$CIFR: Eleven Billion In Leases, Nineteen Million In Revenue, And The One Asset Nobody Else Can Copy
The most hyped stock on fintwit is a ten billion dollar company at an all time high. Strip the rocket emojis and underneath sits the one thing I actually hunt, a physical powerhouse.
Position disclosure: I have no position in CIFR and have never owned it. Shawarma Capital, June 25, 2026. CIFR is $26.22 against Tuesday's close, off 5.1 percent on the session and roughly 12 percent below the all-time high it printed last week. This post is research synthesis, not investment advice. The full disclaimer is at the bottom.
Open the $CIFR feed on X right now and count the rocket emojis. I did. Bernstein just slapped a Buy on it and called the company a "power landlord of AI." A fund manager is bragging about being up seventy-four million dollars on the position. Tier lists rank it in the neocloud cohort. Somebody is posting a Gann angle that unlocks $64. The stock is up something like 600 percent in a year and it is sitting at an all-time high.
This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber.
This is the single most-hyped name on finance Twitter. It is a ten-billion-dollar company. It breaks two of the three rules I run this book on: it is not sub-$1 billion, and it is the opposite of undiscovered.
So why am I writing 4,000 words on it.
Because underneath the rocket emojis there is exactly one thing I actually hunt for a living. A physical chokepoint. Cipher is not an "AI stock." It is a landlord that spent four years quietly assembling energized, grid-interconnected power in West Texas, the single scarcest input in the entire artificial-intelligence buildout, and then signed that power away to Amazon and a Google-backed tenant for $11.4 billion of contracted lease revenue before it had built most of the buildings.
That part is real. The question this post answers is which part is real, which part is a press release, and whether the gap between the two is priced into a $26 stock at an all-time high.
Let's go.
I. The Two Rules This Breaks
I am going to be honest about why this name does not look like anything else I cover, because the honesty is the entire reason to read me instead of the emoji accounts.
Rule one I break: size. My edge is sub-$1 billion names that no sell-side desk covers. Cipher's market cap is $10.7 billion against 409 million shares outstanding at $26.22. There are now multiple analysts on it. Bernstein, the firm that wrote the "power landlords of AI" note, is not a microcap shop. This is a mid-cap with institutional sponsorship. The information asymmetry I usually trade is mostly gone.
Rule two I break: the crowd. I like names the tape has not found yet. The tape found this one. It ran from roughly $5 in the spring to a $30 print last week. Every momentum account on X is long. When I wrote about AmpliTech at $5.04 a month ago, the whole point was that nobody trusted the print. Here, everybody trusts the print. That is a different, and more dangerous, setup.
What I do not break is rule three, the one that matters most. The asset has to be a physical chokepoint with a multi-year catalyst window and a moat a competitor cannot simply copy with capital. On that test, Cipher passes harder than almost anything I own. That tension, a name that fails two of my filters and aces the third, is exactly why it is worth a Part 1. I am not here to tell you to buy a ten-bagger at the top. I am here to tell you what the asset actually is, so that when it pulls back, and a name that has run this hard always pulls back, you already know whether you want it.
II. What Cipher Actually Is Now
Cipher Mining was born as the US bitcoin-mining subsidiary of Bitfury Group. It went public in August 2021 through a merger with Good Works Acquisition Corp, a SPAC, at a roughly $2 billion valuation. Tyler Page has been CEO the entire time. For its first three years it was what every bitcoin miner is: a company that converts cheap electricity into hashrate, sells the bitcoin, and dilutes shareholders to fund the next fleet of machines.
In February 2026 the company changed its name from Cipher Mining to Cipher Digital. The rename is not cosmetic. It marks the pivot that is the whole thesis. Cipher is converting itself from a bitcoin miner that happens to own power into a data-center landlord that happens to still mine bitcoin.
The mechanism is simple to state and brutally hard to execute. Cipher spent years acquiring land and, more importantly, securing large-scale grid interconnection capacity in the ERCOT market in West Texas. Interconnection is the bottleneck. Anyone can buy land. Getting a utility to commit hundreds of megawatts, or a gigawatt, of deliverable grid power to a specific site, on a specific energization date, is a multi-year regulatory and engineering slog that money alone does not shortcut. Cipher did that slog while it looked like a sleepy miner. Now it is renting the result to hyperscalers who need power yesterday and cannot wait three years for their own interconnect queue.
That is the company. A power-interconnection portfolio with a bitcoin mine still running on part of it and AI data centers being built on the rest.
III. The Power
Everything in this thesis reduces to one number that is hard to fake: megawatts of energized, grid-connected power.
Cipher's stated portfolio is approximately 4.2 gigawatts of grid power across operating, contracted, and pipeline sites, with management guiding total capacity toward that 4.2 GW figure by 2030 and beyond. That number is the headline. It is also the number to be most careful with, because "pipeline" is doing a lot of work inside it.
Here is the honest breakdown of where the power actually is today.
Read that table twice. Of the 4.2 gigawatts everyone quotes, roughly 207 megawatts is energized and producing revenue right now, and it is producing that revenue from bitcoin, not AI. The 600 megawatts under contract at Barber Lake and Black Pearl is real, financed, and being built, but it is not yet generating a dollar of rent. Everything past Stingray is pipeline: land and interconnection rights on a 2028-2029 clock, dependent on the same grid queue, supply chain, and capital markets that could slip a year in either direction.
The power is the moat. The energization date is the risk. Hold both of those in your head for the rest of this post.
IV. The Leases
This is where Cipher separated itself from the bitcoin-miner pack. It signed long-dated, large-dollar colocation leases with credible AI tenants and got paid to build.
The Fluidstack deal, announced September 2025 and expanded in November, fills the entire 300 MW Barber Lake site under a 10-year hosting agreement. Total contracted revenue across the Fluidstack relationship runs to roughly $3.8 billion over the initial term, with extension options that management frames as up to $7 billion to $9 billion if exercised. The critical feature: Google is a strategic backer. Google agreed to backstop a portion of Fluidstack's lease obligations, $1.4 billion initially plus an additional $333 million on the expansion, and in exchange received warrants for roughly 24 million Cipher shares, about 5.4 percent pro forma. A trillion-dollar balance sheet standing behind part of the rent, and taking equity to do it, is the strongest validation a former bitcoin miner could buy.
The AWS deal, announced November 2025, is the anchor. Amazon Web Services signed a 15-year lease for 300 MW of gross capacity at the Black Pearl site in Wink, Texas, valued at approximately $5.5 billion. It phases in beginning July 2026 with rent effective August 2026, and Cipher provides both air and liquid cooling. AWS is investment grade. AWS is the tenant that makes the whole portfolio bankable.
A third campus lease with an "investment-grade hyperscale tenant," disclosed on the Q1 2026 call, plus the Stingray financing, brings management's total contracted revenue figure to approximately $11.4 billion. Bernstein, counting extension options and the broader pipeline, frames the order book at roughly $24 billion.
So the lease stack, stated plainly.
That is a genuinely impressive book for a company that was a sub-$1 billion miner two years ago. It is also where the skepticism has to start.
V. The Gap
Here is the number the rocket emojis never put next to the $11.4 billion.
Bernstein, the bull, models Cipher's AI revenue at $19 million in 2026, rising to $1.2 billion by 2030. Nineteen million dollars. This year. Against $11.4 billion of contracted lease value and a $10.7 billion market cap.
That is not a contradiction. It is the entire structure of the trade, and you have to understand it or you will misprice the stock in either direction.
The leases are signed. The rent has not started. AWS rent turns on in August 2026. The Fluidstack site at Barber Lake energizes through 2026 and into 2027. Until the buildings are energized and the tenants move load in, the $11.4 billion is a contract, not a cash flow. 2026 total company revenue is consensus-modeled around $228 million, and almost all of that is still bitcoin mining, the business Cipher is actively winding down. The AI revenue line is a rounding error this year on purpose, because the sites are under construction.
The whole company is a bet on closing that gap on time. If the sites energize on schedule and the tenants take the power, the revenue ramp from $19 million to north of a billion over four years is one of the steepest in the public market, at lease-economics margins Bernstein pegs near 93 percent EBITDA because Cipher is the landlord, not the GPU operator. If the sites slip, every quarter of delay is a quarter of that contracted book pushed right while the stock trades on the assumption it arrives on time.
The bulls are pricing the $11.4 billion. The business is delivering the $19 million. Part 1's job is to tell you that both numbers are true.
VI. The Capital Structure
The most underappreciated thing Cipher did is how it financed the build without torching shareholders.
Bitcoin miners dilute. That is the genre. Cipher itself went from a low-nine-figure share count at the SPAC to 409 million shares today, much of it raised by selling stock to buy machines. The pivot changed the funding model. Instead of issuing equity to build the AI data centers, Cipher financed them with non-recourse project debt secured against the contracted leases themselves.
Black Pearl Compute LLC priced $2.0 billion of 6.125 percent senior secured notes due 2031 in February 2026 to fund the AWS site, plus a $200 million revolving credit facility from a syndicate of global banks. The Barber Lake project carries roughly $1.7 billion of its own project debt. A separate ~$810 million senior secured note financing supports the Stingray/AWS capacity. Total principal long-term borrowings sat around $5.2 billion at the end of Q1 2026, the large majority of it non-recourse and ring-fenced at the project level, against roughly $4.25 billion of cash, equivalents, and restricted cash.
Why this matters: non-recourse project debt means the lender's claim is on the specific data center and its lease, not on the parent. If a project underperforms, the damage is contained to that subsidiary. The parent's equity is not on the hook for the full $5.2 billion. And because the build is debt-funded against investment-grade rent, Cipher is not issuing hundreds of millions of new shares to construct these sites. The dilution that defines the bitcoin-mining sector is, for the AI buildout specifically, largely absent. The Google warrants for ~24 million shares are the notable equity issuance, and Cipher got a Google backstop in exchange.
That is the "masterstroke" the more careful bulls are pointing at. A miner that learned to use its contracted cash flows as collateral instead of using its stock price as an ATM. It is genuinely well-engineered. It also means the company now carries real , and is fine until an energization date slips and a debt-service reserve gets tested.
VII. The Counterparty Problem
The bull case leans on the phrase "investment-grade hyperscaler tenants." Half of that is precise. Half of it is marketing. The distinction is worth real money.
AWS is investment grade. The $5.5 billion Black Pearl lease is as close to a sovereign-quality rent stream as exists in this sector. No argument.
Fluidstack is not investment grade. Fluidstack is a neocloud startup, founded by Gary Wu and Cesar Maklary, that hit roughly $180 million of annualized recurring revenue in late 2024 and is reportedly raising around $700 million at a $7 billion valuation, with Situational Awareness, Leopold Aschenbrenner's firm, in the lead. It manages over 100,000 GPUs and serves AI labs like Meta, Mistral, and Character.AI. It is a real, fast-growing company. It is also a venture-backed business assembling one of the most complex financing structures in the sector, a Macquarie facility sized up to $10 billion, a few hundred million of equity, and Google backstops layered across multiple landlords, Cipher and Hut 8 and TeraWulf among them.
The Google backstop is what makes the Fluidstack rent bankable, and the backstop is partial, not total. Google guarantees a defined slice of the obligation to support the project debt. It does not guarantee the full 10-year, multi-billion-dollar lease. If the AI capex cycle cools, the entity standing behind most of Cipher's Barber Lake rent is a startup whose own survival depends on the same cycle, with a trillion-dollar partner backstopping part, but only part, of the bill.
This is not a reason to dismiss the name. It is a reason to weight the $11.4 billion correctly. The AWS portion is gold-plated. The Fluidstack portion is a high-growth startup wearing a Google-shaped life jacket that covers the torso but not the legs. Treat them differently when you price the book.
VIII. The Bitcoin Engine Being Dismantled
The part of Cipher that actually makes money today is the part it is taking apart.
Odessa runs 207 megawatts of bitcoin mining at roughly 11.6 exahash per second and 17.2 joules per terahash of fleet efficiency. In Q1 2026 it mined 346 bitcoin. The company also holds a treasury of about 1,808 bitcoin and 3,952 ether as of the end of Q1. Q1 revenue was $35 million, down from $60 million the prior quarter, precisely because Cipher is winding down mining at Black Pearl to hand that power to AWS. The net loss was $114.3 million and adjusted EBITDA was negative $48.2 million, both reflecting a company spending heavily to build the AI sites while its only live cash engine shrinks.
That is the awkward middle. For the next several quarters, Cipher is a company whose reported financials get worse, not better, as the high-margin future revenue is still under construction and the legacy revenue is being deliberately switched off. Anyone buying the headline numbers in isolation is buying a company with a falling top line and a widening loss. Anyone buying the thesis is buying the moment the AWS and Fluidstack rent switches on and the income statement inverts. The earnings prints between here and mid-2027 will look ugly to a screen and fine to a model. That divergence is where the volatility lives.
IX. The Comp Set
Cipher does not trade in isolation. It trades inside the neocloud cohort, and the cohort is not one thing.
There are two distinct business models wearing the same "AI infrastructure" label. There are GPU operators, companies like CoreWeave, that buy the chips, take technology and obsolescence risk, and sell compute. And there are power landlords, companies like Cipher, TeraWulf, IREN, Hut 8, and Core Scientific, that own the power and the building and rent it to someone else who takes the chip risk. Bernstein initiated Cipher and TeraWulf together precisely because they are the same model: capital-light at the chip layer, capital-intense at the power layer, paid in long-dated contracted rent.
The landlord model is the safer half of the trade. The landlord does not own a single GPU that depreciates to zero in four years. The landlord owns interconnected power, which only gets scarcer. When you hear "neocloud," separate the ones that own silicon from the ones that own substations. Cipher owns substations. Within the cohort, that is the side I would rather be on, and it is the cleanest argument for the name.
The comparison that should keep a buyer honest is the rest of the cohort has run just as hard. IREN, WULF, CORZ, and the others have all repriced violently off the AI-landlord narrative. This is a crowded, correlated trade. On a sector drawdown, they sell off together regardless of which one signed the best lease.
X. The Bear Case
I give the bear case more room than the bull case in every Part 1, and this name has earned a long one.
One. Execution and energization risk is the whole ballgame. The $11.4 billion converts to cash only if Barber Lake, Black Pearl, and Stingray energize and accept load on schedule through 2026 and 2027. Grid interconnection dates slip. Transformer and switchgear lead times are measured in quarters. Liquid-cooling buildouts at this scale are new. Every quarter of slippage pushes the revenue ramp right against a stock priced for it to arrive on time. This is the single load-bearing risk and it is not in management's full control.
Two. Counterparty concentration. Strip out AWS and a large share of the contracted book sits behind Fluidstack, a venture-stage neocloud whose rent is only partially Google-backstopped. If the AI capex cycle cools or Fluidstack's own financing stack wobbles, the Barber Lake revenue is the first thing markets will discount.
Three. The reported financials get worse before they get better. Falling revenue, widening losses, negative EBITDA, and heavy capex for several more quarters. A macro risk-off that hits before the AWS rent turns on in August 2026 would hit a company that, on a screen, looks like a melting bitcoin miner.
Four. . $5.2 billion of project debt is non-recourse and well-structured, but it is still $5.2 billion against a company earning negative EBITDA today. Debt-service reserves are funded on the assumption the sites energize and pay. A delayed energization tests those reserves. Non-recourse limits the blast radius. It does not make the free.
Five. Valuation and crowding. The stock is at an all-time high, up roughly 600 percent in a year, the most-posted ticker in its cohort, with a fund manager publicly bragging about the gain. Bernstein's own $32 target implies only about 22 percent upside from here, and that is the Street's bull. When the marginal buyer is a momentum account and the marginal analyst target is 22 percent away, the asymmetry is no longer in your favor at this entry. The thing trades like everyone already owns it because, on FinTwit, everyone does.
Six. Bitcoin. Cipher still carries a bitcoin and ether treasury and still runs Odessa. A sharp crypto drawdown dents the one cash engine that is actually live and reminds the market of the genre Cipher is trying to leave.
Seven. Dilution history. This management team raised equity aggressively as a miner. The non-recourse pivot is a real change in behavior, but the track record says watch the share count, watch the ATM, and watch any convertible issuance the moment the project-debt market tightens.
None of these are theses-killers on their own. Together they say the same thing: this is a well-built asset at a fully-priced, fully-crowded entry, with the one risk that matters, energization timing, sitting outside management's hands.
XI. Scenario Math
I do not hand-tune probability weights for a Part 1. I give you the frame and let you weight it.
The market cap is $10.7 billion. The contracted book is $11.4 billion of gross lease revenue over 10-to-15-year terms. At Bernstein's ~93 percent landlord EBITDA margins, the steady-state economics, once fully energized, are a multi-hundred-million-dollar annual EBITDA stream from the first three campuses alone, with the 2.5-plus gigawatts of pipeline as free options on top. On 2030 AI revenue of $1.2 billion at landlord margins, today's $10.7 billion cap is not obviously expensive. That is the bull's math and it is not crazy.
The bear's math is simpler. You are paying $10.7 billion today for $19 million of 2026 AI revenue and a promise that the rest arrives on a construction schedule. The Street's most bullish target, Bernstein's $32, is 22 percent above the print. The average analyst target sits around $32 to $33. You are buying near the analyst ceiling, not the floor.
The honest synthesis: the long-term value is plausibly much higher than $26 if execution lands, and the next 12 months of entry-price asymmetry are poor because the stock has already priced a clean energization. The way you make money on this name is not by chasing the all-time high. It is by owning the asset through a cohort drawdown that resets the entry while leaving the leases intact. The leases do not care what the stock did last week.
XII. The Catalyst Calendar
The dates that move this name from here.
Q2 2026 earnings, August 6, 2026. The first print that should show AWS rent beginning to flow and the first read on whether Black Pearl phase one energized on time.
AWS rent commencement, August 2026. The single most important operational milestone. Rent turning on at Black Pearl is the moment the income statement starts to invert.
Stingray energization, Q2 2026. The 100 MW front-of-the-meter site is the near-term proof that Cipher can bring contracted power online on schedule.
Barber Lake energization through 2026 into 2027. The Fluidstack revenue ramp.
Colchis and the 2028-2029 pipeline. Direct-connect with American Electric Power on the 1 GW JV, the option value on the back half of the 4.2 GW.
The pattern to watch is simple. Every energization that lands on time validates the model and de-risks the gap. Every one that slips a quarter does the opposite. The thesis is mostly a schedule.
XIII. Kill Criteria
What would make this a closed thesis rather than a watch item, if I owned it.
One. A disclosed energization slip of two or more quarters on Black Pearl or Barber Lake. That converts a financing-and-construction story into a broken-schedule story.
Two. Any public sign of distress at Fluidstack, a down round, a financing failure, a lease renegotiation, or a withdrawal of any portion of the Google backstop.
Three. AWS pausing, descoping, or delaying the Black Pearl phase-in. The anchor tenant is the load-bearing wall.
Four. A new equity raise or convertible issuance above a few hundred million at a depressed price, signaling the non-recourse project-debt market has closed to them and the old dilution reflex is back.
Five. A debt-service reserve draw or covenant waiver on any project entity.
Six. A bitcoin drawdown deep enough to impair the treasury and force Odessa offline before the AI rent replaces the mining cash flow.
None of those are tripped today. The first testable one is the August 6 print and the AWS rent commencement in the same month.
XIV. Why I Have No Position
I will tell you plainly. I do not own this and I am not buying it here.
I missed it. It ran from $5 to $30 while I was deep in defense autonomy and compound semis, and I am not going to pretend a ten-bagger at an all-time high is a fresh idea. Chasing a 600 percent move into the analyst ceiling, in the most crowded name in its cohort, is the exact behavior I built my rules to prevent. It breaks my size mandate and it breaks my crowding mandate, and "but the asset is great" is what every top-tick buyer tells themselves.
What would change my mind is not a higher price. It is a lower one with the leases intact. If a cohort-wide drawdown resets CIFR back toward the high teens while AWS rent is turning on and Black Pearl is energizing on schedule, then the chokepoint, energized ERCOT power rented to investment-grade tenants on 10-to-15-year terms, is exactly the kind of asset I want to own for years. At that point the size rule still fails, but the asymmetry comes back, and I would size a position and say so in a Part 2.
Until then I am writing it, not buying it. That is the difference between research and a rocket emoji.
XV. Bottom Line
Cipher Digital is the best-engineered version of the bitcoin-miner-to-AI-landlord pivot in the public market. It spent four years securing the one input nobody can shortcut, energized grid power in West Texas, and it rented that power to Amazon and a Google-backed neocloud for $11.4 billion of contracted, mostly non-recourse-financed lease revenue without torching its share count to do it. The asset is real. The moat is power interconnection, and it is the hardest moat in this entire sector to copy.
The stock is a different question than the asset. It is a $10.7 billion company collecting $19 million of AI revenue in 2026, priced for a flawless energization that has not happened yet, trading at an all-time high inside the most crowded trade on finance Twitter, with its most bullish analyst target only 22 percent away. The contracted book is gold-plated where AWS sits and merely promising where Fluidstack sits. The financials get worse before they invert. The whole thesis is a construction schedule, and the schedule is the one thing management cannot fully control.
The bulls are pricing $11.4 billion. The business is delivering $19 million. Both numbers are true. The gap between them is closed by energization dates in the back half of 2026 and through 2027, and the first one prints on August 6.
I have no position. I am watching the schedule, not the chart. If the cohort breaks and the leases hold, I will have a lot more to say.
Shawarma Capital
I am long the names disclosed in my portfolio. I have no position in CIFR. This is research synthesis, not investment advice. You should not buy or sell securities based on anything I write. I am not a registered investment advisor. I do not owe you a fiduciary duty. My conclusions could be wrong in ways I have not anticipated. Financial projections are model outputs based on publicly available data. They are not guarantees. These names are volatile and can lose 50 percent or more in a single session. Do your own due diligence. Verify every number against primary sources.