Keel Infrastructure

IThe business

Keel Infrastructure Corp. is a company in the middle of changing what it is. Until April 2026 it was Bitfarms, a bitcoin miner founded in 2017 with data centers in Québec, Argentina, Paraguay and, after a March 2025 purchase of Stronghold Digital, Pennsylvania.² It sold or abandoned its Latin American sites, and on 29 June 2026 it stopped mining in the United States.³ It keeps a small mining operation in Québec for now. Its name changed on 1 April 2026, when it moved its home from Ontario to Delaware, and it now files the same reports as any American company.³

What it wants to be is a landlord for AI. The plan in the 10-K is plain. Keel owns or controls power, land and grid connections. It will build data centers on them, lease the finished space to a tenant such as a cloud company, an AI lab or a large technology company, and collect rent for many years.² It says it is an infrastructure developer and not a cloud company: it will not own chips or offer computing, only the building and the electricity.² The pipeline it quotes is 2.2 GW, made of 648 MW it calls secured and 1,513 MW planned.² A gigawatt is a billion watts, about what a large nuclear reactor makes.

The sentence that matters most on this page is in the same filing: Keel had not yet entered into any lease for an HPC data center.² The August 10-Q repeats it for the sites: as of 7 August 2026 Keel had not started data center operations at any of them or recognised revenue from one.³ So today Keel is a company with power sites, cash, debt and a plan, and with no tenant. Everything after that is a forecast.

The sites

Each site is at a different stage, so I list them one by one. Capacities are what the filings call gross megawatts, the power the site can draw. A data center sells the smaller figure that reaches the computers, called critical IT load. I use 70% to convert, which is my estimate: Galaxy’s Helios campus has 526 MW of critical load on 800 MW gross, 66%, and TeraWulf’s Cayuga site is planned at about 320 on 400, 80% E (estimate).⁵⁶

What the company earns while it waits

Not much, and less every quarter. In the second quarter of 2026 revenue was $30.4M, down 50%, and the cost of revenue was $117.2M, of which $84.1M was depreciation, most of it written off faster because the mining machines are being switched off.⁴ General and administrative cost was $31.3M, and the operating loss was $140.8M.⁴ The company’s own measure, Adjusted EBITDA, was −$23.7M, which strips out depreciation, stock pay and bitcoin price changes.⁴ In the first half the cash used in operations was $117.6M, about $0.19 for each share C (calculated).³

Bitcoin matters for two reasons. It paid the bills until now: in 2025 one mining pool operator accounted for 88% of revenue, and Keel ran 113,649 machines at 14.8 exahashes a second at the year’s end.² And it is on the balance sheet. At 7 August Keel held 1,861 bitcoin, $121M, which it is selling down as part of the wind-down, having sold 1,085 since 1 April for $75M.⁴ At the 30 October price of about $77,204 those coins are worth about $144M C (calculated).⁸ The stock did not follow bitcoin. Bitcoin rose 22% between 22 June and 30 October and Keel fell 69%.

How the money is meant to work

The economics of the plan are simple to say and hard to know. A tenant signs a lease of ten to twenty years for a number of megawatts. Keel builds the building and installs the power and cooling, unless the tenant pays for that, and collects rent. A lease that is signed lets Keel borrow against it at a cheaper rate, which the 10-K says directly: a signed lease is generally needed to get project-level debt at a lower cost.² The more of the build the tenant pays for, the lower the rent and the less capital Keel needs. The three published examples below show how wide that range is. They are the only evidence I have on price, and I come back to them in Valuation.

IIThe moat

The question here is what keeps a competitor from taking your customers at your price. I look in three places: scarce inputs, switching costs and scale. Keel has the first, in part, and none of the other two yet.

Scarce power

The case for Keel is that power is the constraint. Chief executive Ben Gagnon said it in the August release: “Power is the constraint.”⁴ The 10-K adds that the PJM grid operator’s 2025 plan implies that the lines needed to serve new data center demand could take more than five years to build, and that Grant County in Washington has capped data center load growth.² If that is true, a site with a signed supply contract is worth more than one without. Keel holds 350 MW of firm supply at Panther Creek under a contract with PPL, with 50 MW due by the end of 2026 and 300 MW by the end of 2027.² That is real, and it is the best thing the company owns.

I can put a rough price on the scarcity, from a transaction rather than a slogan. In May 2026 TeraWulf bought 308 acres in Grayson, Kentucky, with a power agreement for up to 1,000 MW, for $30M in cash, $50M a year later and up to $90M more as power is delivered, a total cost of $167.5M, about $167,000 for each megawatt C (calculated).⁶ That is the price of raw power and land in a deal I can read. Keel’s enterprise value at the close, market value plus notes less cash and bitcoin, is about $1,562M, or $2.4M for each of the 648 MW it calls secured C (calculated).¹³⁴ Even allowing that Keel’s sites have more done than raw land, the market is paying for more than power. It is paying for the chance to turn power into leased buildings, and that has to be earned.

Switching costs and scale

Neither exists. A data center tenant signs a long lease, so once it is in it stays, but that protects a landlord who already has a tenant. Keel has none, and it is competing for the first. Scale in this business is the ability to finance, build and deliver hundreds of megawatts on time. The leaders have shown it. Galaxy delivered 133 MW to CoreWeave on schedule in the second quarter of 2026.⁵ Keel’s only piece of delivered infrastructure is the first Vertiv modules at Moses Lake, a site of 18 MW gross.⁴

What it would take to be wrong about this

Two things would give Keel a defensible position. One is a lease with a very large tenant on terms that show the sites are worth more than others, because the tenant paid a premium for speed. The other is a record of delivering on time at Moses Lake and Sharon, which would make the next lease easier. I will look for both. The history of the industry says a developer’s edge fades quickly: every miner that went into this business bought or built the same things, and tenants can choose.

IIIThe competitive advantage

The last section asked whether the position can be defended. This section asks something narrower: against the companies Keel competes with for tenants, where does it stand today?

Where it is ahead

Its power is energized or contracted in markets tenants want. Three of its four United States sites are in PJM, the largest grid in North America, and the fourth, Moses Lake, sits in a data center cluster in central Washington with cheap hydropower.² It owns about 1,000 acres.² It already runs power plants at Panther Creek and Scrubgrass and has dealt with grid operators for years, which shortens its learning curve.² And its cooler climates allow free cooling for more of the year.²

Where it is behind

Almost everywhere else. The table sets Keel against three of the companies it will meet in tenant negotiations, with numbers I could open in filings dated October 2026 or earlier.

The point of the table is not that Keel is worse. It is that on 30 October the leased and delivered capacity was in other hands, and the tenants who have signed have mostly signed with the companies that already delivered. A tenant choosing Keel is taking a risk on a developer with no completed data center. Keel’s answer, in its words, is to arrive with power, permits and a delivery date, and to be the one that can say yes quickly. Whether that wins depends on how short tenants are of capacity in 2027.

The new president

One piece of evidence points the other way. On 6 July 2026 Keel hired Ganesh Aiyer as President to run commercial work and expansion. Before that he was chief business officer at Digital Realty, where he led commercial strategy for hyperscale, cloud and enterprise customers.¹⁰ A company that expected to wait would not hire that person. It is a signal about intent, not about results.

IVManagement

Ben Gagnon, 37, runs the company. He started at Bitfarms in 2019 as director of business development, became chief mining officer in 2021, and has been a director since 2024.² Before that he founded and ran two bitcoin mining businesses.² His training and his record are in power, land and bitcoin mining. He has not built or run a data center for a third-party tenant. That is true of almost everyone in the business of converting mining sites, and it is the central fact about this team.

How they are paid, and what that pays them to do

Gagnon’s reported pay for 2025 was $5.73M: salary $457,058, stock awards $4.04M, and a cash bonus of $1.15M.² The bonus is set on a mix of company and personal goals, including hash rate, cost, energized capacity growth, revenue, operating cash flow and “strategic milestones related to HPC/AI infrastructure development”, and for 2025 the committee scored Gagnon at 120% on his personal part.² The performance shares vest after three years and pay between nothing and double the target, depending on Keel’s share return against the average of the Russell 3000. Zero if Keel trails by more than ten percentage points, 200% if it beats the average by more than ten.²

What does that pay them to do? Two things. The bonus pays for energized capacity and development milestones, which Keel controls. It does not pay for leases signed, rent per megawatt or return on the capital spent, which are the outcomes that matter. And the share-return test rewards a rising stock over three years, almost whatever the reason. A stock that has fallen 69% in four months has pushed those awards toward zero, which could make the next financing, at any price, look unattractive to staff. I would watch for a change to the plan that adds a lease or return measure. The company has paid for hiring: the new president gets a $500,000 salary, a bonus target of 100% and $1.5M of restricted stock.¹⁰

How much of their own money is in

At 27 March 2026 Gagnon beneficially owned 3,441,770 shares, about $7.1M at the October close; the table counts options he can exercise within 60 days C (calculated).²¹ That is a little over a year of his 2025 pay. All directors and named officers together held 10,467,777 shares, under 2% of the company, about $21.6M C (calculated).² The 10-K lists no related-party deals since January 2025.² For Jeffrey Lucas, the former CFO, the same table shows 3,876,982 shares.² I read this as a team with a stake large enough to care and small enough that a dilutive financing does not threaten them as much as it does an outside holder. After the second-quarter report the two senior officers bought in the open market. Forms 4 show Gagnon buying 58,888 shares on 13 August and 38,888 on 21 August, both at $3.33, and Wilson buying 26,472 shares at $3.78 on 17 August and 30,769 at $3.25 on 20 August, about $0.53M in all C (calculated).¹¹ After the last purchase Gagnon held 1,386,624 shares directly, about $2.9M at the close C (calculated). These are the only open-market trades in the filings I read, and the shares they bought closed at $2.06, 38% below the $3.33 Gagnon paid. I take them as a sign of conviction and not as a forecast; the sums are small next to the $5.7M Gagnon was paid for 2025.

What they did with capital over a full cycle

The honest answer is that this team has run only one phase: the expansion of a bitcoin miner, financed almost entirely by selling shares. Shares outstanding went from 224.2M on 1 January 2023 to 616.7M on 30 June 2026.²³ In 2024 it issued 137.0M shares for $292.5M and in 2025 another 98.0M for $172.8M, some of them to pay for acquisitions.² In 2023 to 2025 cash used in operations was $475.2M, and by 30 June 2026 the accumulated deficit stood at $822.8M C (calculated).²³ It bought Stronghold in 2025, then sold or abandoned the Argentine and Paraguayan sites, which cost $76.0M of discontinued-operations loss in 2025 alone.² It repaid the $116.9M Macquarie loan in February 2026 at a loss of $21.6M, which was cautious with cash at the cost of that charge.³ And it bought back 7.8M shares for $10.0M in 2025, while issuing 98M.² The record is of a company that raises money when its stock is high and spends it on whatever the cycle favours.

There is also a lawsuit. A putative class action in the Eastern District of New York alleges that Bitfarms, Gagnon, a former CFO and a former CEO made misleading statements about the business and its controls between March 2023 and December 2024. Keel says it will defend the case and has recorded no provision.³ I take no view on the merits, and I count it as a risk with no price.

VFinancials

This is where a ten-year section would normally go, and I cannot honestly write one. The audited history under United States accounting covers 2023, 2024 and 2025, because the company reported under Canadian and international rules before then and recast only three years when it became an American filer.² Earlier years are on a different basis and describe a bitcoin miner that is being wound down. I show what exists, the first half of 2026, and a price history since 2021.

What the numbers say

Three things. First, the business that produced all of this revenue is being switched off, so the history says little about the next one. The 2025 revenue of $229.3M was a mining revenue, earned mostly in bitcoin, and in 2026 the machines went dark in stages.²³ Second, the old business never made money for owners. Add operating cash flow to the bitcoin it sold, take away the plant it bought and the stock it paid, and the result is negative in every year, from −$37.0M in 2023 to −$289.2M in 2024 and −$170.8M in 2025 C (calculated).² Third, the loss was funded by selling shares and borrowing, and that is the line that has changed: debt went from $1.6M at the end of 2024 to $1,025.8M by June 2026.²³

The balance sheet, in one paragraph

At 30 June 2026 total assets were $1,415.8M. They included cash of $715.5M, digital assets of $132.4M, plant and equipment of only $287.1M after the write-downs, $81.3M of deposits and prepayments on equipment, $53.4M of restricted cash held against letters of credit, and a $78.8M asset that is the value of the capped calls bought to limit dilution from the notes.³ Liabilities were $1,087.1M, almost all the convertible notes at $1,025.8M carrying value, which is below the $1,046M principal.³ The 2026 notes were priced on 5 June at 1.25% after the offer grew from $350M to $400M, and reached $458M when buyers took the $58M option.¹⁴³ Equity was $328.7M, and it had been $560.4M six months earlier.³

The stock price has told its own story. From a close of $0.68 on 8 April 2025 it rose almost ten times to $6.66 on 22 June 2026, then lost 69% in four months. I looked for a company reason and found none: Keel filed nothing after 21 August except routine ownership reports, and the other AI-infrastructure stocks fell by about the same amount.¹⁷ IREN fell 58%, Cipher 74%, TeraWulf 59%, Applied Digital 61%, Core Scientific 54%, CleanSpark 56%, Riot 55% and Hut 8 51% C (calculated).⁷ I cannot say why the sector fell. The bitcoin price rose, so it was not that.⁸ That makes this a debate about whether AI infrastructure keeps getting built at these prices, and Keel is in the middle of it with the least to show.

VIValuation

First, how this goes badly wrong

Before any number, I ask what would most likely wreck this, and work back. I see three ways.

The first is that no lease comes, or only a poor one. Keel would then be a company with a $1,046M debt, a $13.8M yearly interest bill, overhead of about $125M a year counting stock pay, and no income to meet them.³ Its $757M of cash and bitcoin would be spent before the notes fall due in January 2031 and January 2032, and the shares would be a claim on what is left. That is my bear case. The second is that leases come but the money to build them does not, or it costs too much. A tenant signs, Keel needs $12M to $14M for each megawatt, and it has to raise billions in a market where its stock is down 69% in four months. New shares issued at $2 to build a data center worth much more than $2 a share transfer the gain to the buyers. The third is that the tenant fails. A lease is only as good as who signs it. CoreWeave and Fluidstack, two of the tenants named above, are AI cloud companies whose own income depends on AI customers, and TeraWulf’s Fluidstack leases lean on credit support from Google to be financeable.⁵⁶ If AI spending cools, a tenant without that backing is the first to renegotiate.

What an owner can take out

Today, nothing. Operating cash flow plus bitcoin sold, less plant bought and stock paid, was −$170.8M in 2025, and the first half of 2026 used $117.6M in operations alone C (calculated).²³ Reported earnings are worse, a loss of $210.3M in six months, but most of that is depreciation that did not use cash. So reported loss and real cash are both negative and not far apart, and nothing separates them in a holder’s favour. What an owner can take out in future depends on the build: the model has Keel funding its own growth until about 2031, then paying out cash from 2032. In the base case the first payment to today’s holders is $88M in 2032, and total payouts to 2035 are $1.23 a share, against $3.17 of new money asked of the same holders along the way C (calculated).³

What I cannot judge

Three inputs decide the answer and I cannot know any of them. When, whether and at what rent Keel signs its first lease; whether tenants will fund the build, as at Core Scientific, or the landlord will, as at Galaxy; and what a finished lease is worth to a buyer in 2036. I do not know how many megawatts Keel will lease, so I show how the value moves with the number: each megawatt leased on my base terms adds about $3.6M of equity value C (calculated). The range in the tables is the honest answer. A single figure would be false precision.

Read the last rows carefully. At $2.06 the market is paying for about 559 MW of leases on my terms, which is 1.4 times my base case. If rent is $1.7M, which is nearer to the Core Scientific end, it needs about 875 MW. If rent is $2.1M, closer to Galaxy’s, it needs about 411. That is a statement about hope, not a statement about the company, and I draw no more from it.

How I model it

I do not have ten years of earnings to project. I have sites. The model takes each site, a number of critical megawatts, a first rent year, a rent, a build cost, a running margin and a loan, and adds them. Build spending lands over the two years before rent, half in each. Rent is paid for half the first year, then rises at 3% a year in the base. Overhead is cash plus stock pay. Interest on the converts is paid at the coupon until maturity and then refinanced. Taxes use a $700M loss carry-forward I estimate from the accumulated deficit, then 21%. If the company runs short of $100M of cash it raises new shares, and I charge a cost for the raise: 15% in the base, which is fees and the discount to value. At 2036 I value the finished sites at their net rent divided by an exit yield of 8%, less project debt and the notes. Cash flows are discounted at 12% from 31 October 2026, and equity value per share divides by 637M shares: 617.6M outstanding at 7 August, 19.3M restricted and performance units, and options through a treasury count.³

Bear

Tenants take the safe, finished buildings of the larger developers and Keel signs only about 190 MW, at lower rents, from 2029 on. It spends $2.3B and borrows $1.5B on poor terms. The notes come due in 2031 and 2032 and have to be refinanced at 9.5%, and there is nothing left for the shares. The model value is −$0.16, which I show as $0.

Base

Keel signs Moses Lake, Sharon and the first 140 MW at Panther Creek by 2030, and a further 171 MW across Panther Creek, Scrubgrass and Sherbrooke by 2032, 401 MW in all. Rent starts in 2028, a year later than the company’s own plan, and the build is paid for with $3.3B of project debt and $2.0B of new shares.

Bull

The market for AI space stays short for years. Keel leases 996 MW by 2032, gets better rents, borrows cheaply, and its notes convert because the stock is far above $7.41. Value per share is $10.92, on 799M shares.

Where does the base value come from? Not from the next five years, which cost money. The model’s present value of owner cash through 2036 is −$941M, mostly the equity Keel must raise, and the present value of what the sites are worth at 2036, after debt and notes and with the $100M cash buffer the model keeps, is $1,683M. Together that gives $742M, or $1.16 a share C (calculated). So the base case says Keel creates value only if it finishes what it starts. The 2036 value is a long way off, and I discount it at 12% for ten years, which cuts it to a quarter. A lower rate or a later lease changes the answer a lot.

Discount rate and sensitivity

I use 12%. It is the same hurdle I use for large profitable companies. For a company with no lease and heavy borrowing it is low if anything, so I put the extra risk in the cases and not the rate. The table shows what a different rate does. I also show what leased megawatts and rent do, because those two numbers decide it.

The price against the range

At $2.06, Keel is 77% above my base value of $1.16, and 81% below my bull case of $10.92. My margin-of-safety rule is to set the price where my bear value is three quarters of the price paid. That does not work when the bear value is zero, so I use half the base, $0.58, and I say plainly that this is a fallback and a weak one. To say the price is “fair” I would need to give the bull about a 15% chance: with weights of 40%, 45% and 15% the weighted value is $2.16, almost exactly the price E (estimate). If the bull were 10% likely it is $1.62; at 20% it is $2.71. The market is not crazy and I am not sure it is wrong. It is paying for the bull at a reasonable chance.

The next best use of the dollar, and quality against price

The closest comparable on this site for the power angle is Constellation Energy, whose base return a year I put at 7.7% against a price 20% above its base. Keel’s base return is 8.6% a year if you buy at $2.06 and fund your share of the build E (estimate). That looks better, and the gap is the cost of owning the uncertainty: Constellation’s range is $138 to $300 around a price of $262, Keel’s is $0 to $10.92 around $2.06. The better question for most people is whether they could get a similar return from a business with leases already signed. Quality and price are separate questions. As a business, Keel today is poor: it loses cash and has no customer, though it may become good. As a price it is neither cheap nor dear against my base, and that is the most I can say.

The bull case, which is what the price leans on, needs several things to go right at once, so I multiplied them. Keel signs a first lease by the end of 2027, 55%. That tenant is good enough to get the debt, 70%. The build comes in on cost and on time, 70%. Keel raises $2B of equity without paying more than 15%, 65%. A second wave of leases follows, 60%. Those are my guesses and they overlap, since a strong AI market helps all five. Multiplied they give 11% E (estimate), and allowing for the overlap, 15% to 25%. It matches the weight I give to the bull. Each step looks likely; all of them together do not.

Where this could be in 5 and 10 years

For the base to be right at five years, Keel needs about 230 MW leased and running by 2031 and the rest under construction, so that finished sites are worth $5.09 a share while the funding is still being paid. At ten years it needs 401 MW running, paying cash from 2032, and buyers still paying a yield of 8% for such leases. The drivers are four: megawatts leased, rent of about $1.9M a MW, a build cost near $13M a MW, and the 12% I discount at.

The bear leases about 190 MW, at lower rent and dearer money, and the $1,046M of notes eat the rest. The bull leases about 996 MW at $2.25M, borrows at 7%, and sees the notes turn into shares, which is why its share count is 799M.

The most likely way the base is wrong is timing, not size. If every lease slips a year the base falls from $1.16 to $0.71 today and to $8.07 at ten years, and if new shares are also sold near $2, at a 30% cost, the value today is $0.24 E (estimate).

These are estimates of value, not forecasts of the share price, and the market price can sit above or below value for years.

VIIThe case for the price

What most people miss

Three things cut in Keel’s favour and don’t make headlines. The debt is cheap and distant. The notes pay 1.375% and 1.25%, cost about $13.8M a year in interest, and are not due until January 2031 and January 2032, and Keel bought capped calls, for $41.8M on the second set, that reduce the share dilution if the stock rises as far as $11.88 and $11.86.³ Second, Keel owns generating plants. At Panther Creek and Scrubgrass it has about 123 MW of energized capacity it can supply to the grid while it waits, and the 10-K counts owned behind-the-meter generation among its advantages.³² Third, it already has hardware moving. The first Vertiv modules for Moses Lake were accepted in the second quarter and long-lead items for Sharon arrived, so the schedule is not only on paper.⁴

The bull case

In plain words: the world is short of power for AI, tenants will pay for speed, and Keel has some of the few sites with power already contracted. Management says all three priority sites are nearing full permitting, with several prospective tenants negotiating for each, and that it has $819M of liquidity and uncommitted 2027 capacity in PJM and Washington.⁴ Its chief financial officer says Keel is better capitalized than at any point in its history and expects to finance each site smoothly.⁴ I have only the company’s word on the negotiations. The release names no tenant and gives no rent.

I looked for named investors who have written about Keel. I found no public statement by Gerstner, Baker, Wood, Gurley, Chamath, Sacks, Ackman or any sell-side firm that I could open, so I cite none. What I can show is holdings. Situational Awareness LP, the fund run by Leopold Aschenbrenner, held 26,451,393 Keel shares at 30 June 2026, worth $151,830,996 at that date and 0.75% of the $20.24B it reported, in a 13F-HR filed on 14 August 2026.¹⁵ That is about 4.3% of the shares then outstanding C (calculated). A 13F shows long positions in United States listed stocks at one date; it does not show shorts or foreign holdings, and the filing says nothing about why the position is held or what the fund will do now. Jane Street Group, a trading firm, reported 33,552,042 shares, 5.6%, at the end of 2025.² BlackRock, Inc. filed a Schedule 13G on 29 July 2026 reporting 50,946,789 shares, which it puts at 8.4%, at 30 June 2026.¹⁷ I draw no conclusion from any of these.

Against reality

Today’s facts support the story’s inputs but not yet its outcome. The sites, the power contract, the cash and the hires are real.⁴² The lease is not. If Keel signs about 401 MW of leases at a rent near $1.9M a critical MW, builds at about $13M, and finances most of it with project debt, my base case follows and the shares are worth $1.16 today and more later. That is possible. It requires a first lease, soon, and then more.

The price asks for more than that. To be worth $2.06 on my terms Keel must lease about 559 MW, 1.4 times my base. Today’s number is zero. The distance is the whole of the argument, and what closes it is a signed lease, then a second, then a bank that lends against them.

The bull case needs more, and it helps to multiply it out, as I did in Valuation. The leases, the debt, the build, the equity and the second wave must all come, and together I put them near 11% before allowing that they move together.

What a 10x would take. Ten times $2.06 is $20.60. On the bull case’s 800M shares that is a market value of about $16.5B, thirteen times today’s $1.27B C (calculated). In my model it needs about 1,890 MW of critical load leased on bull terms, which is 125% of the whole 2.2 GW pipeline once converted at 70%, about 1,510 MW C (calculated). In other words, a tenfold move needs the company to lease more than it has told anyone it can. The stock has done something like this before: it rose from $0.68 on 8 April 2025 to $6.66 on 22 June 2026, nearly tenfold in fourteen months.¹ But that began at a market value of about $0.4B and with the sector rising, and the sector has since fallen 51% to 74%.⁷ From here, tenfold is not plausible on what I can see, and I would not build a case on it.

Where I come down

Two things moved me: the sites are real and the cash is substantial, and the insiders bought after the stock fell. The sector selloff did not move me, because it says nothing about Keel’s own leases, and the company’s statements about tenants did not, because they name none. A signed lease at $1.9M a MW or better, with a lender behind it, would raise my values most. A lease that is small, late or tenant-built would lower them, and so would new shares sold near $2.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• Customer concentration after the lease. The first tenant will be most of the revenue. The company’s own list of risks names dependence on significant customers and the risk that one fails to pay.⁴ In 2025 a single mining pool operator was 88% of revenue, so the habit exists.²
• Building on time. Hardware, transformers and cooling take long to arrive. The 10-K says Keel designs for NVIDIA’s Vera Rubin chips, which it expects to need more power per rack than the current generation, and warns the design may prove wrong.² Tariffs on equipment, and duties on mining machines imported earlier, add cost and uncertainty.³
• Power. The sites depend on PJM, Hydro-Québec and the Grant County utility. Power supply in Pennsylvania is contracted in stages: 50 MW by end 2026 and 300 MW by end 2027 at Panther Creek.² Delay in the utility’s transmission work would delay everything. The company also flags rule changes aimed at large electrical loads.³
• Financing. Project debt needs a lease. Equity needs a stock price. Both can close at once if the AI sector falls further.
• Legal. The class action in New York and tax and customs questions on past equipment imports are disclosed with no provision.³
• Bitcoin. Small now: 1,861 coins. Keel recorded a $4.2M provision on a receivable from BlockFills, which filed for bankruptcy in the first quarter.³
• Management bandwidth. A team that ran a miner is running the largest build of its career with a new president and a new CFO.

Geopolitical risk, and what it means for the value

Keel has already left the two countries that carried the old risk. The Argentine plant was abandoned after a power cut-off in May 2025 and Paraguay was sold in April 2026, so neither is in my value.³ What is left is Canada and the United States, where the risk is who decides on power and what imported equipment costs. Nothing here needs a war or a sanction.

Power rules. Pennsylvania matters most: Sharon, Panther Creek and Scrubgrass are 321 of the 401 MW in my base. FERC ordered PJM on 18 December 2025 to write rules for large loads beside power plants, and Keel warns that states are proposing rules aimed at large loads.¹⁸³ In Québec, Decree 88-2026 of 18 February 2026 tells the regulator to rank bitcoin mining below data centers and to raise the mining tariff; Hydro-Québec asked for rises of up to 199%, and Keel is contesting the decree in court.² That hurts the old mining business, which my value ignores. Sherbrooke still needs the province’s energy ministry.³ In Washington, Grant County’s utility caps data center growth and has about 800 MW of requests waiting.²¹⁹

Tariffs. Steel, aluminum and most copper products carry a 50% US duty. Switchgear, power supplies and cooling units are not on the derivative list, but the copper and steel inside them are exposed.²⁰ The Supreme Court struck down some emergency tariffs in March 2026, and Keel expects the rate on Chinese-origin imports near 27.6%.³ My $13M a MW sits just under Galaxy’s $13.5M for Phase I, delivered in 2026; Galaxy estimates Phase II at $15.4M, 14% higher.⁵ So today’s duties are in my cost and a further rise is not.

Bitcoin rules. Keel stopped mining in the United States on 29 June 2026, and its 1,861 coins are worth $144M.³ If bitcoin halves, the base falls from $1.16 to $1.06 E (estimate).

The price already sits $0.90 above my base value before any of this, and my base starts rent a year after the company’s own plan. Not in the price: Sherbrooke, a second round of tariffs, and Pennsylvania’s rules. The odds rest on Galaxy’s own 14% step up in cost, on PJM’s plan that new lines could take more than five years, and on the city’s approval of Sherbrooke without the ministry’s.⁵²³