leschjgen

Bill Ackman Quant
AI, Stocks, Token
NFA, DYOR

[leschjgen @0xgentileschi]

https://t.co/OhYWS6erzr

[leschjgen @0xgentileschi]

What I'm watching: Galaxy's own tripwire is a weekly close under the 50-week average ($76,540) not regained within a month. BTC is ~$79,700, so that line is ~4% away (our calc).

This spell lasted 44 weeks. Galaxy discloses it holds BTC. Source: Galaxy via The Block, Oct 26

[leschjgen @0xgentileschi]

Galaxy's "bitcoin bear market is over" call rests on 4 data points.

Since 2011, BTC reclaimed its 50-week average after 20+ weeks below it 4 times. All 4 were higher a year later.

The other 9 reclaims, after shorter spells? 7 fell 25%+ within a year.

[leschjgen @0xgentileschi]

Sources: Jul 28 guide ($4.1B, ~50% non-GAAP op margin, ~$300M opex) vs Sep qtr 2025 ($2.629B revenue, $763M non-GAAP op income, $291M opex). My calc.

Risk: Seagate credits pricing as well as HAMR, and pricing can turn. Consensus EPS (~$7.42) is already above the $7.30 midpoint.

[leschjgen @0xgentileschi]

Seagate's own guide for tonight: revenue +56% vs a year ago. Operating profit ~2.7x.

My math: ~$1.47B of extra revenue, ~$1.29B of extra operating profit. About 87 cents of every new dollar, with opex near flat.

Tonight I'm watching if the $STX December guide keeps that going.

[leschjgen @0xgentileschi]

8-K, Oct 26: rent paid quarterly in advance, 1/20 of each tranche's cost plus fixed-rate interest. With all $40B live: ~$2B of principal a quarter from H1 2027.

Counterpoint: $53.4B cash (incl. restricted) at Sep 30 after the $85.7B IPO. Capex also funds rockets and Starlink.

[leschjgen @0xgentileschi]

SpaceX's CFO says its AI compute pays back in under a year. Yet it just lined up ~$40B of Nvidia GPUs on a ~5-year lease.

The cash flow statement explains why. Jan-Sep: $48.1B of capex vs $13.1B of operating cash. Capex ran ~3.7x what $SPCX's operations brought in (my calc).

[leschjgen @0xgentileschi]

@gem_detecter on RH maybe Darkwoods and lighthouse storage

[leschjgen @0xgentileschi]

Our math from the Oct 26 filings: Strategy's 8-K (week of Oct 19-25) and BitMine's update (2.04M shares at $22.48 avg, ETH at its $2,356 reference).

Not everyone pulled back: Strive bought 6,182 BTC at ~$82.7K avg on Oct 17-23, roughly $511M.

[leschjgen @0xgentileschi]

Last week the biggest BTC and ETH treasury companies spent ~$251M buying back their own paper and ~$25M buying coins.

Strategy: $204.9M on STRC buybacks, zero BTC.
BitMine: $45.9M on its own stock, 10,452 ETH (~$24.6M).

Discipline, or the flywheel running in reverse?

[leschjgen @0xgentileschi]

Q4 guide is $1.9-2.0B, but Amkor sees computing up low double digits q/q, over 50% y/y. Communications falls high single digits again.

Materials dropped from 57.5% to 50.7% of sales in a year.

Risk: capex steps up in 2027 (2026: ~$2.5B).

Source: Amkor Q3 release + call, Oct 26

[leschjgen @0xgentileschi]

Amkor's revenue was flat vs a year ago: $1.991B vs $1.987B. Gross profit rose ~28% (my calc: $364M vs ~$284M).

The mix flipped. Computing went from 19% to 26% of sales, up ~37% in dollars. Phones and tablets: 51% to 37%.

$AMKR missed the headline, then rose ~5% after hours.

[leschjgen @0xgentileschi]

Only the $12.5B of preferred is cash at closing. The $22.5B of common is a forward sale settling by Nov 19, 2027, and the $60B ATM can't start before Q1 2027.

Meanwhile the share count already rose this year: 2,530M to 2,556M.

Sources: Meta 10-K/10-Q, Oct 26 prospectus

[leschjgen @0xgentileschi]

Meta spent ~$149B buying back its own stock from 2021 through 2025 (my sum of its filings). 2026 so far: $0.

Now it plans to sell up to $95B of new shares.

The 2022 batch cost ~$174 a share. 2025's: ~$653. $META closed at $642.62 before the deal.

[leschjgen @0xgentileschi]

Data: CoinDesk Oct 21 (open interest via SoSoValue), Yahoo daily closes ($544 Oct 1, $591 Oct 26).

It doesn't have to repeat. But since the Oct 22 wick to $655, XMR has closed between $584 and $616 five days running. Leverage building into a stall is what I watch.

[leschjgen @0xgentileschi]

Monero is up ~9% in October (my calc). Its futures grew faster: last week open interest topped $300M, near January's ~$320M record, and went from ~400K to 516K coins in seven days.

January's setup ended with $XMR going from ~$790 to under $300.

Privacy bid, or leverage again?

[leschjgen @0xgentileschi]

The deal: a $63B SPV buys Broadcom XPUs and networking and leases them to Anthropic. 1.7 GW, deliveries from 2027. My math: ~$37B per GW.

The exposure amortizes to zero over 5 years of lease payments.

I'm watching how fast 2027 accelerators lose resale value vs that schedule.

[leschjgen @0xgentileschi]

Broadcom now puts its own peak exposure at ~$43B if Anthropic's AI racks lose value. That's ~3 quarters of $AVGO free cash flow at last quarter's $13.7B.

It pays only if the racks end up worth less than 72.5% of what's still owed on the lease.

[leschjgen @0xgentileschi]

The claim: a deterministic algorithm that exactly counts satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space. 3-SAT is the case that breaks ETH.

Subexponential isn't polynomial, so not P=NP.

Checkers: two of ETH's original authors.

[leschjgen @0xgentileschi]

OpenAI's ETH disproof sits in a repo of 721 AI-written math manuscripts. On Oct 7, that repo withdrew 3 of them over a single sign error and repaired 14 more.

So the line I'd weigh most: the ETH paper comes with a Lean proof. Only ~42% of the repo's top results do.

[leschjgen @0xgentileschi]

Who's taking them: Bitdeer's self-mining hit 82.3 EH/s in September vs 35 a year earlier, ~8% of a 1 ZH/s network alone (my calc). CryptoQuant also points to private and non-US miners.

Difficulty is tracking another ~5% rise this week. Hashprice keeps sliding for whoever stays.

[leschjgen @0xgentileschi]

Five bitcoin miners pivoting to AI, incl. IREN and Core Scientific, mined 7,075 $BTC this year through Oct 19. Same stretch of 2025: 13,263.

Hashrate is back above 1 ZH/s anyway.

Issuance doesn't shrink when miners leave. Those 6,188 coins just went to someone else.