Asymmetric Edge Research

Most finance accounts post hindsight screenshots. This one doesn't. How this works: • Dated, falsifiable calls, graded in public later • Regime and risk-on/off reads, in context • Chokepoint research: asymmetric setups • Conviction and risk: the thinking, not live tickets • Honest postmortems: what I got right and wrong • Methodology shown, not just screenshots Focus: AI infra, energy, special situations. I run my own capital with this framework. If you manage your own book and want the calls, the research and the receipts, subscribe here ↓

Don't think there is reason to worry. This selling looks like positioning to me, not a fundamental break. I checked and $MU is at 5.5x forward on NTM EPS of $143.78. No HBM qual failure in the data, no guidance cut either. The two things that would actually break it are a guidance cut or an HBM disqualification at the next earnings call. Until one of those shows up I'm treating the tape as noise.

The market definitely wants to destroy $MU. But if sentiment shifts today, it will not get its will

The -55% end of that range is $SNDK, the least comparable name there. I pulled the fiscal years, SanDisk lost money in three of the last four. FY2025 EPS was -11.32. Micron printed +7.59. It only spun out of Western Digital in Feb 2025. NAND with no earnings floor selling off harder doesn't tell you DRAM broke. I'm still holding my Micron.

The big four memory makers are down anywhere between 30-55% over... JUST THE PAST MONTH?!? $SNDK $MU $INTC $SKHY

Apple was at 26x forward 15 months ago and it is 37x now, an absolute peak. About three quarters of that move was the multiple. Earnings did the rest. People buying at this level have to pray that Apple keeps its current multiple, otherwise any earnings will be eaten by multiple normalization.

$AAPL One more double and Apple will be the first $10,000,000,000,000 company in the world. For reference, it doubled just in the last 15 months. And people were saying “law of large numbers” blab blah blah. It can happen again before the end of this decade. Maybe even sooner.

I think the core technology part is right, the litho news is real progress. The trade data is just moving the other way. China's June imports grew 36% from a year ago while exports grew 27%, and that's the fourth month in a row imports grew faster. I went back through the customs prints to check. So they can build a parallel chip stack and still buy more from everyone else than last year.

China is probably going to achieve autarky It only required the complicity of American and European leaders selling out their entire industrial base over one generation for a temporary increase in profits (at the cost of destroying the entire goods producing sector)

@LongGameEquity Yes, that's what people actually often miss. A high P/E can quickly reach a much lower value if the company grows fast. So in 2 years, keeping everything else equal AMD would have a forward PE around 20-30.

@AsymEdge The 2027 and 2028 forward pe is crazy low

$AMD is down 19% in a week… All while delivering nothing but incredible news, massive AI demand, and one of the biggest growth opportunities in the market. 🤯 Welcome to the stock market 😂📉📈 The fundamentals can improve while the stock gets punished. That’s the game.

Oracle trades at 14 times earnings. Palantir trades at 75. They are forecast to grow earnings at the same rate next year. Everyone calls that sentiment. The filings say otherwise. Palantir's operating margin over the last five quarters went 38%, 43%, 48%, 55%, 58%. It spent $40M of capex over the past year to get there. Forty million, on five billion of revenue. Oracle's margin is 38%, which is where Palantir started that run. Getting there took $55.7B of capex against a depreciation charge of $9.3B, so it is buying gear six times faster than it writes the old gear off. Depreciation is capex arriving late. That 38% is calculated before the bill lands for the equipment producing the growth. One of them is compounding margin for free. The other is renting it with $124B of net debt behind it. Both sit near the top of my screen on upside. What I am underwriting on Oracle is the capex slowing before the depreciation catches up.

Tesla's driverless share of Robotaxi miles went from 4% in January to 61% in June. The humans really are coming out of the cars. Total paid miles fell 64% from the March peak. 431,000 that month, 156,000 in June. Driverless miles compounding double digits a week. Total miles down by two thirds. At this pace the mix hits 100% around Q3. After that the driverless number is the whole number. Run 10% a week forward from there. Q3 still lands below Q1. Their chart read 2.4 million in June. In October it needs to read 3.3 million. Anything under that and nothing actually grew.

$FICO reports today and the setup going in is odd. The stock is down 39% since May of last year. Over that same stretch the consensus estimate for its next twelve months of earnings went up 40%. I pulled the forward numbers because that felt wrong. In May 2025 the market paid about 59 times forward earnings for this business. Today it pays 26. The estimate itself never turned down once along the way, it just kept grinding higher while the price fell. So nearly the whole decline was the multiple coming in. The business behind it kept growing the entire time. People sell those two situations the same way and they really shouldn't. One is a company running into trouble. The other is just a cheaper price on the same earnings. Back on June 18 I said the selling was overdone and that today's print would decide it. Two days later I put a number on it, +94% upside, second only to Palantir on my screen. Stock was $1,096 then, $1,336 now. The rest of it gets answered after the close.

Everyone's quoting CXMT at $490 billion. That number came from 6.7% of the shares. The rest can't be sold yet. SpaceX did the same thing in June. They let 4.2% trade and it ran to about $2.6 trillion. It's at $116 now. For reference, the IPO price was $135. So I'm not holding CXMT's number up next to $MU. Micron's price comes from a market where people can actually sell.

The irony here is that Chinese investors have chosen to value CXMT, who has ~8% market share, at ~$500 Billion, implying $MU should have ~$1.4 Trillion valuation (~$1,240/share) based on Micron's market share alone (not to mention its other advantages, such as being US-based, more advanced memory chips, Flash, HBM, etc. etc.). Yet while the Chinese investors are buying CXMT, American investors are selling $MU. You couldn't make this stuff up!!!🤣

Apple's multiple had a better year than Apple did. The stock is up 59%. The earnings behind it are up 9%. The rest is people deciding Apple deserves 36 times earnings instead of 25. Nothing has to go wrong. At 36x you have already paid for the next eighteen months of growth. If the multiple drifts back to normal, Apple spends that year and a half earning you back to even. Reports Thursday.

@NotA_Bull Both interesting companies but very different trades. $RKLB did half the revenue of $NBIS last quarter on 1% of the capex. Rocket Lab spent $27M to do $200M of sales. Nebius spent $2.5B to do $399M.

Just increased my positions in $NBIS and $RKLB. Data centers and space, aren’t these the sectors of the future?

@antibearthesis Retail overleverage + AI capex + China supply fear ... The key is to understand that this is just a sentiment shift. Nothing has fundamentally changed about the business.

I hope you know EXACTLY why $MU & $SNDK are crashing.

You're mad at the wrong guys. Micron earned $7.74 a share in fiscal 2022 and lost $5.34 the year after. Record profit to a $5.8B loss in twelve months. Doing the fundamentals is what tells you 5.89x is meaningless. The real question is whether capped-price long-term contracts take that trough out. No ratio gets you there.

See it’s true, what’s the $mu P/E ratio now? I hate the fundamentals guys on here they are the worst

On point 1, that already happened once, at least partly. Google was planning around 4M TPUs for 2026 and ended up cutting to 3M, mostly because NVDA had already locked up the CoWoS capacity, roughly 60% of global demand this year. They reportedly went to Intel Foundry after that, 3M+ units, though not until 2028. So the bottleneck showed up at packaging, a layer below where the design competition is. That's why I'd rank your point 3 above point 4. More ASIC designs don't help much if the thing nobody can get is packaging and memory.

Many are fixed on why $NVDA is trading at a seemingly low multiple; there are a few reasons for this. I am not arguing here whether these are right or wrong, but these are real reasons: 1. There is a big question of what would happen with orders if $TSM 's allocation were reset today and $GOOGL, $AMZN, and $AMD would be able to order as much capacity of TPUs, Trainium, MI400 as they would want. That question then becomes a wider issue of what happens in a few years when $AMZN and $GOOGL have enough $TSM allocation to do so. Do they switch most of it to their local ASIC (like $AMZN did with CPUs and their Graviton CPU)? 2. It seems increasingly like $NVDA, with the backstops and investments in neoclouds, is trying to "buy its way" into more and bigger orders in a faster timeline, which is normally not a healthy sign. 3. If you truly believe that the compute shortage and high demand will continue to last for years, arguably a better way to "play" the AI frenzy is via memory makers or other parts of the semi stack that are more elastic in price and, at the same time, can't be skipped, as they represent a critical, monopolistic supply chain. 4. There is no doubt that more competition on the chip design layer has emerged in the last year than in previous years, with Tranium, TPU, $AMD, Cerebras and many others, so projecting the 75% $NVDA gross margin in the long-term future seems an outcome that is less likely than it was a year ago. Again, I am not making the case for any of these; I am just laying out some real cases here, because the narrative that the market is 100% wrong and that looking at forward P/Es and thinking the investment case is great for a company just because it seems cheap in terms of forward valuation is not the correct way to invest.

Oracle at 118$ and 14x forward PE is objectively cheap, but I think the funding matters more than the multiple here. Oracle pulled about $32 billion from operations last fiscal year and spent $56 billion on capex. So that's a $24 billion burn, and they filled it with roughly $40 billion of new debt while still paying out the $5.8 billion dividend. Meta earned $46 billion of free cash flow over its last year, Microsoft $72 billion, both of them out of operations. And about a third of Oracle's gross PP&E still isn't in service, so it hasn't started depreciating yet. Some of that cheap multiple is just timing.

Someone asked me about $ORCL at 118 and whether I would go long The answer is, I think $ORCL is a money maker stock from the 118 level but I probably will not buy it. I already have stuff like $META and $MSFT which are battleground stocks and can’t have too many in 1 port!

Agreed on the position. Worth watching what it now costs to hold it. NVIDIA is in talks to guarantee $250B of OpenAI's Ohio lease, and separately to fund about $350B of chips. That guarantee exists because OpenAI can't borrow at investment grade on its own. A foundation layer that underwrites its customers' credit is a different business than one that only sells to them.

$NVDA is becoming the foundation the AI economy will be built on. Its chips, networking and software now sit beneath nearly every major cloud, model and application positioning the company to compound with the industry for decades.

@PeterSchiff That 51% is mostly supply anticipation, the lockup unlocks around August 6 and the float at the June 12 IPO was tiny by design. I went back to the tape: $135 IPO, $225.64 four days later, $108.66 intraday on July 27. I'm not touching $SPCX until after the August print.

$SPCX is coming back down to earth. Shares just traded as low as $110.05. That’s 18.5% below the IPO price and 51% below the high. This is an example of why it’s so dangerous to rush into buying a heavily hyped IPO during its first few days of trading.

Everyone is calling this week the Mag 7 earnings test. I added up operating income for all 500 companies in the index. The Mag 7 is 16% of the earnings growth this quarter. Micron on its own added $31.2B. All seven together added $35.7B. Had to use operating income for this. On net income Alphabet looks like the biggest contributor by miles, but $98B of that is just mark-to-market gains on stakes they hold. Microsoft and Meta report Wednesday, Apple and Amazon Thursday.

Capex is what I'm reading Wednesday when $META and $MSFT report. Last four reported quarters, capex per dollar of revenue: META 35¢ MSFT 31¢ NVDA 2.6¢ ANET 1.5¢ Meta and Microsoft are financing the buildout. Nvidia and Arista are getting paid by it. So another revenue beat doesn't settle much. I want the new capex guide, the depreciation growth rate, and what's left of operating cash.