Asymmetric Growth OS

My money. Obsessive research. Long-term growth stocks. Every holding must fight for its place regularly. Tracked vs the S&P 500. Live portfolio on Savvy Trader.

[Asymmetric Growth OS @AsyGrowthOS]

Stocks are not football teams. You're allowed to change sides.

The Next Dollar series: companies fighting for a place in my long-term, real-money portfolio. Any sector.

Research, decisions and results vs the S&P 500. I'll show the bad calls too.

Live portfolio in bio.

[Asymmetric Growth OS @AsyGrowthOS]

@http_413 @nishffx Why should I reevaluate priorities? They should make cheaper compute... wasn't that their goal from the start?
"Trillion token context and compute to cheap to meter"

[Asymmetric Growth OS @AsyGrowthOS]

@cmsinvests because not everyone has 321,200$ in between their couch cushions

[Asymmetric Growth OS @AsyGrowthOS]

@drayinvests 22.5% so around 10% outperformace YTD

[Asymmetric Growth OS @AsyGrowthOS]

$NBIS: a ~40%-a-year scenario that needs $48bn of other people's money

— The short version —

The GPUs don't care how bullish your spreadsheet is. They still get old, and it turns out they're really fucking expensive to replace.

My latest research modelled Nebius's growth plan properly: the data centres, the chips, the refresh cycles and the financing. One scenario comes out looking incredible: about $1,203 a share from a $224.55 starting price. That's roughly 40% a year, or about 5.4x, to around the end of 2031.

But even that best case needs about $48bn of future project loans and customer advances (in the version with a six-month delay built in). That's not new shares; it's loans and prepayments. If you strip out all future outside funding and keep the build commitments, 2027 alone comes up about $9.6bn short.

And here's the contradiction that made me sit up: the worse things go, the more new shares $NBIS needs. In the weak scenario (pricier hardware, and rental prices squeezed at renewal) the model needs about $30.5bn of new equity and ends up around $111 a share, roughly −13% a year. The middle case needs about half that ($14.8bn) and ends around $452, roughly +15% a year. Bad economics don't just mean lower profits here. They mean selling more of the company at worse prices to keep the build going.

So the outcome is decided less by "is AI demand real?" (it clearly is) and more by "will lenders and customers keep funding the build at a sensible price?"

— The nerd cave —

What's reported:
• Q2 2026: group revenue $582.3M (cloud $574.9M). Operating loss $175.9M and net loss $190.4M, against adjusted EBITDA of +$236.2M. That gap is mostly depreciation, i.e. the GPUs getting old.
• H1 operating cash flow $4.504bn. It sounds amazing, but $4.395bn of it was customer advances: customers paying up front for capacity. That's real cash, but it isn't profit. It's a promise Nebius now has to deliver on.
• H1 net income was flattered by a $780.6M gain on an equity stake. It won't recur.
• 2026 guidance: ARR $7–9bn, revenue $3–3.4bn, capex $20–25bn. Read that last one again: capex this year is several times revenue.
• Target of 800MW–1GW "connected" by year-end 2026. Connected isn't billable; customer onboarding runs into H1 2027.
• $12.054bn of leases signed but not yet started.
• Resources: about $14.5bn pro-forma. That's June cash of $8.04bn, plus a $0.775bn secured loan in July, plus ~$5.68bn of notes in August. It's before second-half spending, so it isn't "cash in the bank today".
• Accounting: in the Q2 letter, Nebius said it now depreciates servers and network gear over five years instead of four. Maybe that's reasonable, but it lowers the annual charge without creating a single extra dollar of cash.

What I assumed (not guidance):
• 0.5GW billable from Jan 2027, adding 0.5GW a year of owned capacity, reaching 3GW owned plus 1.5GW partner-owned by 2031.
• Capital cost of $40–60bn per GW: 80% hardware with a 5-year economic life, 20% buildings and infrastructure over 20 years.
• Customer advances cover 35% of each new cohort's capital cost. Loans cover 60% of hardware at 7.5%.
• Stock-based pay at 5% of revenue, paid in new shares.
• $3bn of value for the non-core businesses (ClickHouse, Toloka, Avride). It's unverified, so treat it as a guess.

What comes out (from $224.55 on 11 Sep):
• Strong ($40bn per GW, strong pricing and margins): ~$1,203, ~40% a year. No new shares, but ~$48bn of outside loans and advances.
• Middle ($45bn per GW): ~$452, ~15% a year. Needs ~$14.8bn of new shares plus ~$54bn of loans and advances.
• Weak ($50bn per GW, renewal prices compress): ~$111, ~−13% a year. Needs ~$30.5bn of new shares.
• Re-run at the 24 Sep price, after a ~7% jump that day, the weak case is about −14% a year.

If the middle case can't raise new shares at all, it runs out of cash in 2028, about $1.3bn short. Capped at the share-sale programme that's already approved (~$2.4bn left), it's about $3.6bn short in 2029.

What makes the bull case: cheap capital per GW, prices that hold at renewal, very large paid scale, and debt markets staying open for ~$48–54bn.
What breaks it: financing drying up, renewal prices falling faster than hardware costs, chips going obsolete before the five-year accounting life, customer cancellations, or fixed lease and power bills that keep running even if the money stops.

$48bn of loans and advances is not $48bn of new common shares. It's a different claim with a different risk. But it's still $48bn somebody has to be willing to lend.

Source: Nebius Q2 2026 shareholder letter and financial statements; 2026 guidance.

[Asymmetric Growth OS @AsyGrowthOS]

Selling a winner feels disloyal.
Selling a loser feels like admitting you fucked up.
Congratulations. You've found an emotional reason to keep absolutely everything.
The Next Dollar series isn't fantasy football. No weekly transfer quota. No need to kick someone out because the audience is getting bored.
A good business can have a bad quarter. Patience can be exactly right.
But “the business needs time” and “I don't want to admit the story changed” aren't the same thing, however convincing they sound at 2am.
Existing holdings don't get to stay out of sentiment. New companies don't get in just because they're shinier.
Sometimes the whole exercise ends with no changes. The challengers didn't offer a good enough reason to move the money.
No shock signing. No dramatic eviction.
Your portfolio doesn't need a season finale.

[Asymmetric Growth OS @AsyGrowthOS]

The market doesn't give a shit where you bought.
“I'll sell when I get back to even” sounds like a plan. Sometimes it's just a grudge with a brokerage account.
A new buyer gets the same future business as you. No penalty for arriving late. No loyalty bonus for having suffered through six earnings calls.
There may be excellent reasons to keep the stock. Your old receipt is not a valuation.
Cost basis can matter for tax. The company still doesn't owe you a recovery.

[Asymmetric Growth OS @AsyGrowthOS]

@hany99dev Only an Astra 6.1 with Astra 6 level intelligence or higher and usage comparable to Opus 5.5 would make me come back...i will not settle for anything less

[Asymmetric Growth OS @AsyGrowthOS]

@AcedKnisley @melvindvivas you still have something to burn? i burned the bank one upon dropping in 24h

[Asymmetric Growth OS @AsyGrowthOS]

@YoYInvestor @Ryan_M_Hartman @amitisinvesting Not trying to sound smart...It's just that I've seen so many baseless hype posts that never mention dilution and I wrongly assumed this was also one of them so I didn't even bother reading the whole post...my bad

[Asymmetric Growth OS @AsyGrowthOS]

@buildwithrajath Astra is still the king for everything but is unusable...if they would just release Astra 6.1 that has Opus 5.5 usage i would give them any amount they ask

[Asymmetric Growth OS @AsyGrowthOS]

@johhnyWalkerAZ i held for around 2 years and got out....the opportunity cost of not owning other things started to get much bigger than betting that Ondas will maybe wake up someday...if it starts waking up ill be the first one to get back in but for now its not worth the wait

[Asymmetric Growth OS @AsyGrowthOS]

@xikhar Unfortunately Opus 5.5 is not the best....it is the cheapest for sure and good enough for everyday tasks....but for long complex reasoning work that requires many complex instructions and rules its like a kid with ADHD on LSD with handcuffs

[Asymmetric Growth OS @AsyGrowthOS]

@post_human_pr @Layton_Gott Tutorial? They dont even know the tutorial exists

[Asymmetric Growth OS @AsyGrowthOS]

@Layton_Gott Its like we live in parallel universes...I've reached a point where i can only have an interesting conversation with people on here and the AI itself...everyone else is asleep at the wheel

[Asymmetric Growth OS @AsyGrowthOS]

@LeifInvests Ok my bad then...sorry

[Asymmetric Growth OS @AsyGrowthOS]

Maybe they only give resets to 5x plan... although I can tell you i got the 20x plan because I used to spend the 20x pro codex plan with Astra in a single day...now with Opus 5.5 it seems like it will last me almost the whole week with same workload....so maybe reset is not even needed