Pinned Tweet
OMG! That amazing feeling when your idol follows you back! #pabrai
@MohnishPabrai
You made my day sir!
Finding Value retweeted
“So Nvidia gives these people money to buy its GPUs?”
“Yeah.”
“And then guarantees the debt they take on to buy more GPUs?”
“Yeah.”
“What happens if these people don't make any money?”
“That's where the life insurance companies come in.”
How about $10B in debt and $15.5 B in liabilities. Also 4.5 billion per year cash burn rate. $IREN
Dont fall for these clickbait accounts, check the numbers yourself
$IREN
$5.73B in cash.
$6B worth of physical buildings and land completely debt-free.
Current market cap: $5.33B.
Readers added context they thought people might want to know
IREN's latest 10-Q shows ~$5.73B cash+restricted cash, $12.3B PPE (incl. buildings/land) but $10.2B total debt and $15.5B liabilities vs $5.1B equity. sec.gov/Archives/edgar… stockanalysis.com/stocks/iren/fi…
13
1,058
Finding Value retweeted
Retail traders celebrating a 2.2x forward price-to-earnings multiple on memory producers are misreading cyclical accounting.
When SanDisk trades at 2.18x, SK Hynix at 2.55x, and Micron at 2.82x while Costco commands 44x, retail investors conclude the market is irrational. Claiming that SanDisk can buy back half its market capitalization in twelve months misunderstands the physics of semiconductor capital expenditure.
Three mechanical realities explain why memory trades at low single-digit multiples:
The Denominator Collapse:
The forward earnings in a 2x multiple are calculated against peak cyclical profitability. Throughout 2025 and 2026, memory suppliers enjoyed gross margins exceeding 50% on severe HBM and enterprise NAND shortages. However, as confirmed by TrendForce contract price forecasts flipping negative in 1Q27, pricing leverage has peaked.
Memory fabrication involves massive fixed-cost depreciation on advanced cleanrooms. When wafer ASPs drop 10% to 15%, operating income experiences roughly 80% downward operating leverage. The denominator in that 2.2x multiple evaporates over four quarters. What looks like a 2x multiple today transforms into 20x or negative earnings during cyclical downshifts.
The Share Buyback Fallacy:
Retail models assume peak operating cash flows can be directed toward equity retirement. In heavy manufacturing, cash flow generated at cycle peaks must be preserved to fund multi-billion-dollar technology migrations. To maintain competitive bit densities, producers must continuously capitalize next-generation cleanroom tooling, including 300-layer NAND stacks and HBM4 advanced packaging. When contract prices decline and operating cash flow contracts, boards immediately suspend buyback programs to defend liquidity and debt covenants.
Consumer Annuity vs Commodity Depreciation:
Comparing memory producers to Costco fundamentally confuses business models. Costco trades at 44x because membership fee revenue operates like an annuity, supported by negative working capital and stable consumer demand. Memory producers manufacture an undifferentiated commodity subject to boom-bust inventory gluts. Peak cyclical earnings represent temporary windfalls, not durable compounding.
The Cyclical Inversion Rule:
The classic rule of memory investing remains unchanged: never buy memory equities when their price-to-earnings multiple looks historically cheap. Commodity semiconductor producers consistently print their lowest forward multiples at the cyclical peak, immediately before margin compression accelerates. Buying memory at a 2x multiple is the standard entry point of a cyclical value trap. $SNDK $MU $WDC
At what point do these Fwd PE bros realize that Fwd PE is made up from an estimate put out from analysts who just know to extend the current unsustainable earnings to the future (to the moon)! $SNDK $MU $WDC
2
9
836
Finding Value retweeted
Ok, I’m going to be as polite as possible even though I’m tired of explaining this. Forward earnings are horrible for assessing valuations when we’re at the end of the cycle.
Analysts extrapolate recent growth and elevated profit margins instead of anticipating mean reversion or recessionary contractions. As a result, the denominator in forward price to earnings ratios stay artificially high while the business cycle is already turning, making valuations appear more reasonable than trailing or cyclically adjusted ratios.
Students of financial history understand this. During the GFC, actual earnings collapsed about 92%, yet consensus estimates only fell about 6%. This disconnect happens at the end of every cycle and the beginning of every recession (although the GFC is an extreme example). Downward revisions to estimates only come after the damage is already visible.
This is a bad chart crime!
Analyst Estimates follow price narratives!
In 2020, analysts estimates were based on the world ending.
Today, analysts are estimating AI will take over the world.
They were wrong before, they will be wrong again.
BREAKING: US technology stock valuations have hit their lowest level since March 2020 amid the market correction.
The S&P 500 Info Tech Index P/E ratio is now at its lowest since March 20th, 2020, at 16.4x.
3
330
Finding Value retweeted
Every $MU conference goes like this:
> Memory will stay tight until 2065
> We're going to make more money than is possible
> There's so much money to be made that we might make too much money
> We'll invest $10 trillion in CapEx and probably buy back the whole company
I strongly disagree that Dev was unlucky
First he held onto stocks like $FICO at sky high valuations. And then when the prices crash, he doesnt buy. He has even sold a bit.
He then proceeds to buy AI stocks at sky high valuations.
Valuation matters! Humility matters.
I’m still thinking about Dev Kantesaria. It’s pretty sad.
He spent years avoiding the AI stocks, watching them all rocket upwards. He finally decides to buy them a month before they drop 30%.
The stocks he sold to fund the AI purchase (like Mastercard) are also likely up another 10-15% based on when he sold them.
One of the most unfortunate timings I have seen in a portfolio rotation.
5
443
Exactly!
“Buying a cyclical after several years of record earnings and when the P/E ratio has hit a low point is a proven method for losing half your money in a short period of time.”— Peter Lynch, Beating the Street
Memory stocks at rock-bottom valuations:
$SKHY — 2.3× forward P/E
$SNDK — 3.4× forward P/E
$MU — 3.9× forward P/E
Do what you want with this information.
2
322
Finding Value retweeted
"The market trades the stock like nothing has changed". Hmm. The stock is up 10x fold since mid 2025 despite having fallen 50% from its highs.
"The market doesn't know what to do with earnings that go up that fast". Correct. It just naively sends the stock up 20 fold alongside strong earnings and then investors end up losing money as a result. They've already lost 50%.
"Particularly for a stock that has historically been very cyclical". There is an important clue as to why the P/E multiple is low.
Market has now consolidated? That's what people thought last cycle. The industry still overbuilt and lost money in 2022-early 2024. Now you have the Chinese players hot on the industry's heels so you'll end up with aggressive Chinese competition long term.
"What's changed... is that you have HBM... that's on an annual cadence". So was traditional DDRRAM, DDR1, DDR2, DDR3, DDR4, DDR5...
"DDR is good for 7-8 years". The AI capex bulls are arguing right now old generations of GPUs are good for 7-8yrs also. Presumably they mean the HBM memory attached to those GPUs as well.
HBM is indeed harder and takes longer to produce. It will elongate the cycle a bit. But all the major memory players are adding HBM capacity as fast as they can and if and when mushrooming AI compute capex slows, HBM will tip into oversupply all the same.
"HBM ties memory more closely to the compute roadmap". Yes it's tightly linked with AI compute but that's where the risk lies because AI capex is frenetic atm and might not be sustainable at current levels. Moreover, memory has always been broadly tied to the level of compute. More compute needs more memory.
We're told a useful analogy is logic tells you not to touch the stove, memory makes you remember. Yes, that's indeed how computing works. Stop the presses.
"The market doesn't see that memory has become more valuable..." except that the stock went up 20-fold in one year. It seems like someone did notice.
"[Investors] are just saying, well the earnings have gone up a tonne, they're gonna probably crash, so we're going to put a very low multiple on it". Precisely, and they are correct to do so, because in all probability that is exactly what will happen.
The old adage that you should buy cyclicals on high PEs and sell them on low PEs is borne of hard won investor experience - mostly losing money buying cyclicals near the peak because the P/E looked low. Micron earned $8 per share as recently as 2025 and that was a good year for memory. In 2024 they made $1 per share and in 2023 they lost $4 a share.
"The earnings multiple is very close to the book multiple". The book multiple is 5.6x which is high (even after the stock halved), and the fact that that is similar to the P/E tells you they are massively overearning (100% ROE).
Moreover, MU has almost always ended the cycle at or below book, because they usually lose money for a while. Yeah the book value will rise fast during the boom years, but it still has to earn and retain $570/share for BV to catch up with where the stock price currently is, which is 3-4yrs worth of current omega-cyclically-inflated earnings. And a lot of those earnings will be invested in capacity expansions which will eventually glut the market.
A lot of the putative valuation anomaly disappears if you try to value a company more thoughtfully than just looking at one year's earnings/growth and the near term earnings multiple. If that is the depth of your process you'll get, as Munger used to say, the decidedly poor results you deserve.
Whoa - UBS Managing Director who covers Micron $MU on a podcast just released on Youtube.
Here comes the roller coaster ride:
Finding Value retweeted
THIS AGED LIKE FINE WINE
My advice in February was simple: run from tech into energy, and treat the AI stocks like the dot-com stocks of 2000.
Since that interview, our AI shorts have been crushed. CoreWeave is down 57%, Astera Labs is down 56% and Oracle is down 39% from where we shorted them.
On the other side of the book, Valaris is up 48% and Cenovus is up 61% from where we called them in December.
The internet changed the world and the dot-com stocks still crashed. AI is running the same play, and the crowd buying every dip is learning that the hard way.
Which side do you want to be on?
For the AI stocks, it will only get WORSE from here.
If you think $mu is so undervalued and Fwd PE is 3, why are insiders selling their stock hand over fist?
Insiders have sold over 250 mil in the last 6 months and not bought a single dollar.
Finding Value retweeted
Here’s a simple question on $MU. If this is truly a generational buying opportunity the likes of which have rarely been seen before, who would know it better than management? So, why aren’t they buying any stock? So what if they already own a lot, it’s free money right?

































