Applied Digital

IThe business

Applied Digital owns buildings full of electrical and cooling equipment, and rents them to companies that fill them with computers. The tenants pay per megawatt of power the building can deliver to the computers, for 15 years, whether or not they use it. The company calls the buildings AI factories. It designs them, builds them, owns the land and operates them, from Dallas, Texas.³ Fiscal years end on 31 May, so “FY2026” means June 2025 to May 2026.

It began as a bitcoin-hosting business, Applied Blockchain, in 2021.⁹ It still runs two such sites in North Dakota, 106 MW at Jamestown and 180 MW at Ellendale, for one customer whose contract had about a year and a half left in May 2026.³ Its April investor deck names that customer as Marathon Digital.⁶ In the first quarter of FY2027 the sites earned a segment profit of $13.3M on $37.8M of revenue.² Everything else it has done since 2023 points the other way. A second business, the cloud service, rented out graphics processors; in May 2026 it was merged into a listed shell called ChronoScale, in which Applied Digital kept about 96%.²

What it has signed

As of 31 August it had leases for about 1,410 MW across five campuses, worth about $36B of rent over the 15-year base terms, or about $86B if every renewal option were used.⁴ The leases are take-or-pay: the tenant owes the rent whether it uses the space, and ending early means paying the full remaining value.³ The 10-Q gives a schedule of minimum payments: $468M in FY2027, $1,580M in FY2028, then $2,361M and $2,371M a year from FY2029, and $26.6B after FY2031, in total $35.7B R (reported).²

Only the first campus has a named tenant. Polaris Forge 1, 400 MW, is leased to CoreWeave in three buildings. The first 100 MW opened in October 2025. A second building of 150 MW was handed over in two halves, on 1 July and 1 October 2026, so 250 MW is live.⁴ The third, 150 MW, is under construction. In March 2026, CoreWeave restructured two of the leases: the term on two of the four data halls in one lease was suspended and re-let to a new CoreWeave subsidiary on the same terms, and the other building’s lease was assigned to that subsidiary. CoreWeave added springing guarantees and agreed to post a $50M letter of credit.³ The company describes this as an improvement in credit quality following a rating of A3 on a CoreWeave refinancing.¹⁰ I could not tell what makes the guarantees “spring”, because the exhibits set that out and I did not read them.

What the first quarter looked like

Revenue for the quarter to 31 August was $341.9M, up 322%. Only $65.8M of it was base rent. $183.5M was tenant fit-out work that Applied Digital organises and bills to the tenant, against $176.1M of cost, and $13.3M was costs passed back to the tenant.² The company’s own measure of the rental business, net operating income, was $58.8M, an 89% margin. Adjusted EBITDA was $64.4M. The reported operating loss was $62.4M, after $66.5M of stock pay and $41.9M of depreciation.² So this is a business that has just started to earn rent, and for now most of its revenue is pass-through.

What is still to build

About 1,160 MW are not yet earning rent C (calculated).⁴ The campuses are in two groups. Polaris Forge 1 and 2, 600 MW, sit inside a subsidiary, TopCo 2, which carries $5.9B of debt and $2.0B of Macquarie preferred stock, and held $2.9B of cash and $0.5B of restricted cash at 31 August.² Delta Forge 1 and 2 and Polaris Forge 3, 810 MW, sit outside it, and I found no project financing yet for them in the filings. Management’s deck puts the cost at $11M to $13M a megawatt.⁶ Applied Digital also reports an active pipeline of over 3 GW and a further 5 GW of earlier discussions.³ In October it signed an agreement for up to about 1 GW of potential power in Finland, its first site outside the United States.⁴

Power, and a related company

On 4 October 2026 it signed a 15-year power purchase agreement with Base Electron for the output of a planned 1,200 MW gas plant in North Dakota, to supply Polaris Forge 3. It pays fixed monthly amounts based on contract capacity, with a credit support agreement, and Base Electron can cancel it if the plant has not reached financial close by 31 March 2027.² Base Electron is a related party. The proxy says it is owned and managed by certain officers and directors of Applied Digital, including Wes Cummins, Jason Zhang and Richard Nottenburg, acting in their own right, and Applied Digital owns about 10%.⁷ Applied Digital has also lent it $59M. The board set up a committee of directors to review the transaction.²⁷

IIThe moat

I ask two questions. Could a rival take the tenant at the same price, and would the tenant bother? For a landlord whose tenants are among the largest companies in the world, the second matters more than the first.

What is hard to copy

Power. The company says getting power and a grid connection ahead of demand is the main limit on new capacity, and that this is where it differs from most rivals.³ I can check part of this. It holds sites in North Dakota, where the deck says electricity costs about 24% less than the national average and there is surplus generation, and it has a gas plant in prospect through a company its leaders partly own.⁶² A tenant that needs 300 MW by 2028 cannot easily get that anywhere. That is real.

Speed and procurement are the second item. It says it can build about 150 MW in 14 to 18 months, and that it bought electrical equipment from ABB, cooling from BASX and generators from Caterpillar in bulk before shortages.³ The chief executive said on a podcast in January that locking in the supply chain 18 to 24 months earlier was a major advantage.¹² The record so far supports him: the first 100 MW opened in October 2025, the 150 MW building in two steps in 2026, against the 2H 2025 and 2H 2026 windows given a year earlier.³⁴ Three buildings is a short record.

What is not

The rent. The first campus pays $1.83M per megawatt-year. The three newest campuses pay $1.67M, and the latest, Delta Forge 2, $1.65M C (calculated).³ That is about 9% less, in a year when the company says power is the gating factor.⁴ A seller with a scarce product usually raises the price. The tenants are few and large: three customers in the first quarter were 56%, 21% and 11% of revenue.² The 10-K lists IREN, Cipher, TeraWulf, Hut 8, Riot, CleanSpark, Core Scientific, Galaxy and others as rivals, as well as Digital Realty and Equinix and the hyperscalers building their own.³ Several of the bitcoin hosts on that list are trying to do what Applied Digital did.

The shares moved with the group

Between 28 May and 30 October Applied Digital’s share price fell 64%. CoreWeave fell 49%, IREN 63%, Cipher 70%, TeraWulf 56% and Core Scientific 52%. Oracle fell 36% and Vertiv 26%. NVIDIA fell 4% and Microsoft rose 22%.¹⁸ The fall was shared with the companies that are financed with debt to build AI capacity, not with the companies that sell chips or own the software. A news search found no single cause. One press account points to the debt, rising yields, a falling bitcoin price and debt-funded builders going out of favour, after a record quarter.¹³ I do not claim to know the reason.

My judgement: it has a good position in a scarce thing, power and a quick build, and a short record of using it. What would make this a wall instead of a position is evidence that tenants pay more for it, and the rents so far point the other way.

IIIThe competitive advantage

The advantage is the spread. A megawatt costs about $12.5M to build and earns about $1.46M a year of net operating income, so on my base numbers it earns about 11.7% on cost for 15 years E (estimate).⁶³ Money to build it costs about 7% to 9% from lenders and 12.75% or more from Macquarie.² The deck’s own figures give a range: net income over cost runs from 11% to about 15%, and 15% needs the lowest cost and the highest rent together.⁶ A press account says management expects new lease rates to rise by more than 15%; I come back to it in Section VII.¹³

Against the alternatives

The honest comparison is with a landlord who finances the same building with long debt. Applied Digital’s project debt on the first campuses cost 9.25%, 6.75% and 7.0%, and the notes have traded below par for the newer two: the 10-Q gives fair values of about $2.0B for $2.15B of 6.75% notes and about $1.5B for $1.59B of 7.0% notes, which I read as yields of roughly 8% to 9%, with the usual caution that the filing rounds to $0.1B.² The tenant is the credit. For the three newest leases the tenant is investment grade by the company’s account, and for the first it is CoreWeave, which had $35.6B of debt and lost $626M in the second quarter of calendar 2026.¹⁴ Lenders therefore see a rent stream from a very good credit on three campuses and a leveraged one on the first.

The second part of the advantage is the structure. Macquarie put $1.825B into TopCo 2 as preferred stock that accrues 12.75% a year and carries a 1.8 times minimum payout, together with 13.5% of the common equity of that subsidiary.³ That bought the company time and cash. It is also the most expensive money in the structure, and it ranks ahead of the shares.

Here is how I see it. The assets are good: powered land, a design the tenants have qualified, and long contracts with credits that are mostly strong. The question is how much of the value goes to the lenders and to Macquarie before it reaches the common shares. Valuation, below, takes this apart.

IVManagement

Wes Cummins, 49, is the chairman and chief executive and a co-founder. He came from technology investing, at Nokomis Capital, and founded an investment firm, 272 Capital.⁶ Jason Zhang, the other co-founder, became president in January; Saidal Mohmand is the finance chief and Laura Laltrello the operating chief.³⁷ The company had about 256 employees on 31 May 2026.³ I judge a management by how it is paid, what it owns, and what it did with other people’s money.

How they are paid

The proxy reports Cummins’ total pay for FY2026 at $189.2M. Of that, $185.9M is the grant-date value of stock awards. Cash pay was $2.1M, a salary of $875,000 and a $1.2M bonus, and other pay of $1.2M included $1.1M of personal use of the company aircraft. For FY2025 it was $27.7M and for FY2024 $1.2M.⁷ Zhang’s was $100.0M and Mohmand’s $40.5M.⁷

What the stock awards reward is clear. In January 2026 Cummins was granted 1.5M restricted stock units worth $45.4M and 4.5M performance units worth $122.1M. The performance units vest in thirds if the average price over any 90 days reaches $50, $75 and $100 within five years, while the price at the grant was $30.26.⁷ Today’s $17.82 is below all three. The board’s own arithmetic is that at the full $100, he would realise $450M against about $21B of added market value, a share of 2.15%, which it calls below the 25th percentile of the awards it compared.⁷ Zhang’s units vest on signing leases and getting buildings into service on time; Mohmand’s on net operating income.⁷

That pays them for the share price, for leases signed, and for megawatts delivered. It does not pay anyone for return on the money put in, for rent per megawatt, for dilution or for keeping debt low. A building that is leased at a thin spread and delivered on time earns the award in full. I note too that the company paid $99.4M in cash in one quarter to cover taxes on vested stock, so stock pay is not free to the other shareholders.²

What they own, and what they sold

Cummins beneficially owns 23.36M shares, 7.8%. Of those, 17.59M are held by a family company and are pledged against a line of credit with JPMorgan; at $17.82 that block is about $313M C (calculated).⁷ I do not know the loan size or terms, so I cannot say how close it is to a margin call. Directors and officers as a group hold 9.5%. The two largest outside holders listed are Hood River Capital at 7.4% and Jane Street at 7.0%, the latter including 15.96M warrants.⁷

Insider trades in the year to 30 October tell two stories. Between October 2025 and October 2026, Cummins and four directors sold 534,368 shares for about $18.3M, at $21.84 to $38.57, 165,000 of them Cummins’ at $36.42 on 16 January. On 16 October 2026 Cummins bought 55,803 shares at $18.04 and Zhang bought 300,000 at $18.86, about $6.7M in all C (calculated).¹⁵ A purchase at $18 after sales at $36 is a signal. It is also small beside the $313M of pledged stock.

What they did with capital

The cycle is short: five years as a listed company. Weighted shares were 57M in FY2022 and 292M in the first quarter of FY2027, and 302M at 6 October C (calculated).⁵² The company has sold stock through the market and through a Series G preferred line that converts at 95% of the recent low price, which it can draw down in $150M amounts every seven days, and which it raised to $2.0B in June.³² It has 28.1M warrants outstanding at an average $8.61, 21.5M of them issued to CoreWeave with its leases and since assigned to banks and trading firms.²³ In FY2025 it sold $450M of 2.75% convertible notes and paid $51.8M for a capped call and $52.7M for a prepaid forward.¹¹

The record of strategy is mixed. A shareholder class action filed in August 2023 alleges the company overstated the profit of its hosting business and its ability to become a cloud provider; the court stayed the case in September 2025 while a motion to dismiss is pending, and the company cannot estimate a loss.² The cloud business it pushed was split off in 2026. The buildings pivot has worked on delivery so far.

Related-party dealings need more care than usual. Applied Digital paid $37M in May 2026 to end a guarantee it had given on a roughly $2.4B Babcock & Wilcox contract for Base Electron, in return for Base Electron shares; it lent Base Electron $59M, and it holds B&W warrants whose fair value fell from $90.8M to $34.7M in a quarter, a $56.1M loss.⁷² Six family members of Cummins are employees, paid $3.3M since 1 June 2025.⁷ The four top executives were also granted 5.25% of ChronoScale’s shares in units.⁷ None of this is hidden, and each is approved by a board committee. It does mean that the people running a tenant-funded landlord also have personal stakes in the power company it depends on for its next campus.

Following the incentives, then: the leaders are paid in large amounts of stock for price and growth, they own a lot, and a good part of what they own is borrowed against. That pays them to build more and to keep the share price up. It does not stop them issuing shares at a discount to fund more. The delivery record is good. The capital record is of a company that has raised every kind of money it can find.

VFinancials

The company went public in 2021 and moved to the Nasdaq in 2022, so there are five fiscal years of operating results, not ten, and the first three are a different business: bitcoin hosting. Revenue rose 72 times from $8.5M to $611M in FY2026 and the company has had an operating loss in every year C (calculated).³¹¹⁵ Figures are the latest as recast; FY2024 revenue was $136.6M when first reported and $165.6M in the FY2026 10-K, because of how the cloud business is now shown.³

What an owner could really take out

Nothing. In the twelve months to August, operations produced $235M of cash, the company spent $4,691M on buildings and equipment, and it paid its people $271M in stock, so the owners’ cash after stock pay and spending was −$4,727M C (calculated).²³ That is not a fair measure of the business, since most of the $4.7B went into campuses that are not yet earning, but it is the plain fact: every dollar came from lenders, Macquarie and new shares. Reported net loss over the same period was $351M. The gap is the building programme.

A better question is what the finished campuses leave for owners. The whole of Valuation answers it. In short, rent at full build is about $2.4B a year, net operating income about $2.1B, and before the owner sees anything come interest of about $1.0B, debt repayment, overhead of about $0.3B including stock pay, and the Macquarie claim. The finished business is a thin slice.

Quality of earnings

The company’s adjusted measures exclude ChronoScale, stock pay ($59.4M in the first quarter), deal costs and the non-cash write-down of customer incentives. Adjusted EBITDA of $64.4M compares with a GAAP operating loss of $62.4M.² The adjusted net loss was $4.1M against a reported $221M loss to common after continuing operations. Three items make up the gap that I cannot dismiss as one-off: stock pay of $66.5M a quarter, 42% of revenue other than fit-out billings C (calculated); a net $49.5M loss on the Babcock & Wilcox warrants and Macquarie derivatives; and interest of $77.4M, with another $35.7M capitalised into buildings.² Interest will rise as buildings open and capitalised interest stops.

The balance sheet

At 31 August the company had $2.95B of unrestricted cash and $0.73B restricted, and $6.4B of debt on the balance sheet net of costs.² Most of the cash sits in TopCo 2, which held $2.9B of unrestricted cash and $0.5B restricted; it is tied to Polaris Forge 1 and 2.² Total assets were $11.7B, of which property and equipment $6.3B, $3.7B of it still under construction. The Macquarie preferred stock is shown outside equity at $2.0B, and shareholders’ equity was $1.6B R (reported).² The company says it has enough liquidity for the next twelve months and all its known plans.² I read that as covering the plans it has financed, which do not include the 810 MW outside TopCo 2.

VIValuation

First, how this goes badly wrong

Before any numbers I ask what would most likely hurt the business or the investment, and work back from there. Here are three, and they feed each other.

The financing does not hold. About $11.5B has still to be spent to finish the 1,410 MW E (estimate). The 600 MW in TopCo 2 are mostly funded, but the 810 MW outside it are not, and the company has no project loans for them in the filings I read. The cheapest source it has is a preferred-stock line that turns into shares at 95% of the recent low price, and the shares have fallen 64%.²³ A $58.5M Texas Capital loan requires the market value of the shares to stay above $2.0B at every close, against about $5.4B now.² If the lenders will not fund the unfinished campuses at 70% to 75% of cost at 7%, the shareholders face dilution that may not stop.

A tenant fails, or the contract is restructured. Polaris Forge 1 is $11.0B of the $36.2B, and its tenant is CoreWeave, which had $35.6B of debt, $103.7B of contracted revenue still to deliver and a loss of $626M in its latest quarter.¹⁴ The leases were already restructured in March.³ If that tenant struggled, Applied Digital would hold a building worth much less than the debt on it.

The price paid for money. Macquarie’s preferred accrues 12.75% a year in kind and has to be repaid at 1.8 times its $1.825B, about $3.3B, with rate step-ups after October 2030 and a right to force a sale after October 2032.³ Add $6.1B of notes at 6.75% to 9.25% and the project loans to come. The interest on all of that is a large part of the rent. Of the three, I think the third is the most likely to cost the shareholders, because it takes effect even if everything else works.

What an owner gets today, and later

Today, nothing: owner cash after stock pay and spending was −$4.7B over twelve months C (calculated).² On my base, the first cash to the shares arrives in FY2032, from the 2030 to 2031 notes’ refinancing and the Macquarie payout, and over FY2032 to FY2039 it totals about $0.57B, or $1.83 a share before discounting, then the shareholders are asked for money for three years while the buildings are refitted E (estimate). Most of the cash comes in FY2045 and later, when the loans are repaid. That is the shape of a lease: the rent pays the lenders first, then the owner of the preferred, and then the shareholders.

What I can’t judge

Three inputs decide the answer and I cannot know any of them. The first is what each megawatt really costs. The company says $11M to $13M; the balance sheet of TopCo 2, which carries $10.3B of assets for 600 MW, $3.4B of it cash, points higher.⁶² I use $12.5M and show the range. The second is what happens after the leases end. The company says renewals would take total contracted revenue to $86B from $36B, which implies rent about 1.4 times higher if every option is used.⁴ I do not know who decides: the tenant holds the option. The third is the rate at which buyers will value finished campuses; I have no sourced private-market figure for a data centre leased to a hyperscaler, so I show a table instead of a number.

How I value it

I build the cash that reaches the shareholders year by year to fiscal 2062, in US dollars, from the contracted rent in the 10-Q, a net operating margin, overhead and stock pay, the cost still to be spent, the interest and repayments on the debt, the payout to Macquarie and the loans for the unfinished campuses, and then discount it at a rate for equity.² Stock pay is a cost, because it is paid with shares that are not available to anyone else. Equity calls, the cash shareholders would have to add if the model runs short, are subtracted at a cost of 8% for the discount on the preferred-stock line. Per-share values use 302.4M shares at 6 October, 7.8M restricted units and 3.0M performance units, plus warrants and converts when the value is above their strikes of $8.61 and $9.75. The chief executive’s 4.5M performance units, with hurdles of $50, $75 and $100, are left out, since the price is under all three.

The rent for FY2027 to FY2031 is the 10-Q schedule; my base contract rent after FY2031 totals $26.6B against the 10-Q’s $26.6B, within 0.1% C (calculated).² The convert is $450M of debt repaid in June 2030 if the value is below $9.75 and becomes shares above it, reduced by the capped call that covers the gap to $14.72.¹¹

The discount rate. I take it from the company’s own securities. The 2031 notes were worth about 93 to 94 cents on the dollar, a yield of roughly 8% to 9%; the most junior outside money, Macquarie’s, costs 12.75% and rises; the common stock ranks below both.²³ A rate under 13% for shares behind that much senior claim would not be honest, so I use 13% in the base, 16% in the bear and 12.5% in the bull, which is a little under my own floor because the bull case also assumes cheaper money. I show 9% to 16% below.

Bear

Building costs about $15.5M a megawatt, the level TopCo 2’s $10.3B of assets across 600 MW would imply if all were spent, and the 810 MW outside TopCo 2 are financed at 70% of cost at 8%. CoreWeave stops paying on Polaris Forge 1 in FY2031, and the 400 MW are re-let two years later at about half the rent. Macquarie’s preferred has to be paid off at its 1.8 times minimum. The common holders would be asked for more than the buildings could return, so I put the value at zero.

Base

Everything signed is delivered on the company’s schedule at $12.5M a megawatt. The rent is as contracted, flat, to FY2042 to FY2044. After that tenants use about three quarters of their renewal options at rents about 1.4 times the original and the rest is re-let lower, which I average to 1.2 times for 15 years. Overhead and stock pay run at about $300M a year. Shareholders get their first cash in FY2032, after the preferred is paid off, then less each year as the loans are repaid, and the larger sums only after FY2045.

Bull

Cost falls to $11.5M. Tenants use all their renewals at 1.4 times the rent and the buildings are re-let for 20 years. Applied Digital also signs a further 1,500 MW on the same terms and finances it at 75% of cost at 6.75%. The shareholders’ rate of return required drops to 12.5%.

The price and what it needs

Turn it round. At $17.82, with the converts and warrants it brings, the market values the common at about $6.4B and the whole enterprise at about $10.9B, 31% of the $35.7B of rent contracted C (calculated).² That sounds cheap. But the buildings are not finished: another $11.5B has to be spent, and by completion the claims ahead of the shares are about $15.1B E (estimate). Allowing for 3.6 years until then at 13%, the price needs the finished campuses to be worth about $25B, which is 7.1% of their net operating income after overhead. My base is worth about 11% on the same figure.

A rate of 7.1% is a rate for a building that will be rented for ever at rising rents. The rent in these leases is flat for 15 years, then depends on renewal. If you think a hyperscaler campus deserves a 7% to 8% rate because tenants will keep renewing and rents will rise, the price is about right. If you think a flat 15-year contract and a refit deserve 10% or more, the common is worth little. The table shows how quickly that choice moves the answer.

How much the answers move

The base value is a thin slice and it moves a lot. The move is not even. A cost $1M a megawatt lower adds about $1.90 a share, taking the base from $1.50 to $3.43 at $11.5M; a cost $0.5M higher takes it to zero at $13M E (estimate). The re-letting assumption moves it from zero, if tenants leave and the buildings sit empty, to $3.00 if rents stay at 1.4 times for 20 years.

A few other changes, one at a time on the base: no stock pay, $4.86; overhead at $100M, $2.78; margin of 90%, $2.76; no maintenance spending, $3.06; interest at 9.5% when notes are refinanced, $0.10; CoreWeave stops paying, zero; and adding 1,500 MW of new leases at today’s cost and rent, zero, because new buildings at $12.5M a megawatt earn about what the money costs E (estimate). New buildings only help if they cost less than $11.5M a megawatt or rent for more.

The price against the range

The price is 1,084% above my base and 2.1 times my bull. Measured to October 2036 as in the table above, my base returns −13% a year from $17.82, and even the bull returns 7% a year E (estimate). I find it hard to write a value this far below a quoted price, so I say what could be wrong. I may be too harsh on the discount rate: 13% for a leveraged claim on investment-grade rent may be high. I may be too harsh on the tail, as the tenants’ renewal options carry rents about 40% higher by my arithmetic from the company’s totals. And a market that pays 7% for finished buildings would be right. What I cannot do is produce a cash flow to shareholders that gets to $17.82 unless the buildings are valued like infrastructure, and I would rather say that than move the numbers to meet the price.

Against the alternative

A Treasury bond pays a fixed amount and needs none of this. Within this site, the closest case is Oracle, which also builds capacity for AI tenants with debt: its price is 7% above my base value. The supplier side is Vertiv, at 89% above. Applied Digital’s price sits much further above its base value, because its shareholders sit behind more money.¹

Quality and price, separately

How good is the business? Better than the stock suggests: delivered buildings, long contracts, strong tenants, and a rent that is paid whether the tenant uses the space. How good is the price? On my numbers, poor, because the claims ahead of the shares absorb almost all of the rent. This looks like a good set of assets with a poor claim on them, at a price that already assumes the assets are worth an infrastructure rate.

How the odds multiply

The price needs several things together. My guesses are in the table. They are not independent, but each is far from certain, and the product is about 9%.

Where this could be in 5 and 10 years

These are my values for the common at two later dates: the present value, at that date, of the owner cash that comes after it, per share on today’s count of 313M plus warrants and converts where the value is above their strikes, after interest, debt repayment, the Macquarie preferred and stock pay, in each year’s dollars. “Owner cash” is cash paid out less money the shareholders would have to add.

For the base to hold at five years, all 1,410 MW must be running by 2028, the Macquarie preferred must be paid off around October 2030 without new equity, and the loans must amortise on schedule. Its value then grows at about the 13% rate I discount at, because only small payouts are made before FY2040. At ten years the base has paid out about $1.59 a share and holds about $2.70, because the shareholders’ cash arrives mostly after FY2045.

The bear differs on cost, $15.5M a megawatt, and on CoreWeave, which stops paying on Polaris Forge 1. The bull differs on rent and life after the leases: 1.4 times for 20 years, 1,500 MW of new leases, and money at 6.75%. The numbers that drive all three are cost per megawatt, rent after the leases end, and the discount rate. The most likely way my base is wrong is cost: $0.5M a megawatt more takes today’s value to zero, and $1M less adds about $1.90 a share.

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

VIIThe case for the price

What most people miss

Three facts cut in the company’s favour and are not the usual talking points. First, the enterprise value at $17.82 is about $10.9B, 31% of the $35.7B of rent contracted in the leases, and about $2.4B a year of that rent is flat and take-or-pay from FY2029 to FY2042 C (calculated).² Second, the renewal options are worth more than the first terms: the company puts contracted revenue at $86B if all are used against $36B for the base terms, which by my arithmetic means rent about 1.4 times higher on the renewal years C (calculated).⁴ Third, the chief executive and the president bought shares on 16 October at $18.04 and $18.86, after the chief executive had sold 165,000 at $36.42 in January.¹⁵ I do not read intent into it. It is a fact.

The bull case

In plain words: AI computing is short of powered, finished buildings, hyperscalers want them fast, and few people can supply them. Applied Digital has leases for 1,410 MW with large tenants, a repeatable design, a queue of more than 3 GW and a gas plant in prospect. If it builds on time, finances at about 7% and renews its tenants, the leases are worth more than the stock.

Who says so, and what I could verify:

Management. Wes Cummins told The Data Center Frontier Show on 27 January 2026 that 2026 and 2027 are “execution years,” that many announced projects will not become operating facilities, and that Applied Digital had secured its electrical and mechanical supply 18 to 24 months earlier, which he called a big advantage, in my words.¹² The show’s write-up says the company is focused on six customers, five hyperscalers and CoreWeave. The filings name only CoreWeave and describe the rest as investment-grade hyperscalers, so I cannot check the write-up’s list. On 29 June 2026 he gave a 6:32 interview on CNBC’s Squawk on the Street; the page has no transcript, so I cannot say what he said and do not characterise it.¹⁶ A press account of the October results says management expects about 250 MW of expansion leases by the end of 2026 with lease rates up by more than 15%.¹³ I could not find the call transcript, so I do not know who said it or what the 15% is measured against. The newest leases pay $1.65M to $1.67M per megawatt-year, so 15% more would be about $1.9M, and no such lease is in the filings yet. I have no time-coded transcripts, so I give episode dates, not timestamps.

Analysts. As reported by 24/7 Wall St. on 26 October, Rob Brown at Lake Street has a $90 target, Jones Trading $70, Wells Fargo $55, Morgan Stanley $41 and Rothschild Redburn $22, the last citing execution, power and pipeline risk, against an average target of $62.44 across 15 analysts.¹³ I did not open the notes. A $62 price would put the common at about $23B, which is 3.5 times today’s price C (calculated).

Holders. Situational Awareness LP, Leopold Aschenbrenner’s fund, listed 15,384,616 Applied Digital shares in its 13F-HR for 30 June 2026, filed 14 August, with a reported value of $469.0M, 2.3% of its $20.24B of reported holdings.¹⁷ That value works out to $30.49 a share, 18% below the 30 June close of $37.30. It is not a different quarter-end price: the fund’s 31 March filing divides to $23.74, the close on that day, and for six other holdings in the 30 June table (Micron, CoreWeave, Core Scientific, IREN, Riot and CleanSpark) value divided by shares equals the 30 June close to the cent.¹⁷¹ I could not reconcile this one line and found no 13D or 13G for Applied Digital that explains it. At the 30 June close the shares would be worth about $574M C (calculated). I use the filing as filed. The 13F does not show short positions or holdings outside the United States, and I do not know what the fund holds now. This is a dated fact, not a view. The proxy lists Hood River Capital at 7.4% and Jane Street at 7.0%, the latter including 15.96M warrants.⁷

Not found. I looked for what Chamath Palihapitiya, Jason Calacanis, David Sacks and David Friedberg of the All-In podcast, Gavin Baker and Brad Gerstner have said about Applied Digital and about AI data-centre hosting. A search of the web and of the shows’ listings turned up no episode, clip or transcript that discusses the company, so I cite none. If one exists, I did not find it.

Against reality

If Applied Digital builds the 810 MW outside TopCo 2 for about $12.5M a megawatt and finances them at 75% of cost at 7%, if CoreWeave keeps paying, if tenants use their renewals, and if buyers of finished campuses pay a rate near 7%, the price is possible. Today the first condition is unproven: no financing is in the filings for those campuses. On my base the shares are worth $1.50; the gap is the distance between a 13% required return on a flat, finite contract and a 7% rate for a campus that never stops earning.

What a 10x would take. Ten times $17.82 is $178, or about $64B of market value on today’s share count, which would grow. At a 7% rate on finished income, that needs campuses worth about $79B and income after overhead of about $5.6B, 3.1 times the $1.77B my base reaches in FY2030 C (calculated). That is roughly 4,420 MW of leased campuses at today’s rents, against 1,410 MW signed, or about $38B more to build at $12.5M a megawatt, with no fall in rents and no rise in rates. It is also 3.6 times the May high. For a company with this much debt ahead of its shares, tenfold is close to impossible. I would not build a case on it.

Where I come down

The delivery record and the strength of the tenants on the newer campuses moved me; the leases are better than the stock’s fall implies. The price did not move me, because it needs finished buildings valued at about 7.1% of their income, and the shareholders sit behind $6.3B of debt, $2.0B of Macquarie preferred stock and a further $11.5B to be spent. What would raise my values most is a loan on the 810 MW at a rate and cost like the first campuses, a cost per megawatt under $11.5M, a tenant using a renewal early at a higher rent, or a sale of a finished campus at a rate near 8% or lower. Any of those would show me what a buyer pays.

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. Three tenants were 88% of first-quarter revenue.² One is CoreWeave, which lost $626M in its latest quarter and had $35.6B of debt.¹⁴ A change at either would show up quickly.
• A lease is only as good as the delivery. Rent starts when buildings are ready for service. Delays in transformers, switchgear and generators were a theme of the January podcast, and the company says dates are subject to construction, permitting and interconnection risk.¹²³
• Regulation and local opposition. The 10-Q says several states have passed moratoriums on new AI data centres and others are considering them.²
• Power for the next campus. Polaris Forge 3 depends on a gas plant from a company the chief executive and others own and run, and the agreement can be cancelled if it does not reach financial close by 31 March 2027.²
• Dilution. The preferred-stock line converts at 95% of the recent low price, above a floor price the board can lower; 128,750 Series G shares became 5.2M common shares after the quarter.² A lower price means more shares for the same money.
• Pledged shares. 17.6M of the chief executive’s shares secure a bank credit line.⁷
• Lawsuits. A securities class action and a derivative claim are pending; the company cannot estimate a loss.²
• Key people. Cummins and Zhang hold both the company’s top jobs and stakes in Base Electron.⁷
• Legacy hosting. One crypto customer provides $154M a year and its contract ends in about a year.³

Geopolitical risk, and what it means for the value

Applied Digital has little direct exposure to other countries. The buildings are in North Dakota, Louisiana and Alabama, the tenants are American, and the tenants buy the chips. The one foreign step is Finland: an agreement for up to about 1 GW of potential power, nothing built.⁴ I rerun my base for four routes, one at a time.

CoreWeave. It is the tenant on 400 MW, $11.0B of the $36.2B. If its subsidiary stopped paying on Polaris Forge 1 from FY2031, my base goes to zero. I put that at about one in seven, from my odds table and CoreWeave’s $35.6B of debt and $626M quarterly loss.¹⁴

North Dakota and US power policy. Tenants pay for their own power, so a higher power price is their cost.³ The exposure is permission and delay. A North Dakota legislative committee on data centres met at the end of August, and a candidate for the Public Service Commission called in May for a halt on new data centres until the 2027 session.¹⁸¹⁹ The legislature and the commission decide; I found no law or order that stops a leased campus. The nearer risk is Polaris Forge 3, whose gas plant has a 31 March 2027 financial-close deadline. I put a miss at about one in three, because the filings show no financing with five months left.² A year’s delay takes the base to $0.60.

Tariffs on equipment. The 10-K says tariffs have affected power and cooling gear and that the company cannot predict trade policy.³ A May 2026 analysis says copper has carried a 50% duty since 1 August 2025, and that switchgear and large generators were left off the April 2026 list of derivative products.²⁰ The company says it bought early; I cannot tell how much of the gear for the 810 MW outside TopCo 2 is covered. Rents are fixed, so extra cost falls on the shareholders.

GPU export controls. They bind the tenants, not the landlord. In January 2026 the US moved some chip sales to China to case-by-case review, while Blackwell-class chips stayed under a presumption of denial.²¹ Contracted rent is take-or-pay and does not move. What moves is demand for the next lease: with none, my bull falls from $8.34 to $7.11.

What is in the price? At $17.82 the price needs a 7.1% rate on finished income, and it sits above my bull, so I see no sign it has priced any of these. The rows are one at a time and do not add up.