Latest / Deep Values: Value Investing Deep Dives / [Q3 2025] NVIDIA Financial and Valuation Analysis
Transcript
- 0:00Welcome back to The Deep Dive, the show where we cut through
- 0:02the noise to get to the core knowledge of the world's most
- 0:05fascinating companies. Either.
- 0:07For years, the story around NVIDIA, you know, ticker NVDA,
- 0:12was all tangled up in this AI hype cycle.
- 0:15It was this soaring valuation built on, well, futuristic
- 0:18promises. Right, it was all about what
- 0:20could happen. Exactly.
- 0:23But what happens when the hype doesn't just show up?
- 0:26What happens when it becomes this undeniable, concrete
- 0:30financial reality that seems to defy the fundamental laws of
- 0:33business? And that's our challenge today,
- 0:35isn't it? We're looking at a company that
- 0:37has entered a period of, I mean, you have to call it hypergrowth.
- 0:40It's so massive, so intense, that it really challenges this
- 0:43idea of the law of large numbers.
- 0:46Explain that for a second. Sure.
- 0:48It's just the basic principle that, you know, growth has to
- 0:50slow down as a company's revenue base gets bigger.
- 0:53It's just harder to AA billion dollars than a million.
- 0:56So when you're already making 10s of billions, 1/4 doubling,
- 1:00or even just keeping 60% growth going, it just shouldn't be
- 1:04possible. It shouldn't be, and yet here we
- 1:07are. Nvidia's doing it.
- 1:09So our mission today is to go deep.
- 1:11We've got these comprehensive financial research reports that
- 1:14cover two really distinct pivotal periods, Q3 fiscal year
- 1:182025 and Q3 fiscal year 2026. We're not just going to look at
- 1:23the final numbers. We want to track almost quarter
- 1:26by quarter how this company completely transformed its
- 1:29business model, how it managed these unprecedented operational
- 1:33challenges and you know critically how the CEO, Jensen
- 1:36Huang, plans to sustain this pace.
- 1:38With things like the new Blackwell architecture and this
- 1:41whole concept of sovereign AI. Exactly.
- 1:44And our sources are, I mean they're incredibly detailed.
- 1:47We're talking 10 years of historical metrics, deep
- 1:49operational insights and these really forward-looking strategic
- 1:53assessments on everything from like supply chain constraints to
- 1:56future competition. So we're not just.
- 1:58Chasing momentum here. No, absolutely not.
- 2:00We're here to rigorously figure out if this is just a
- 2:02historically great stock that got bid up or if it's truly
- 2:05great business with a, you know, an unassailable economic Moat
- 2:09and a real margin of safety for the years ahead.
- 2:11And before we dive in, we have to give our crucial reminder.
- 2:16All the analysis we're about to present is based on publicly
- 2:19available financial sources. It is strictly for educational
- 2:23purposes. Yes, we're here to synthesize
- 2:25knowledge. We are not offering investment
- 2:28advice. You absolutely have to conduct
- 2:30your own independent due diligence.
- 2:31Precisely and look analyzing a company is only one part of the
- 2:35equation. You have to understand the
- 2:37broader market context right the the overall temperature of the
- 2:41economy. That's a great point.
- 2:42So if you want a non emotional, data-driven tool to track that
- 2:46total market froth, we'd really encourage you to search for and
- 2:50download the iOS App Buffett Indicator from the Apple App
- 2:53Store. A very useful tool.
- 2:55It is, and for those of you who want to join our full deep dive
- 2:58into these more advanced topics or get access to more strategic
- 3:01content, you're always welcome to join our Deep Value Club
- 3:05membership. OK.
- 3:06Let's unpack this growth story. I want to start with just the
- 3:08sheer velocity of change. I mean, if you look at the
- 3:11financial separated by just one year, 4 quarters, the
- 3:14acceleration is, well, it's breathtaking.
- 3:16It really is. Let's start with Q3 fiscal year
- 3:192025. At the time, everyone, and I
- 3:22mean everyone, thought that was the absolute peak.
- 3:24That's right. That quarter, which wrapped up
- 3:27late in 2024, I just set all these historical records,
- 3:30revenue hit $35.1 billion. And that wasn't, you know, just
- 3:36a little bit of growth. That was a phenomenal 94% year
- 3:40over year increase. And it was driven almost
- 3:42entirely by the massive, massive demand for their Hopper H100 GPU
- 3:47platform. That was really the moment the
- 3:49AI super cycle went mainstream. OK.
- 3:51So 94% growth. Now Fast forward just four
- 3:53quarters to Q3 fiscal year 2026. This is where that law of large
- 3:58numbers we talked about was supposed to slam on the brakes
- 4:00instead. Instead, it it actually
- 4:03accelerated In Q3 of 26, revenue soared even further to an
- 4:07astounding $57 billion. 57 billion in a single.
- 4:12Quarter in one quarter. Now that's a 62% year over year
- 4:15jump, which is remarkable enough on that scale.
- 4:17But the number that really tells the story is the 22% sequential
- 4:21growth. Meaning from the quarter just
- 4:22before it? Right.
- 4:23It means their growth actually speed up even coming off the
- 4:27enormous $46.7 billion base of Q2.
- 4:31I mean, to get your head around the scale of this, that single
- 4:34$57 billion quarter is more than the company's entire annual
- 4:39revenue from just fiscal year 2023.
- 4:43Let's pause on that. The 22% sequential jump, that
- 4:46kind of growth of a 40 plus billion dollar base isn't just a
- 4:50financial marvel, it's an operational 10.
- 4:52Absolutely. You're talking about managing
- 4:54logistics, securing components, manufacturing, and then shipping
- 4:58an additional $10 billion worth of product in three months.
- 5:02What does that kind of velocity tell you about their supply
- 5:04chain? It says they're absolute master
- 5:07class operators. I mean, for a while there and
- 5:09video wasn't constrained by demand at all.
- 5:11They could sell everything they can make.
- 5:12The problem was manufacturing capacity at the very bleeding
- 5:15edge. Specifically, the constraints
- 5:17were around Kawa's packaging. And just remind us what that is.
- 5:20It stands for Chip on Wafer on Substrate.
- 5:23It's basically the really advanced way they stack and
- 5:26connect multiple chips together, like the main GPU and the Super
- 5:30fast memory right next to it. That and the HBM 3 E high
- 5:34bandwidth memory itself were the bottlenecks.
- 5:36So these are the technical choke points of the entire AI era.
- 5:39They are. So the fact that sequential
- 5:41growth jumped from 16% in Q2 to 25% in Q3, that is the clearest
- 5:47signal you can get that those bottlenecks were finally
- 5:50resolving. They could open the floodgates.
- 5:52They could crank the spigots wide open, and they did it just
- 5:55as they were starting the transition to their next
- 5:57generation platform, Blackwell. It just confirms that the whole
- 6:01story has shifted from a company with potential to a company with
- 6:04frankly undisputed and overwhelming dominance.
- 6:08And that dominance is really concentrated in one place now,
- 6:11the data center segment. This is the whole story.
- 6:13It is the business, absolutely. The data center segment isn't
- 6:16just the biggest part of NVIDIA, it is NVIDIA.
- 6:18Now if you go back to Q3 of 25, data center revenue was over 30
- 6:23billion, which is about 85% of the total.
- 6:26And within that, the compute GPU sales were 27.6 billion.
- 6:31That was up 132% year over year. And by Q3 of 26, that
- 6:37concentration just got even more intense.
- 6:40It did. Data center revenue shot up to
- 6:42$51.2 billion. It now makes up nearly 90% of
- 6:47the company's total revenue, 90%.
- 6:49The insight here is really crucial.
- 6:52NVIDIA is no longer a gaming company that just happens to
- 6:55have a cool AI side business. It is, as the research reports
- 6:59put it, the sovereign foundry of the AI age.
- 7:03What did that mean? Sovereign Foundry.
- 7:05It means they manufacture the essential foundational
- 7:08infrastructure that every nation, every major company,
- 7:11every researcher needs if they want to participate in this
- 7:14revolution. It's like they're the only one
- 7:15selling the shovels in a global gold rush.
- 7:17So they have the scale, but this is where it usually gets tricky.
- 7:21Maintaining profitability while you're scaling complex hardware
- 7:24this fast is usually where the wheels come off.
- 7:26So let's talk about margins and operating leverage, because
- 7:29somehow they are managing this with the financial efficiency
- 7:32you'd expect from a pure software company.
- 7:34This to me is the most stunning financial metric of all.
- 7:38Despite transitioning to the new Blackwell systems, which are
- 7:41more complex, use more expensive HBM 3E memory, you know,
- 7:45inherently higher cost materials, NVIDIA somehow
- 7:48managed to maintain and even expand their gross margins.
- 7:52OK, So what were the numbers? They were at about 72% gap gross
- 7:56margin in Q3 of 25 and that actually increased to 73.4% in
- 8:01Q3 of 26. Expanding margins while you're
- 8:05ramping up a brand new higher cost product line, I mean that
- 8:10just screams pricing power, doesn't?
- 8:11It it's phenomenal pricing power.
- 8:13It means they're not just passing along those higher
- 8:15memory and packaging costs, they're capturing even more
- 8:18value on top of it. But the operating leverage,
- 8:21that's an even better story. OK, define operating leverage
- 8:23for us. It's just how fast your
- 8:25operating income grows compared to your revenue.
- 8:27If your profits grow faster than your sales, you have great
- 8:30leverage. And in Q3 of 26?
- 8:32Well, while revenue grew by that 22% sequentially, their
- 8:36operating expenses, OpEx, you know, R&D, sales, admin costs
- 8:41grew by only 8%. This massive difference is pure
- 8:45leverage. It drove their operating income
- 8:48to a staggering $36 billion. That translates to an operating
- 8:53margin of over 63%. 63% for a company selling physical goods
- 8:57that's that's unheard of. It's the characteristic
- 9:00footprint of a software monopoly.
- 9:03It's where you're fixed costs, like all the money they spent on
- 9:06R&D for CEDA over the years, are spread across this exponentially
- 9:10larger base of paying customers. So it's the closest thing you
- 9:13can get to a software as a service financial model, but in
- 9:16the hardware space. Indeed, it just proves that the
- 9:19vast majority of their cost structure is fixed, so every
- 9:22single extra dollar of revenue they bring in, minus the direct
- 9:25cost of the chip, drops almost straight to the bottom line.
- 9:27We can't let the data center story completely over shadow
- 9:30everything else, though, because while they're now, what, just 10
- 9:34of revenue, the nondc segments are actually really important.
- 9:38There are these highrofit cash cows that are strategically
- 9:42vital. That is a great point.
- 9:43I mean, take gaming, it's their legacy business, right, But it's
- 9:47still delivered a really robust $4.3 billion in Q3 of 26.
- 9:53Which is a 30% year over year increase.
- 9:55So that's still a healthy business.
- 9:56It's a huge business. 4.3 billion is a number most tech
- 9:59companies would kill for, and for NVIDIA it's just the side
- 10:02gig. And then you have automotive.
- 10:05It's still the smallest segment, 406,000,000, but it's surged
- 10:08123% year over year. Their Drive and Orange systems
- 10:13are becoming the standard brain for advanced autonomous driving
- 10:17and AI cockpits. So you have these two other
- 10:19businesses, one that generates billions and another that's
- 10:22growing in the triple digits and they're basically funding the
- 10:24core R&D. Exactly.
- 10:27They are funding that core $15 billion annual run rate R&D
- 10:32budget that's needed to maintain the data center lead.
- 10:35It gives them a financial resilience that their
- 10:37competitors just can't match. OK.
- 10:39Let's talk about the Moat, because if the hardware
- 10:42eventually becomes a commodity, this whole high flying growth
- 10:45story has to end. And CEO Jensen Huang has been
- 10:49very clear about this strategic transition from just being a
- 10:53component vendor, you know, just selling a graphics card to being
- 10:56a platform provider. What does that really mean and
- 10:59where does the real lock in lie? The Moat is really layered and
- 11:02you're right, the hardware performance is only the most
- 11:05visible layer. Yes, they have staggering market
- 11:08share. Something like 92% in discrete
- 11:11GPUs and over 80% in AI accelerators, but the real
- 11:14fortress that's built out of software.
- 11:17We are of course talking about CUDA.
- 11:19CUDA, It stands for Compute Unified Device Architecture and
- 11:22you have to remember they developed this way back in 2006.
- 11:25This was years before the current AI boom was even a
- 11:28whisper. So it's the programming
- 11:29platform, the API. It's kind of like the operating
- 11:32system for their GPU's. That's a perfect way to put it.
- 11:35It's what makes their hardware intelligent and programmable.
- 11:39And why is that so powerful? Why does it create such high
- 11:41switching costs? Because it became the industry
- 11:44standard. There are millions of developers
- 11:46around the world, the people who are actually writing the code
- 11:49that drives the AI revolution, and they are all trained on
- 11:53CUDA. Every major machine learning
- 11:55framework, whether it's Tensorflow or Pytorch, is
- 11:58optimized for it. So if you're a company and
- 12:00you've spent years and millions of dollars developing your AI
- 12:04models on that stack. Right.
- 12:05Migrating that whole code base, retraining all your developers,
- 12:08reoptimizing your models for a competitor's chip, it is a
- 12:12massive, massive undertaking. It's expensive and it's risky.
- 12:16So it's like having a proprietary operating system
- 12:18that runs 90% of the world's most critical applications, even
- 12:22if a competitor comes out with a cheaper or maybe a slightly
- 12:24faster physical box. The cost of rewriting your
- 12:27entire software foundation makes that competitors offering
- 12:30functionally irrelevant for most customers.
- 12:32It's the definition of a walled garden.
- 12:34It is, but it's a garden that's so fruitful that everyone is
- 12:37willingly moving inside. The whole investment thesis
- 12:40really relies on the fact that this locking creates inertia and
- 12:44that insulates NVIDIA from competitors who are only focused
- 12:46on hardware. And they're just making that
- 12:48Moat deeper with the next generation.
- 12:51Blackwell. The reports call this an
- 12:52architectural revolution, not just a speed bump.
- 12:55What's the key difference here? The key shift is that Blackwell,
- 12:59especially the GB 200 platform, is all about rack scale
- 13:02computing. Rack scale.
- 13:03Yeah, they're moving away from selling you individual chips and
- 13:07towards selling you these fully integrated super computing
- 13:10systems. Take the GB 200 NVL 72 system.
- 13:14It connects 72 Blackwell GPU's and 36 of their Grace CPU's into
- 13:18one massive cohesive unit that basically functions as a single
- 13:22computer. So instead of customers having
- 13:24to piece together thousands of components themselves, Nvidia's
- 13:27just delivering a pre integrated, pre optimized AI
- 13:30factory. Precisely, they're controlling
- 13:33more of the total system value and they're removing all that
- 13:36integration complexity for the customer.
- 13:38And it plays right into their supply chain advantages,
- 13:41especially with that COAS packaging we talked about, which
- 13:44is essential to stitch all those components together for maximum
- 13:47performance. Let's focus on one crucial
- 13:50metric the reports highlight. Blackwell offers a 10 times
- 13:53throughput per MW improvement for inference compared to
- 13:57Hopper. Yes.
- 13:58Why has this per MW metric suddenly become more important
- 14:02than just raw clock speed or FLOP?
- 14:04S because it's the critical operational necessity of our
- 14:08time, data centers are running up against a really severe
- 14:11physical constraint power, right?
- 14:14Hyper scalers are building these centers so large that they're
- 14:17literally hitting the capacity limits of the local electrical
- 14:20grid. And inside the data centers, the
- 14:22racks themselves are constrained by power delivery and,
- 14:24crucially, heat dissipation. So they're either running out of
- 14:27electricity or they're generating so much heat they
- 14:29can't cool the building efficiently.
- 14:31Exactly. Running a huge AI cluster
- 14:34demands megawatts of power, and all that power creates an
- 14:37enormous amount of heat. So if a new chip gives you 10
- 14:40times the performance for the exact same amount of
- 14:42electricity, it solves 2 huge problems at once.
- 14:46It lowers your electric bill. Dramatically, and it defers the
- 14:49need for these incredibly expensive and complex liquid
- 14:52cooling systems or trying to secure these multi year
- 14:56contracts with utility companies.
- 14:58Blackwell basically solves the power crisis for hyperscalers,
- 15:02and that strategic advantage is massive.
- 15:04That makes the hardware so much stickier than just being the
- 15:06fastest. It's about total cost of
- 15:08ownership and the ability to scale your operations.
- 15:11And that leads to the final element of the Moat, their
- 15:14control over the supply chain. I mean, Nvidia's scale gives
- 15:17them this unparalleled logistical.
- 15:19Power, we mentioned the Cowess packaging capacity was tight.
- 15:22It is the absolute bottleneck in the entire AI supply chain and
- 15:26because NVIDIA is this high volume leader in both gaming and
- 15:29data center, their scale gives them capacity assurance.
- 15:32They get preferential treatment at the leading foundries,
- 15:35specifically TSMC. So when these cutting bring edge
- 15:38manufacturing processes are constrained globally, only a
- 15:41handful of customers like Apple and NVIDIA get guaranteed access
- 15:45at the front of the line. Which means any competitor is
- 15:49not only chasing a performance benchmark that's constantly
- 15:52moving, but they're also stuck in line behind NVIDIA just for
- 15:56the ability to physically produce their chips.
- 15:58It's a huge barrier to entry. It is, and it's all sustained by
- 16:01their aggressive R&D spend, which is running at about $15
- 16:05billion a year. They're setting the pace of
- 16:08innovation, ensuring they have this multi year lead.
- 16:11That sustained investment is the ultimate testament to their
- 16:14business quality. All right, let's shift to how
- 16:16they're diversifying their demand, because if you're an
- 16:19investor, the biggest fear over the last couple of years was
- 16:21customer concentration. Right, that the whole business
- 16:24relied on the spending of the Magnificent 7 hyperscalers.
- 16:27Exactly, and NVIDIA is actively cultivating these two new
- 16:31pillars of demand to deal with this.
- 16:33The first one is called Sovereign AI.
- 16:35Sovereign AI is a really strategic, almost geopolitical
- 16:38concept. Nations all over the globe have
- 16:40suddenly recognized that AI infrastructure is now the same
- 16:43thing as national security and economic independence.
- 16:46They don't want their culturally specific language models, their
- 16:49critical medical data, or their defense systems trained and
- 16:53housed on public cloud infrastructure owned by, you
- 16:57know, foreign companies. Exactly.
- 16:59So this is demand that's driven not by corporate ad revenue, but
- 17:02by government budgets and strategic policy.
- 17:05It's a totally different, much steadier class of Customer.
- 17:09And the reports have an example. Yeah, the RIKEN deployment in
- 17:11Japan. RIKEN is this major scientific
- 17:14Research Institute. They focus on climate science,
- 17:17industrial research and they announced a huge deployment
- 17:20built on the GB 200 platform. We're talking 1600 Blackwell GP
- 17:24US. It's a massive multi year
- 17:26strategic so. The significance here is that it
- 17:29reduces that concentration risk and this government driven
- 17:32demand is way less cyclical than corporate spending.
- 17:36It introduces a lot of resilience.
- 17:38Government spending on critical infrastructure tends to be
- 17:40really sticky. It's driven by multi year
- 17:43political cycles, not quarterly interest rate changes or, you
- 17:47know, a temporary cut in an ad budget.
- 17:49OK. So that's pillar number one.
- 17:50The second pillar is the compounding nature of AI itself,
- 17:54this virtuous cycle driven by the explosion of inference.
- 17:58This is the long term exponential growth story.
- 18:02The virtuous cycle idea is that advanced AI models are being
- 18:05used to create even more advanced AI models, which then
- 18:08require progressively more compute power.
- 18:11And we're shifting from the training phase, the big upfront
- 18:14cost to build the models to the inference phase.
- 18:16The operational consumption phase and inference is where the
- 18:19growth really compounds, because every single time a user makes a
- 18:22query, every time an AI takes an action, it burns compute time.
- 18:25Exponentially so, right? Yes.
- 18:27We're moving beyond simple, you know, ChatGPT question answer
- 18:31models. We're moving into what's called
- 18:32a gentic AI. Autonomous software agents.
- 18:35Right agents that can reason, plan, execute multi step
- 18:39workflows. Think of an AI that could manage
- 18:41your entire supply chain or design a new protein from
- 18:44scratch. These agents are complex,
- 18:47they're resource intensive, and they just reinforce the
- 18:50stickiness of the CUDA ecosystem.
- 18:52And the numbers are huge. The industry estimates that for
- 18:55every dollar you spend on initial training, you'll
- 18:57probably generate three to five times that in revenue from
- 19:00inference over the models life and we are just at the beginning
- 19:04of that inference explosion. We have to turn to the risks
- 19:06now. We have to starting with
- 19:08geopolitical decoupling, the China story.
- 19:11This was a massive headwind they had to navigate.
- 19:13It was decisive after the US put in all those export controls on
- 19:17high end ships. NVIDIA was forced to create
- 19:19these, well these compliant stripped down ships like the H20
- 19:23to try and keep some market access.
- 19:24And what are the Q 326 reports say about how that's going?
- 19:27It revealed that these compliant ships have seen zero share in
- 19:30the Chinese market in recent quarters.
- 19:32Zero, And that's largely due to rising domestic competition,
- 19:36especially from Huawei's Ascend ships, and just all the
- 19:39regulatory ambiguity making big purchases too risky for Chinese
- 19:43companies. So the implication is clear.
- 19:45NVIDIA has effectively lost what was historically 20 to 25% of
- 19:49its entire data center market. We should have absolutely
- 19:52cratered their growth figures. It should have, but here's the
- 19:54critical silver lining. This is the strongest possible
- 19:58de risking event you could imagine.
- 20:01NVIDIA achieved 66% year over year data center growth despite
- 20:05completely losing that massive China market.
- 20:07Wow. So the demand from the US, from
- 20:09Europe, from allied nations, it's so voracious that it
- 20:14completely absorbed the loss of a multi billion dollar market
- 20:17and still drove phenomenal growth on.
- 20:19Top that is the core take away for risk management.
- 20:21Future U.S. sanctions can't really hurt a revenue stream
- 20:24that's already gone to 0. The stock is in a sense now de
- 20:27risked from future geopolitical hits related to China.
- 20:29It just proves the extraordinary uncapped appetite for AI
- 20:33infrastructure everywhere else. OK, let's talk about the other
- 20:36two big risks, cyclicality and competition.
- 20:40First, the air pocket theory. The air pocket theory comes from
- 20:44looking at historical infrastructure booms.
- 20:46The risk is basically that once the hyperscalers are 50 billion
- 20:50a quarter customers finished building their first massive
- 20:53training clusters, they might just pause, take a breath,
- 20:57right? Pause capital expenditures for a
- 20:59year or two to digest that new capacity and that would create a
- 21:03temporary but very severe demand Cliff for NVIDIA before that
- 21:07inference explosion we talked about fully ramps up to
- 21:10compensate. It's a very real cyclical.
- 21:12Threat and the second, more existential threat, The
- 21:15customers becoming the competitors, the ASIC threat.
- 21:19This is the big long term strategic threat.
- 21:21You've got companies like Google with their TPU, Amazon with
- 21:24Inferentia and Tranium, Microsoft with Maya.
- 21:26They're all designing custom chips, basics, application
- 21:29specific integrated circuits. Chips that are optimized for one
- 21:32thing and one thing only. They're specific AI workloads.
- 21:35Exactly, and over the long term, if their workloads stabilize,
- 21:40these custom chips can offer superior performance per dollar
- 21:44compared to buying a general purpose NVIDIA GPU.
- 21:47It's the ultimate pair Paradox's success, right?
- 21:49You innovate so well that you motivate your biggest clients to
- 21:52spend billions trying to build their own alternative.
- 21:56What's Nvidia's counter argument?
- 21:57It's really twofold. First, as we've said, the CUE
- 22:01software mode makes shifting to a new hardware architecture
- 22:04really complex and costly. But second, and this is probably
- 22:08more important, algorithmic volatility.
- 22:11Meaning the AI field is still changing really fast.
- 22:13Incredibly fast. An ASIC that's perfectly
- 22:16optimized today for, say, a transformer architecture might
- 22:19become functionally obsolete if the whole industry pivots to a
- 22:22fundamentally new model architecture next year.
- 22:25The GPU, because its general purpose and programmable through
- 22:28CUA, is the safe, flexible bet in a field that's defined by
- 22:32radical, rapid change. So the GPU is the Swiss army
- 22:34knife, while the ASIC is the highly specialized but
- 22:37inflexible scalp. That's a perfect analogy.
- 22:40If AI algorithm stabilized for a long time, the ASIC probably
- 22:44wins on efficiency, but if they keep evolving rapidly, the
- 22:47flexible GPU wins. NVIDIA is betting everything on
- 22:51continued evolution. And finally, we just have to
- 22:53reiterate that supply concentration risk, the Taiwan
- 22:57situation. Yeah, this is a low probability,
- 23:00but extremely high impact risk. NVIDIA is fabulous.
- 23:04They design the chips, but they don't manufacture them.
- 23:06They rely almost entirely on TSMC for that advanced wafer
- 23:09fabrication and critically, the Kawas packaging.
- 23:13So any geopolitical conflict in the Taiwan Strait?
- 23:16It represents an existential threat to their ability to
- 23:18deliver product. Now, TSMC is trying to diversify
- 23:22its footprint, but the vast majority of leading edge
- 23:25capacity is still concentrated in Taiwan.
- 23:27It's a crucial geopolitical risk that any investor just has to
- 23:30keep an eye on. OK.
- 23:32We've established this as an extraordinary company
- 23:34operationally and strategically. Now we have to subject it to the
- 23:37rigorous lens of the classic value investor.
- 23:40We're using the 10 year metrics table from the research.
- 23:42This is where we figure out if the quality actually justifies
- 23:44the price. And this section really reveals
- 23:47the ultimate dichotomy of NVIDIA.
- 23:48It's a growth company priced for hyper growth, which means it
- 23:52instantly fails the strict conservative filters of classic
- 23:55deep value investing. Let's run through the failures
- 23:59quickly, but explain why they fail.
- 24:01PE ratio, the classic measure. It should ideally be below 15.
- 24:05Where's NVIDIA? Based on the latest trailing 12
- 24:08months earnings, the PE is around 50, and historically it's
- 24:1110 year median PE is already high, hovering over 50.
- 24:15The company is basically never traded anywhere near that strict
- 24:18PE of 15 or less, so by that one rule it's a decisive fail.
- 24:22It is not a cheap stock. OK, what about price to book PB
- 24:25that should be below 1.5? PB is currently sitting around
- 24:2935. I mean trying to value in video
- 24:31with price to book is kind of pointless.
- 24:33Why is that? It's an asset light company.
- 24:35It's value is tied up in intellectual property in the
- 24:39CUDA software in its future earnings power, not in physical
- 24:43factories or equipment on the balance sheet.
- 24:45But still, a number like 35 is clearly way outside the
- 24:48traditional range. The failure of these metrics
- 24:51just tells us this isn't a stock you're going to find by
- 24:53screening for low asset ratios. OK, now for the fun part.
- 24:57The quality metrics, things that are undeniable.
- 25:01What does NVIDIA ass comfortably?
- 25:03It asses virtually every single operational quality check with
- 25:07flying colors. The first measure of quality is
- 25:10gross margin. The requirement is typically
- 25:12above 40%. Nvidia's has been consistently
- 25:15over 50% for a decade, and thanks to all that pricing power
- 25:19has now soared to about 75%. 75%, that's world class
- 25:23profitability. That's what you see from
- 25:25software giants. It is.
- 25:26It's a measure of how proprietary and defensible their
- 25:29product is. They're capturing the vast
- 25:31majority of the value they create.
- 25:32OK. What about efficiency?
- 25:34How they use capital? So we look at return on equity,
- 25:36Roe and return on invested capital ROIC.
- 25:40The benchmark for is usually around 15% for Roe, 12% for
- 25:45ROIC. And Nvidia's performance?
- 25:47Their 10 year average Roe is about 26% and thanks to this
- 25:51recent AI earnings surge, they're trailing 12 months Roe
- 25:55has skyrocketed to a staggering 69%.
- 25:59It just tells you that management is exceptionally
- 26:01efficient at turning shareholder equity into profit.
- 26:05And for ROIC, they consistently deploy capital at very high
- 26:08rates of return, well above that 12% benchmark.
- 26:11When a company can reinvest its earnings at such high rates, it
- 26:15just creates this enormous compounding machine.
- 26:18And finally, the cash machine itself, free cash flow.
- 26:21This is the lifeblood. The rule is it has to be
- 26:24positive every year with a healthy margin.
- 26:26NVIDIA has generated positive free cash flow every single year
- 26:29for the past decade. The margin has recently hit 44%.
- 26:33It's an incredible cash generating engine and that
- 26:36excess cash is what allows for the aggressive R&D and the
- 26:39massive buybacks we're about to discuss.
- 26:41So the conclusion for the Traditionalist is great company,
- 26:46but the market has always known it's a great company, so you
- 26:49almost always have to pay a premium.
- 26:50Exactly. Which brings us to the modern
- 26:52valuation method, the discounted cash flow analysis, the DCF,
- 26:57which tries to quantify the value of all that future growth.
- 27:00Right. The DCF is essential here
- 27:02because it looks past historical ratios and focuses on the
- 27:05present value of future cash. First, we need a discount rate,
- 27:10the WACC weighted average cost of capital.
- 27:13The reports calculated this at 9.14%.
- 27:16So what does that 9.14% signify? It's the minimum required rate
- 27:20of return. It's a hurdle rate.
- 27:22We have to project all the future cash flows and then
- 27:24discount them back to today's dollars by at least that much
- 27:27per year to find their present value.
- 27:29OK, so let's run 3 scenarios to test the sensitivity.
- 27:32Let's start with the most pessimistic 1 scenario one, the
- 27:35no growth case. This assumes that from their
- 27:38current massive free cash flow run rate, about 60 billion a
- 27:42year, that FCF just stays flat forever.
- 27:47No more AI growth, no new markets, just 60 billion a year.
- 27:51Discounting that back gives you an intrinsic value of about $243
- 27:56per share. Wait, hold on a second.
- 27:58If the stock was trading around $186 at the time of the report
- 28:02and the no growth intrinsic value is 240 three, that means
- 28:06the stock was actually trading about 23% below its value in the
- 28:09most pessimistic scenario possible.
- 28:11That is the astonishing finding. It suggests that the market
- 28:14isn't pricing in zero growth, but that the sheer volume of
- 28:18cash they're generating right now already covers a huge
- 28:21portion of the market cap. In other words, the company has
- 28:24grown so profitable so fast that its current earnings power
- 28:27provides this large inherent margin of safety and that.
- 28:30Completely changes the expense of stock narrative.
- 28:32OK, what about scenario 2, conservative growth?
- 28:35Let's assume FCF grows at a modest 5% a year, right?
- 28:38Roughly in line with long term GDP, a very conservative
- 28:42assumption. In that scenario, the calculated
- 28:45intrinsic value jumps to about $347 per share.
- 28:49So if you demand a comfortable 30% margin of safety on that,
- 28:53your buy price is around 243. Given the market price was 186,
- 28:57this conservative 5% growth scenario suggests the stock was
- 29:01trading with a healthy margin of safety.
- 29:04This is the key take away from the DCF.
- 29:06If NVIDIA can sustain even very modest mid single digit growth
- 29:11over the long term, the stock, based on its current earnings,
- 29:14appears to be undervalued by this framework.
- 29:16And what about the last one, Scenario 3, nominal growth.
- 29:19Growth, say 10% a year for the next decade.
- 29:22If they can sustain 10% growth, which seems highly achievable,
- 29:25the intrinsic value shoots up to about $493 per share.
- 29:29In that scenario, the stock is trading at roughly half its
- 29:32intrinsic value. It really suggests the market
- 29:35isn't pricing in those crazy 60% growth rates, but rather a
- 29:38moderate sustained growth path. So to sum up the valuation part,
- 29:43it looks expensive on the classic metrics, but the DCF
- 29:46analysis suggests the sheer volume of cash flow means the
- 29:49price is defensible and maybe even offers a margin of safety
- 29:52against pretty moderate growth expectations.
- 29:54And the research reports we're looking at reached a similar
- 29:57conclusion. They assigned a 12 month target
- 30:00price of $235, which was derived from a probability weighted mix
- 30:04of different cases, showing a lot of confidence in the
- 30:07durability of this growth. The quality of a company is
- 30:09always tied to its leadership, and in Nvidia's case, you have
- 30:13Jensen Wong, CEO since 1993 / 30 years.
- 30:18That kind of continuity is almost unheard of in tech.
- 30:21And that continuity isn't just a footnote, it is the fundamental
- 30:24reason for their mode. Only a leader with that kind of
- 30:27long term vision could have invested billions into something
- 30:31like GPU computing for AI starting with CU Day back in
- 30:342006, years before it showed any real return.
- 30:37I mean, that's arguably the single greatest strategic
- 30:40decision in modern semiconductor history.
- 30:42And his interests are totally aligned with shareholders.
- 30:45He still owns about 3 1/2% of the company, a stake worth 10s
- 30:49of billions, which just dwarfs his compensation package.
- 30:53He's a true owner operator. That alignment is key.
- 30:56It ensures his focus is on long term value creation, not just
- 30:59short term quarterly results. Now we do have to acknowledge
- 31:02the blemish, the 2022 SEC settlement over the crypto
- 31:07mining revenue disclosures, right?
- 31:09But the reports do indicate that transparency and governance have
- 31:12improved a lot since then, which shows management is responsive
- 31:15to accountability. OK, let's look at capital
- 31:17allocation, which has shifted pretty dramatically.
- 31:20Historically, they paid a tiny token dividend.
- 31:23Yeah, the annual dividend is $0.16 a share.
- 31:25It's negligible yield, but it always signaled a core
- 31:28principle. They preferred to reinvest every
- 31:31single dollar back into that high return R&D to fuel future
- 31:35growth. But now the sheer volume of cash
- 31:37being generated, that 60 billion a year run rate, means they can
- 31:40do both. They can reinvest massively and
- 31:42return capital to shareholders. And the big capital allocation
- 31:46story now is the massive aggressive share repurchase
- 31:49program. This is management's way of
- 31:51telling the market we think our stock is cheap.
- 31:53Shouting it from the rooftops, in August 2023, the board
- 31:57authorized a huge $25 billion buyback.
- 32:01Then, in August 2025, they authorized an additional
- 32:05jaw-dropping $60 billion program with no expiration date.
- 32:10So that's 85 billion in total authorized buybacks.
- 32:14That's an enormous figure. It provides this huge structural
- 32:17support for earnings per share. It's so significant because it's
- 32:20so different from their past. The total cumulative buybacks
- 32:23over the previous decade were only about 15 billion.
- 32:26This new magnitude just shows management's absolute conviction
- 32:29that the stock, even with its high valuation multiples,
- 32:32represents a good long term value based on its intrinsic
- 32:35value. And institutional support is
- 32:37still strong, which confirms this confidence.
- 32:40Yeah, institutional ownership is very high.
- 32:42About 69% is held by major players like Vanguard and
- 32:45BlackRock. It just indicates general
- 32:47satisfaction with management and their performance.
- 32:50We have covered a lot of ground today.
- 32:52Let's try to summarize the essential takeaways from this
- 32:55deep dive into the NVIDIA machine.
- 32:58First, I think the business transformation is absolute.
- 33:01NVIDIA is now a data center powerhouse.
- 33:03It's generating nearly 90% of its revenue from that
- 33:06infrastructure, and it's achieving this hyper scale
- 33:09growth even after effectively losing the entire China market.
- 33:13Second, their Moat is getting deeper and it's centered less on
- 33:16just having the fastest chip and more on the stickiness of that
- 33:20decade old CUDA software ecosystem, which is now
- 33:24reinforced by the operational efficiency of the Blackwell
- 33:27platforms 10 times throughput per MW.
- 33:30And 3rd, their future growth is being diversified into these
- 33:32really resilient markets, especially sovereign AI that's
- 33:36driven by national security and strategic policy, and it
- 33:38significantly reduces that customer concentration risk we
- 33:41were worried about. And finally, valuation wise,
- 33:44while the stock fails all the strict classic value investor
- 33:47test because of the high multiples, the DCF analysis
- 33:50suggests that the sheer volume of cash being generated combined
- 33:53with the high probability of even moderate sustained growth
- 33:57of five to 10% a year more than justifies the current price.
- 34:01It confirms its status as a truly great company.
- 34:04Which brings us to the ultimate provocative thought for you to
- 34:07consider as you digest all of this.
- 34:09NVIDIA has successfully navigated 2 massive industry
- 34:12transitions in its history, from PC graphics to gaming, and then
- 34:16from gaming to the AI data center.
- 34:18So the question is this. Can the company, relying on its
- 34:21software mode in this rapid architectural evolution,
- 34:24successfully fend off its largest customers, the
- 34:26hyperscalers like Google and Amazon, who are simultaneously
- 34:29becoming its largest competitors by developing their own custom
- 34:32silicon, those ASICS? It is the highest stakes game in
- 34:36the entire tech world. How long can NVIDIA keep its
- 34:39lead when its best clients are actively trying to
- 34:41disintermediate them? That tension, the race between
- 34:45cities, flexibility and the ASICS optimization, is the long
- 34:48term risk that will define the next decade for this business.
- 34:52Thank you for joining us for this deep dive.
- 34:54And remember, this content is for educational purposes only.
- 34:58It does not constitute investment advice.
- 35:00We always encourage you to conduct your own independent
- 35:02research. And if you do want to track the
- 35:05total market temperature and get a data-driven perspective on
- 35:07market cycles, welcome to search for and download the iOS app
- 35:11Buffett Indicator from the Apple App Store.
- 35:13For listeners who are interested in more in depth content just
- 35:15like this, please welcome to join our Deep Value Club
- 35:18membership. We will see you on the next deep
- 35:21dive.