To The Moon

$iShares Bitcoin Trust(IBIT)$  

Prices to the US close on Fri 30 Oct 2026, the latest available.

Over the last three months, Bitcoin rose 21.6%. Over the same three months, an equal-weight basket of nine AI-pivoting miners (IREN, Cipher, TeraWulf, Core Scientific, Hut 8, MARA, CleanSpark, Bitdeer, Applied Digital) fell 46.2%.

If these were still Bitcoin stocks, that gap should not exist. So what are they now? Our answer, from the filings and from our own return data: they trade as AI-credit stocks. CoreWeave, not Bitcoin, sets their daily direction.

1) What the miners actually signed

The seven miners we could check in filings have about 3.9 GW of AI and HPC capacity under contract. The chart lists each lease with its customer, value and start date.

Core Scientific leads with about 1.1 GW. Of that, 590 MW is leased to CoreWeave and the rest to AMD or a neocloud with AMD credit support.

Hut 8 has 949 MW, all leased to or backstopped by investment-grade names, worth $26.6B over the base terms. Part of it is leased to Fluidstack with a Google backstop.

TeraWulf signed a 20-year Anthropic lease for 401 MW, worth about $19B. It also hosts Fluidstack at Lake Mariner, again with Google behind the rent.

Cipher leases 300 MW (gross) to AWS for 15 years, worth about $5.5B. Its Barber Lake site for Fluidstack is now contracted for 20 years, with a leading AI lab taking the second decade.

IREN runs GPU clouds instead of leasing shells. Its Microsoft deal is about $9.7B over five years on 200 MW; Nvidia added $3.4B more.

CleanSpark signed 175 MW for 20 years, with deliveries from Q4 2027. Bitdeer signed a 16-year lease in Norway worth about $4.7B. MARA had not disclosed a signed AI lease in its filings.

The timing matters more than the totals. Much of this rent starts in 2027 and 2028, so the market is paying today for buildings that bill later.

2) How big is that, really?

Headline lease totals mislead because they add up 15 to 20 years of rent. A fairer yardstick is the yearly run-rate. Dividing each disclosed base-term value by its term gives about $8.5B a year across the seven miners at full delivery. Math is mine; inputs are in the sources comment.

CoreWeave's own revenue is running at about $10.3B a year ($2,575M in Q2 times four). So the miners' contracted book, once built, is close to one CoreWeave in yearly size. Two cautions: most of it is not billing yet, and IREN's figure is GPU-cloud revenue, not rent.

The tighter link is CoreWeave itself. About 41% of CoreWeave's roughly 3.7 GW of contracted power sits on ex-miner and hosting sites (Core Scientific, Applied Digital, Galaxy).

Those three sites hold 590 MW at Core Scientific, 400 MW at Applied Digital and 526 MW at Galaxy's Helios campus.

Why miners convert: a long AI lease pays a fixed rent with yearly escalators and no halving, while mining revenue moves with Bitcoin and network difficulty. Hashprice is near the lows of its recent series.

3) Who is pulling whom? Our own data

We took daily returns for the last 126 trading days. The miner basket's correlation with CoreWeave is 0.72. With Nvidia it is 0.53. With Bitcoin it is 0.40.

Put another way, CoreWeave alone explains 52% of the basket's daily moves. Nvidia explains 28% and Bitcoin 16%. A year ago the link was loose: the 63-day correlation with CoreWeave was 0.27 on 1 Oct 2025.

Crash days tell the same story. Since 1 Jul there were 19 days when the basket fell more than 5%. On those days Bitcoin averaged -1.0%, while CoreWeave averaged -7.6%.

What the data does not show: we found no lead or lag. Over one to five days, neither CoreWeave nor Bitcoin moves first in a usable way. So the honest read is a shared driver on the same day, AI credit, not miners dragging CoreWeave or the other way round. One caveat: Bitcoin's daily close is struck at midnight UTC, after the US close.

4) What caused the drop, ranked by evidence

Strongest: the repricing of AI debt. On 14 Oct, CoreWeave fell 15.5% and the miner basket 11.0%, while Bitcoin fell 1.8%.

That same day, CNBC reported the yield on CoreWeave's 2032 notes at about 14.8%. The bond market, not the Bitcoin market, was sending the signal.

Strong, but only for the build-up: rates. The 10-year Treasury yield rose from 4.44% on 30 Jun to 5.29% on 30 Sep.

US high-yield bond yields went from 6.99% to 8.16% over the same months, raising the cost of every new build.

Yet in October the 10-year eased to 5.05% (28 Oct) while the basket fell 34.9% in a month. Rates loaded the gun; credit fear about AI borrowers pulled the trigger. That is my read, not a proven cause.

Medium: worry about AI capex and its returns, with Oracle as the visible case. Oracle is 48.6% below its 12-month high, and its correlation with the basket is 0.49.

Weak for this drop: Bitcoin itself. Over three months Bitcoin rose while the basket fell by almost half. Low hashprice squeezes the mining cash that helps fund builds, but it does not explain the timing.

Not yet tested: share supply from convertibles and at-the-market programs, and lock-up dates. We have not run a dated study, so we make no claim.

5) The cost of money is the real scoreboard

Within CoreWeave, the oldest GPU loan costs about 15% and the newest non-recourse one about 7%. Lambda borrowed at 6.78% through an A(low)-rated contract vehicle.

Miners show the same pattern. Core Scientific replaced a term loan with $3.3B of 7.75% secured notes due 2031. Lenders price the tenant's credit, not the landlord's story.

That is why a Google-backed or investment-grade lease can borrow at a lower rate than CoreWeave's own bonds. It is also why the market lumping every miner together may not last.

6) Depreciation: whose problem is it?

If CoreWeave went back from 6-year to 5-year lives on its $33.8B of technology equipment, depreciation would rise by about $1.13B a year. Math is mine: one-fifth minus one-sixth of the gross cost.

That is about 10.9% of CoreWeave's annual run-rate revenue. Real, but not fatal on its own.

Landlord miners such as Core Scientific, Hut 8 and TeraWulf do not own the GPUs, so this risk sits mainly with CoreWeave and with IREN, which buys its own chips. Price action has treated them the same anyway.

7) How long could this last?

Base case, my view: one to two quarters of range trading while the market sorts names by tenant credit and delivery dates. The Q3 prints over the next two weeks start that sorting.

Better case: a few weeks, if CoreWeave shows new borrowing at or below 9% on 11 Nov and high-yield yields fall back toward 7.5%.

Worse case: two to three quarters or more, if a tenant delays, CoreWeave's bond yields push past 15%, or weak hashprice forces miners to raise equity at low prices.

The case against this view

The leases are long, many tenants are investment-grade or Google-backed, and new data-centre space remains tight. The assets may be fine and only the valuation reset. Our correlation may also reflect fund flows that fade once rent starts. And 126 days is one market regime. I could be wrong because the next six months may look nothing like the last six.

Dated tests (SGT)

Cipher reports tonight, Mon 2 Nov, call at 21:00 SGT. Watch: Barber Lake first rent in Q4 as promised, any cost overruns, and any new lease.

Hut 8 on Wed 4 Nov (21:30), Core Scientific on Thu 5 Nov (21:30), IREN on Fri 6 Nov (06:00) and TeraWulf on Fri 6 Nov (21:00). Watch: MW energised, rent start dates and funding costs.

Bitdeer on Tue 10 Nov (21:00), CoreWeave on Wed 11 Nov (06:00) and Nebius on Thu 12 Nov (21:00). The signal: CoreWeave's rate on new borrowing at or below 9%.

The decoupling test: if the 63-day correlation between the miner basket and CoreWeave falls below 0.5, the market is finally pricing each landlord's tenant on its own.

After reading this you can: check whether a "Bitcoin stock" still trades on Bitcoin with one correlation and one R², and turn a multi-decade lease total into a yearly number you can compare.

Question for you: which do you trust more as the turn signal, a falling CoreWeave bond yield or the miners decoupling from CoreWeave?

Not advice. I may hold positions in the names mentioned.

Stop guessing. Start calculating.

Live to fight another day. 🤙

#options #tradingpsychology #riskmanagement #optionstrading