ANTI DEEP VALUE REPORT — NVIDIA (NVDA)
Hidden value in plain sight — by showing you where it isn’t
Everyone online is treating NVIDIA like a religion.
Every chart. Every meme. Every portfolio that suddenly “gets AI” has the same new tattoo: $NVDA.
Right now, NVDA is:
One of the most profitable tech companies on the planet.
Throwing off tens of billions in net income and free cash flow.
Trading at luxury multiples on what are very clearly super-cycle numbers.
Sitting at a valuation in the multi-trillion-dollar range.
From a Deep Value brain — think Cundill, Marks, Burry, Zell, Buffett, Munger — that’s exactly what an Anti Deep Value stock looks like:
Fantastic business. Peak-cycle numbers. No asset floor.
Crowd already all-in. Price assumes perfection.
This is not a “NVIDIA is garbage” piece.
This is a Crash Lab for the most loved stock in the world, written from the other side of your brain:
The side that thinks in cycles, not just narratives.
The side that cares about salvage value, not just growth.
The side that asks “what’s left if the story stops working?”, not just “what if the story keeps going?”
1. What this report is (and what it isn’t)
Let’s get one thing out of the way.
This is not:
A fraud call.
A “short this now” pitch.
A “AI is a bubble, everything is fake” rant.
It is:
A ledger-first look at NVDA at today’s price.
A cycle-aware and asset-aware read: what happens if this goes back to “normal”?
A reality check on a simple Deep Value question:
“If Peter Cundill woke up in 2025 with this balance sheet and this price in front of him, would he ever call this Deep Value?”
You already know the answer.
2. The market’s fairy tale about NVDA
The story the market is telling itself is clean and extremely seductive:
AI is the new electricity.
GPUs are the shovels in the gold rush.
NVIDIA is the shovel monopoly.
Therefore, it “deserves” any valuation the market wants to give it.
The recent numbers seem to justify the worship:
Revenue has exploded in just a few years.
Net income has gone from “nice” to “state-level budget.”
Free cash flow looks like a firehose.
Gross margins are sitting in the 70–75% zone.
Net margins hover around 50%+.
On the back of that, the market thinks it’s being rational when it pays:
A high double-digit P/E on trailing earnings.
A still-rich forward P/E on already aggressive forecasts.
A price-to-sales multiple most industrial companies will never see.
A price-to-book multiple that only makes sense if you believe cycles don’t matter any more.
Between the lines:
The market isn’t pricing “a great business at a fair price.”
It’s pricing peak-cycle profits at a peak-cycle multiple — and calling that “reasonable.”
That’s exactly the kind of setup Deep Value is designed to walk away from.
3. The ledger, not the meme: what the last two years really say
Forget the hype. Look at the income statement and cash flow for the last two years.
3.1 Income statement: everything at 11/10
At a high level, the P&L looks like this:
Revenue: more than doubles in a very short window.
Net income: goes from “strong” to “absurd,” tens of billions per year.
Net margins: float in the ~50–55% range.
Gross margins: in the low- to mid-70s, driven by one engine: AI data center.
The core reality:
Data center (Compute & Networking) has become the whole story.
Gaming, auto, and everything else are now side quests from a valuation standpoint.
Quarterly results repeat the pattern:
Data center revenue prints new highs.
Margins remain obese.
Guidance and commentary keep telling you, “Yes, this is the new normal. Don’t worry.”
On the surface, this looks like a perfect business at a perfect moment in the cycle.
From a Deep Value point of view, that’s exactly why you slow down and grab the flashlight.
3.2 Cash flow: a monster, but a super-cycle monster
Cash flow from operations and free cash flow have gone vertical:
CFO is now in the tens of billions per year.
Free cash flow is in the same ballpark.
Some quarters, the company generates more free cash than many countries’ annual tax intake.
That’s the part everyone clips for social media.
Here’s what most people skip:
a) Customer advances — tomorrow’s demand, booked today
A meaningful slice of that cash comes from customer prepayments and advances:
The biggest customers — hyperscalers and cloud giants — are prepaying billions to secure their place in line.
Those advances show up in working capital; revenue gets recognized against them over time.
It’s real money.
It’s just not pure “we sold, we collected” money. It is:
Future demand pulled into the present.
The headline FCF is front-loaded by these advances.
b) Inventory write-downs & purchase commitments — the China punch
Then there’s policy risk made visible:
U.S. and allied export controls on high-end AI chips to China forced NVIDIA to:
write down inventory, and
take charges on purchase obligations tied to chips it suddenly couldn’t ship as planned.
We’re talking multi-billion-dollar hits from one policy change in one key market.
The important part is not the exact dollar number.
The lesson is:
If your business is built on huge fixed commitments to supply hardware into a politically sensitive demand curve, one law can move billions of future profit into the “oops” bucket overnight.
Between the lines:
The free cash is very real.
It’s also super-cycle free cash — boosted by:
prepayments, and
a willingness to sign very large, very path-dependent supply deals.
A Deep Value brain is allowed to admire the cash machine and still ask:
“What happens when the prepayments slow, the charges repeat, or the cycle just goes back to ‘normal’?”
4. Balance sheet vs. market cap — no salvage-value floor
Now look at the balance sheet.
You don’t see:
timberland,
mills,
ships,
real estate portfolios,
steel plants you can shut and sell.
You see:
A modest amount of property, plant, and equipment compared to the valuation.
A large pile of current assets (cash, marketable securities, receivables, inventory).
Equity that is a small fraction of the market cap.
At today’s price, you’re paying roughly:
Dozens of dollars for every dollar of book value.
Hundreds of dollars for every dollar of hard, physical assets.
This is the purest version of:
asset-light, obligation-heavy.
If things go wrong, there is very little “stuff” you can sell to recover value.
Your “floor” is almost entirely intangible franchise value — the thing the market is already paying a peak multiple for.
From a Cundill/Zell perspective, it’s simple:
This is not an asset play.
This is not a salvage-value story.
This is a bet that:
earnings stay enormous, and
the market continues paying luxury multiples for those earnings.
That can work.
It just has nothing to do with Deep Value.
5. The Michael Burry lens — late-stage supply mania
Michael Burry has written and spoken for years about the shape of late-stage bubbles.
His core point:
The top is not just about crazy prices.
The top is when supply goes vertical —
everyone funds the theme,
everyone ramps capacity,
everyone celebrates capex,
and almost no one reads the obligations.
What does that look like in AI hardware land?
NVIDIA planning hundreds of billions of dollars in electronics spend over a few years.
Hyperscalers committing hundreds of billions to AI data centers.
Governments throwing money at “AI sovereignty.”
Startups and incumbents all raising capital for more compute.
All at once.
All into the same theme.
From a Burry lens, the job of a serious investor right now is boring and unsexy:
Not retweeting capex plans.
Not drooling over top-line growth.
Just going back to the 10-K and reading:
purchase obligations,
customer concentration,
policy risk,
how much of today’s demand is prepaid,
how much of tomorrow’s capex is already locked in.
That’s what an Anti Deep Value report is.
You read the obligations while everyone else is celebrating the projects.
6. When was NVIDIA “fairly priced” in the last three years?
There’s an obvious follow-up question:
“Okay, but was there a moment in this AI run where NVDA was actually fine — or even attractive — on price?”
6.1 For a growth investor
Yes.
If you go back roughly three years:
Late 2022:
Semis were hated.
Macro fear was high (inflation, rates, recession talk).
AI was a theme, not the theme.
NVDA traded much lower than today, before the truly vertical move.
At that point:
You could reasonably say:
“This is a dominant chip company, with a fortress balance sheet, in a space that will matter. I’m paying a premium multiple, but the AI upside isn’t fully priced in.”
For a growth / quality investor, late 2022 and early 2023 were probably okay entry points:
Still expensive on classical metrics,
but with more optionality and less priced-in perfection.
6.2 For a Deep Value investor
Even then, it was never Deep Value:
No hard-asset margin of safety.
High multiples on current and future earnings.
Business model dependent on:
technology leadership,
a concentrated customer base,
and a politically sensitive supply chain.
A Cundill-style operator in 2022 might have said:
“Great business, yes. Deep Value, no.
Call me when the market is disgusted, not just nervous.”
Between the lines:
There were times when NVDA was “not insane” for a growth buyer.
There has not been a time in the last three years when it was a classic Deep Value situation.
Today, at current prices, that distinction isn’t even close.
7. Three years ago vs. three years from now — competition and cycle
Three years ago:
NVIDIA looked almost like the sole fully-scaled shovel vendor in AI training.
AMD existed but lagged.
Google’s TPUs were mostly internal.
Amazon, Microsoft, and Meta were primarily customers, not serious chip competitors.
Three years from now, the map will look different:
Google:
Mature TPU generations targeting cost and efficiency for internal workloads.
Amazon (AWS):
Trainium and Inferentia positioned explicitly as ways to cut GPU bills.
Microsoft:
Custom accelerators (e.g., Maia) with the stated goal of reducing dependence on third-party GPUs.
Meta:
In-house AI chips aimed at lowering cost and power consumption versus external options.
OpenAI + foundries:
Working on custom accelerators to support enormous power footprints.
China:
Domestic champions like Huawei building local AI silicon with full policy backing.
NVIDIA will still be massive.
But the landscape shifts from:
Near-monopoly with scarcity premium
to:
Oligopoly with big, motivated customers building their own alternatives.
Deep Value doesn’t require you to bet against NVIDIA’s engineering talent.
It simply asks:
“Do I want to pay ‘monopoly forever’ prices at the exact moment when every large customer has both the incentive and the budget to compete with that monopoly?”
8. Mini SOTP — how much are the pieces actually worth?
Let’s do something the memes don’t:
treat NVDA like two separate businesses plus a balance sheet.
AI Data Center (Compute & Networking) — the crown jewel.
Everything else — gaming, auto, pro-viz, etc.
8.1 Block 1 — AI Data Center
We’ll be generous.
Mid-cycle AI data center revenue: $100B.
Mid-cycle net margin: 45%.
That gives you $45B in mid-cycle net income from the core franchise.
Now ask: what multiple does a strict Deep Value brain give to a world-class, but cyclical and political-risk-exposed, franchise?
Let’s say 20× — that’s already a Buffett-tier honor.
Value for core AI data center: ≈ $900B
8.2 Block 2 — The rest
Now everything outside core AI:
Gaming, auto, pro visualization, the long-tail of products.
We’ll be generous again:
Mid-cycle revenue: $20B.
Net margin: 25%.
That’s $5B of net income.
Give that a healthy 15× multiple (solid business, not irreplaceable):
Value for everything else: ≈ $75B
8.3 Net cash and securities
Add a ballpark of $60B+ in net cash and marketable securities (cash plus investments, minus debt).
Add it all together:
$900B + $75B + $60B ≈ $1.0–1.1 trillion
That’s your generous Deep Value “sum of the parts” for NVDA.
Compare that to the current valuation, which sits many multiples above that.
Even with friendly assumptions on:
revenue,
margins, and
multiples,
you end up roughly 75% below today’s quoted value.
That’s not a precise target.
It’s a sanity check on how far away we are from anything that looks like a margin of safety.
9. Scenario Map — Base / Best / Bubble Hangover
Now flip from SOTP to scenarios.
This is where Deep Value lives: not “what’s your price target,” but:
“What does this look like in different states of the world?”
9.1 Table: three futures, three prices
Rough numbers, for intuition, not precision:
Interpretation:
Base case:
The world still wants a lot of AI, but at some point:growth slows,
margins normalize,
competitors chip away.
A 20× multiple on a still-fantastic $36–40B of net income gets you under $1T.
Best case:
The bullish forecasts happen. Revenue climbs to ~ $290B; margins stay heroic; net income hits ~ $130B.
Then the question becomes:Does the market still pay 30× that number?
Or does it compress to 20× as the story matures?
At 20× you’re under $3T.
At 30×, you’re roughly in line with today’s valuation.
Notice: even best-case is not obvious upside from here.Bubble hangover:
Growth slows faster than bulls expect.
Revenue peaks around $130–150B.
Margins drift back toward mid-30s.
The market gets tired of paying extreme multiples for hardware.Suddenly, 15–20× is the right range.
You’re right back into the $675B–1T zone.
Between the lines:
Today’s price only makes real sense if the Best case happens and the market keeps applying a “story stock” multiple to what will then be a very large, very cyclical global hardware and software franchise.
That’s the opposite of a margin of safety.
10. Crash Lab — Why your brain wants to love NVDA at any price
This section is not about NVIDIA.
It’s about you.
If you’re honest, NVDA hits almost every button in your investor psychology:
10.1 FOMO and social proof
Every serious investor seems to own it.
It’s in every big benchmark.
The smartest funds talk about it on every call.
Your feed is one long NVDA highlight reel.
The quiet script in your head goes like this:
“If I don’t own it and it keeps going up, I’m the idiot.”
“If I do own it and it blows up, well… everyone else was long too.”
That’s not analysis.
That’s career risk and herd risk backstage, wearing an “investment thesis” costume.
10.2 The Narrator in your head
You almost never tell yourself:
“I bought NVDA because I was scared of missing out, and I didn’t want to look stupid at the next meeting.”
You say:
“Given the structural role of AI and the quality of the platform, it made sense to own the leading operator.”
Same behavior.
Different story.
The first version is about fear and ego.
The second version is about identity — you as the visionary operator who “sees where the world is going.”
Deep Value forces you to strip the identity out and ask:
“What did I actually buy?”
“At what multiple?”
“What is protecting me on the downside?”
If the answer to that last question is “hope the story keeps working,” it’s not Deep Value.
10.3 Story vs. price
With a stock like NVDA, the story is so good that your brain quietly swaps the question:
From: “Is this a good trade at this price?”
To: “Is this an amazing company with a huge future?”
Those are not the same question.
You can be completely right on the second and completely wrong on the first.
Cisco and Intel in 2000 were:
Real companies,
Real profits,
Real innovation.
They were also terrible buys at bubble multiples.
Tesla 2021 was the same exam with a different logo.
10.4 Benchmark brain vs. owner brain
Benchmark brain says:
“If the index is full of NVDA and I’m not, I’m underperforming and I look dumb.”
Owner brain says:
“If I buy a world-class business at a price that bakes in 10–15 years of perfect execution, my risk is not just a bad quarter. My risk is owning a great business at the wrong price for a very long time.”
Deep Value lives in the owner brain, not the benchmark brain.
It is more comfortable looking stupid alone with a cheap salvage-value asset
than looking smart with everyone else at the top of a mania.
10.5 The illusion of safety in size
NVDA feels safe because:
It’s big.
It’s liquid.
It’s popular.
Everyone talks about it.
Your brain hears: “too big to really hurt me.”
Reality:
Size does not create a floor if there are no hard assets to liquidate.
Size does not protect you from multiple compression.
Size does not stop your customers from building their own chips.
The only real safety is buying cash flows and assets at a price that already assumes disappointment.
Right now, the price assumes:
no big policy surprises,
no serious competitive erosion,
no sharp normalization of AI capex,
and a market that keeps rewarding the story.
That can happen.
It just isn’t Deep Value.
Between the lines:
The Deep Value question is not “Is NVDA amazing?”
It’s:
“If this stock dropped 60–70% and I saw the same ledger, would I finally be interested?”
If the answer is yes, then today’s price is your ego trying to front-run your own patience.
11. History Rhymes — Walmart, Cisco 2000, Tesla 2021
History doesn’t repeat. It rhymes.
NVDA can rhyme with more than one past story at the same time.
11.1 Walmart then vs. NVIDIA now
Walmart in its early hyper-growth phase:
Built a gigantic tangible footprint: stores, DCs, trucks, land.
Even if growth slowed, you owned something real.
The stock could be expensive, but it wasn’t 40–70× earnings on peak-cycle numbers.
NVIDIA today:
Has very little in hard, saleable assets relative to its market cap.
The “assets” are:
CUDA,
the software ecosystem,
relationships with hyperscalers.
All real. All already fully understood.
That’s exactly what the market is paying peak multiples for.
From a Deep Value angle, NVDA 2025 looks much more like Cisco 2000 than Walmart 1985.
11.2 Cisco and Intel in 2000
Cisco in 2000:
Dominant in routers and switches.
Real profits, real growth.
The internet wave was not fake.
Intel:
Dominant in CPUs.
Massive R&D, deep moat.
Bought at bubble multiples:
They became dead money for a decade or more for anyone who bought the peak.
The businesses did fine.
The stocks spent years digesting the overpayment.
Lessons:
You can be right on the technology and wrong on the price and still lose.
“Will this company be around and profitable?” is a different question from “Will I make money from this entry point?”
11.3 Tesla 2021
Tesla is the recent case everybody remembers:
Genuine innovation.
Serious operational achievement.
A giant narrative about changing the world.
For a while:
The stock price and the story were the same thing.
The multiple implied a future where nothing ever went wrong.
Then:
Competitors arrived in force.
Pricing power wasn’t infinite.
The multiple compressed, even as the business continued to operate.
NVDA today rhymes with Tesla 2021:
Beloved stock.
Crowded positioning.
Real business and real moat,
priced as if nothing can dent it.
Deep Value’s job is not to call the exact top.
It’s to recognize when the mood and the math no longer line up with a margin of safety.
12. Meme Corner — lines you can steal for X/Twitter
You want traffic and memes? Here are a few lines designed for that:
“If your FCF yield starts with a 1 and your ticker starts with NV, you’re not buying Deep Value. You’re buying a religion.”
“NVIDIA is what happens when a truly great business meets a market that thinks cycles were outlawed in 2022.”
“Your GPU cluster is not a risk-free bond. It’s a leveraged bet on AI capex never getting boring.”
“This is not an anti-NVIDIA report. It’s an anti ‘I don’t care what I pay as long as it’s AI’ report.”
“Deep Value doesn’t hate NVDA. Deep Value just hates paying for 15 years of perfection upfront.”
Pick one for the opening tweet.
Use the others as replies, quotes, or screenshots.
13. Final verdict — Anti Deep Value, not anti NVIDIA
So where do we land?
NVIDIA is:
A world-class operator.
A genuine innovator in hardware and software.
The central node in the current AI capex boom.
A machine that prints an insane amount of profit right now.
At today’s price, though, you are buying:
Super-cycle earnings at
Super-cycle multiples, with
Almost no hard-asset floor, in a world where
Every major customer is also a rising competitor, and
Policy risk is embedded in the core product line.
For a Deep Value investor, that combination is not a “maybe.”
It’s a no.
Great business.
Very bad hunting ground for a brain that needs downside protection.
The point of an Anti Deep Value Report is simple:
To show you where Deep Value does not live,
To use a market darling as a case study in:
cycle risk,
narrative risk,
and psychological risk,
And to remind you that:
The real edge is not calling the top of the mania.
The real edge is being willing to walk away from it.
If NVDA ever trades at a price where:
the SOTP math,
the scenario map,
and your owner brain all say “yes”…
…you’ll know what to do.
Until then, the Deep Value part of your collective brain keeps this one on the watchlist, not in the portfolio
Brilliant breakdown of the asset-light risk most people completly gloss over. The Cisco 2000 parallel is spot-on because back then people dunno how expensive "infrastructure dominance" could get at peak multiples. What really stands out here is how customer prepayments inflate that FCF picture while simultaneously those same hyperscalers are all buildng their own silicon alternatives.