework.
Disclaimer: The contents of this document are NOT meant to serve as investment advice. Read our full disclaimers below.
Quick-Dive: NVDA’s Historical Performance
Let’s dive straight into it. NVDA’s business actually has a really straightforward economic model — there’s no complicated jargony stuff in their business model to trip up non-accountants. Hence, their valuation really boils down to forecasting revenues and margins.
NVDA’s Revenues
The bulk of NVDA’s recent revenue growth came from their Data Center segment, due to the AI sector’s explosive growth. This has allowed them to keep raising prices on their AI GPUs to nefarious levels — resulting in the upwards inflection of Total Revenues (yellow line with marker).
This is the same chart as the one above, except it expresses each segment’s revenue as % of Total Revenues. We can see how the Data Center segment (orange bars) is currently killing it, representing 78% of FY24 revenues.
Meanwhile, the legacy Gaming segment’s performance (green bar) has fallen markedly, representing just 17% of FY24 revenues. Recall that before all the AI hype, NVDA’s business was mainly about selling upscaled GPUs to video game enthusiasts.
I’ve also expressed Total revenue growth YoY (yellow line with marker) in the same chart for convenience.
The chart above shows how most of that revenue growth has taken place within US shores (green bars); whereas revenue growth in mainland China (dark blue bars) has been falling. To better illustrate this phenomenon, I’ve turned it into the 100% stacked chart below — we can see how the bulk of NVDA’s FY23 revenue growth has occurred in the USA (green bars):
This has some real-world implications. As most of you are aware by now, the US government has sanctioned sales of their H100 and A100 GPUs to China.
Given that the bulk of NVDA’s revenue growth has and will continue coming from the AI sector, we should expect their China revenue growth to be flatter going forward. However, this should be easily offset by continued US revenue growth.
NVDA’s Margins
If we take a look at their margins, you’ll immediately be taken by FY24’s sterling above-average performance.
FY24 Gross Margins: 73% (blue line)
FY24 Net Margins: 49% (green line)
Both are far above their LT historical averages: 63% GM, 31% NM (rightmost)
We can pretty much take these profit margins at face value since there’s not much complexity in their economic model. This demonstrates how much operating leverage exists in their business model — when revenues rise, their GP and NP rise by more.
This is great today, but it implies that the reverse is also true — when AI GPU prices inevitably fall in the future, the resulting impact on revenue growth will definitely be felt on the bottom-line (i.e. exponentially falling NM).
NVIDIA’s Conundrum
While I’ll be the first to admit that I haven’t really studied this in-depth, a cursory search reveals that NVDA sold roughly half a million H100’s and A100’s apiece in FY24. Another cursory search reveals that they tend to be sold at an average price of $40,000-$50,000.
Multiplying their sales price x volume for both GPUs, and adding them together, gets us in the ballpark of their Data Center segment revenues of $47B in FY24. Hence, these volume and price estimates should be reasonably accurate.
Considering that most of NVDA’s future revenue growth will likely come from AI, we can pretty much guesstimate their future revenue growth = AI GPU units sold x price.
This leads us to a conundrum. While everyone can agree that NVDA will continue selling an ever-increasing number of AI GPUs, we can also agree that their prices will inevitably drop over the long-term. This is a problem when their 100x PE ratio imputes the next 30 years of growth.
NVDA is only able to jack up its prices to criminal levels today because it’s the only company which can produce such AI GPUs at scale in the world. This is unlikely to remain so in 5 years time; and once other competitors start appearing to service the practically unlimited demand for such AI GPUs, prices could fall by up to half.
This puts a ceiling on NVDA’s revenue growth. While it’s true that sales volume growth will continue to grow, it’s likely to be severely offset by continuously declining prices beginning 5 years or so from now. I just don’t see how it can maintain high double-digit revenue growth beyond that — it’ll likely be significantly less once competitors start flooding the market with alternatives. Keep in mind that there’s also a limit on how much it can continue raising prices before those 5 years are up.
I’m not going to try and put hard numbers down. The point I’m trying to make is that today’s exponential revenue growth is unlikely to persist into perpetuity — it’ll last until FY30 at best. As aforementioned, that’s a problem when your stock is priced at 100x trailing PE and imputing growth for the next 30 years.
Forecasting NVDA’s Future Revenues & Margins
As aforementioned, NVDA’s valuation boils down to just two things — revenues and margins.
Let’s get margins out of the way first. As I’ve mentioned, it’s quite unlikely that their AI GPU prices will stay at today’s prices into perpetuity. My conservative approach is to assume that their normalized Net Margins will fall closer to their LT historical average of 35% over the long-term.
Forecasting normalized revenue growth for the next 30 years is no straightforward affair either — especially when your prices are expected to start collapsing from the present-day baseline only 5 years in. To address this, I’ve compartmentalized their revenue growth into two categories:
20% - 30% revenue CAGR (2025-2029): Prior to any substantial competition challenging NVDA’s price hikes.
5% - 10% revenue CAGR (2030 and beyond): Once other competitors start appearing, prices of AI GPUs should start collapsing.
The low 5% - 10% CAGR assigned to revenue growth beyond 2030 is to account for both normalizing prices & volume growth to equilibrium levels over the long-term.
If you have difficulty visualizing my description, the chart below illustrates what my Net Profit forecast would look like under a 20/5% and 30/10% revenue growth assumption respectively:
For those of you who think that 20% - 30% revenue CAGR prior to 2030 is too low, keep in mind that we’re trying to be conservative. Feel free to discount this valuation if it doesn’t align with your revenue growth expectations.
Valuation Method 1: PE ratio
To appease both the bulls and bears, I’ve opted to present ALL reasonably possible PE ratio scenarios representing all the revenue growth scenarios as aforementioned over the next 30 years1.
Revenue growth: As mentioned, I assumed both 20% and 30% revenue growth prior to 2030; and subsequently 5-10% revenue growth after 2030.
Net Margins: To forecast earnings, I assumed 35% normalized net margins for all years (inclusive of 2025-2029). It would simply add too much complexity to this forecast to model NMs gradually falling to equilibrium levels over time. Hence, I just went with the most conservative assumption.
PE34, PE44, PE54: To demonstrate valuations over different time horizons, I used FY34, FY44 and FY54 earnings to demonstrate the PE ratios in 10 years, 20 years and 30 years time respectively.
The resulting chart should give you a bird’s eye view of all plausible NVDA valuations based on different revenue growth assumptions for the next 10, 20 and 30 years respectively:
Here’s how to read the chart above, for the 5% and 10% normalized revenue growth assumptions (leftmost and rightmost) respectively:
5% normalized revenue growth: NVDA would be valued at 44x PE34, 27x PE44 and 17x PE54 in 10, 20 and 30 years time respectively.
10% normalized revenue growth: At the end of the curve, the most optimistic revenue growth scenario based on the above assumptions would result in 35x PE34, 13x PE44 and 5x PE54 in 10, 20 and 30 years time respectively.
Based on 20% revenue growth prior to 2030 and current market cap of $3T.
Okay now let’s try another one based on 30% revenue growth prior to 2030:
5% normalized revenue growth: NVDA would be valued at 29x PE, 18x PE44 and 11x PE54 respectively.
10% normalized revenue growth: NVDA would be valued at 23x PE34, 9x PE44 and 3x PE54 respectively.
Based on 30% revenue growth prior to 2030 and current market cap of $3T.
While that green line (i.e. PE54) may imply that NVDA is undervalued today, keep in mind that not everyone is comfortable with waiting for 30 years to breakeven on their investment (i.e. PE ratio).
The vast majority of people would do a 10 year revenue projection on their DCF model, and then just use the PV of annuity formula to estimate all revenues beyond that. Such a valuation would fall somewhere between the middle of both the blue and orange lines.
At the midpoint assumption of 7.5% normalized revenue growth after 2030, that would imply something in the ballpark of:
18-40x PE for the first scenario (20% revenue growth before 2030;
12-27x PE for the second scenario (30% revenue growth before 2030).
Taking the midpoint of both ranges would give us roughly 25x PE.
Keep in mind that the further out into the future you go, the lower the normalized revenue growth assumption should be to reflect future competition putting increased pressure on AI GPU prices.
It’s worth remembering that for all the hype around AI GPUs today, these are ultimately commoditized products. For instance, it shouldn’t be too difficult for Google or another similar Tech giant to replicate them in 10 years time.
And that assumes everything goes according to plan. What if something changes in the wider tech landscape? 20 years is a veeeerrry long time — and even AI could be disrupted by then.
Once you assign a reasonable discount rate to account for the risk of the unknown, you can see how quickly their valuation scenarios could potentially collapse.
Valuation Method 2: Cumulative FCF
There’s actually another easier way to understand their valuations. If you imagine NVDA as a private business, how would you determine what the fair price to pay to acquire it would be? It’d simply be their cumulative FCF right? (i.e. add up their FCF for each following year until the cumulative FCF hits their current market cap of $3T)
This tells us how long it would take to recoup your initial investment via FCF, which can be taken as a rough proxy for future dividends. It basically just amounts to calculating the PE ratio in a roundabout fashion.
The good news is that both their OCF/NP (blue line) and FCF/NP (purple line) average close to 100% over time. This implies that we can simply take their FCF at face value.
And since their LT average FCF is almost identical to their LT average NP (103%), we can simply use NP as a proxy for FCF.
In that case, we shall simply replicate the same forecasting model as the one in the previous section. Except this time, rather than estimating a PE ratio, we can cumulatively add up all their NP of future years until it sums up to their current market cap of $3T. Then we see how many years it takes — i.e. a proxy for PE ratio.
Once again, if we take the midpoint assumption of 7.5% normalized revenue growth, we can see that it takes roughly 25 years to breakeven based on cumulative NP — implying an equivalent 25x PE ratio.
That would reconcile exactly with the valuation that we arrived at using the earlier method.
At the very least, I’d say that NVDA is already pretty fairly priced, even in the best-case scenario where everything goes according to plan.
Conclusion
Keep in mind that:
This is a screening-level exercise, and far from exhaustive.
At most, you should use this as a stepping stone for determining your own valuation. At worst, some of my assumptions could be totally incorrect.
We also have not considered risk at all.
If you add a standard 30% margin of safety, the same assumptions would lead to a valuation of 33x normalized PE instead (25 x 1.3).
However, it surprises me that even at a face-ripping $3T market cap, it remains theoretically possible to make a case for NVDA being fairly valued today. I honestly wasn’t expecting that at all.
Check out our previous stock reports:
Disclaimers:
Disclaimer: This document does not in any way constitute an offer or solicitation of an offer to buy or sell any investment, security, or commodity discussed herein or of any of the authors. To the best of the authors’ abilities and beliefs, all information contained herein is accurate and reliable. The authors may hold or be short any shares or derivative positions in any company discussed in this document at any time, and may benefit from any change in the valuation of any other companies, securities, or commodities discussed in this document. The content of this document is not intended to constitute individual investment advice, and are merely the personal views of the author which may be subject to change without notice. This is not a recommendation to buy or sell stocks, and readers are advised to consult with their financial advisor before taking any action pertaining to the contents of this document. The information contained in this document may include, or incorporate by reference, forward-looking statements, which would include any statements that are not statements of historical fact. Any or all forward-looking assumptions, expectations, projections, intentions or beliefs about future events may turn out to be wrong. These forward-looking statements can be affected by inaccurate assumptions or by known or unknown risks, uncertainties and other factors, most of which are beyond the authors’ control. Investors should conduct independent due diligence, with assistance from professional financial, legal and tax experts, on all securities, companies, and commodities discussed in this document and develop a stand-alone judgment of the relevant markets prior to making any investment decision.
I didn’t model revenue growth higher than 10% normalized revenue growth, as I just don’t see that happening as a persistent concern over the next 30 years.
I know it's just surface level, but at least you can see that it's possible here. I'm gonna attempt to do my own surface-level valuation of NVDA pretty soon, and I'll try to remember to report back here. Thanks for helping get that ball rolling!
Great piece. I like your conclusion "However, it surprises me that even at a face-ripping $3T market cap, it remains theoretically possible to make a case for NVDA being fairly valued today. I honestly wasn’t expecting that at all."
I wrote a similar article (https://oninvesting.substack.com/p/the-price-is-right) although a lot less complex and came to a similar conclusion to yours. Although probably way too high, it's not as crazy as it seems.