Digi Power X
A former bitcoin miner becoming a landlord for AI: it builds power-connected, liquid-cooled halls and rents each one whole to a single tenant for a fixed monthly fee.
The scores
Growth1/4 ↑
Revenue fell 7.6% in 2025 and 18.3% in Q2 2026 as hosting winds down; the growth exists only in a contract that bills from December.
Business quality2/4 →
Power-secured sites are real, but every converted miner named sells the same hall to the same tenants; the one contract was won on a date.
Financial position2/4 ↑
No debt and $142M of cash, but Phase 2 needs an unsigned loan and the 10-Q says the agreement is at risk without it.
Execution record2/4 →
The substation and pod were delivered, but the $40M spending guide became $110M, cash includes crypto, and a share-sale statement contradicted that day's 10-Q.
Per share capture1/4 ↓
Shares tripled since 2024, the sale programme is spent with no loan signed, share pay is half of revenue, and Phase 2 may be share-funded.
Potential2/4 ↑
Phase 1 is committed, funded and in the ground; the 25 MW carrying most of the contract awaits a loan, and the rest is unsigned.
Risk profile2/4 →
One tenant, one building, a ready-for-service date with termination behind it, resolving inside the horizon; plus a father-son executive pair on a 17-person staff.
Valuation2/4 ↑
$275M above cash buys one contracted building plus a second nobody signed; Fair only if Phase 2's loan closes at a rate the fee carries.
Evidence quality3/4 →
Central facts are filed: the 10-Q note, the redacted agreement, the buyer's filings; the economics are estimates because charges, prepayments and power price are undisclosed.
If it works 5.1× · if it fails you keep up to 84%.
The Verdict
We rate Digi Power X unattractive. Once a bitcoin miner, it is building a hall in Alabama that an AI chip company has leased on a long contract. The contract is real, but the business is weak: one tenant, one building, and a share count that has tripled while paying for it. The price is full, not cheap: it counts that building finished plus a second nobody has signed, paying now for a tenant not yet found. As a bet it is speculative: if a second tenant signs and the loan closes, the business could be worth about four times today; if not, much of the money could go.
The long caseIf AI tenants keep running short of halls built for hot, water-cooled chips, a small owner with power and a finished building can lease it on long contracts.
The evidence under this - 8 claims
- If the Cerebras non-recurring charges and prepayments are large, the building costs the company less than the $110M spent so far suggests, and the equity's share of the fee is higher than this reading allows (C10).
- The New York sites' 141.7 MW, the North Carolina land and the West Virginia letter of intent are carried here at nothing; if any is contracted the second building is not hypothetical.
- A project loan on Phase 2, if signed at a rate the fee carries, would leave the $66M to the equity net of interest, which at the market value above cash is a plausible return rather than a full price.
- A large-load contract with Alabama Power may price below or above the state industrial average; the delivery agreement's rate is not disclosed, and the Alabama PSC tariff was not fetched (open question).
- The company's 10-K reports 89% of its New York electricity as zero-carbon and a 60 MW hydro allocation, but the Cerebras hall is in Alabama, served by Alabama Power, so the New York power does not help this contract.
- The contracted piece is one agreement whose second phase is conditioned on financing not yet raised (C27), and the 50 MW and 10 MW beyond it have no signed customer.
- The GPU run-rate of $100M on 10 MW implies about $10M of revenue for each megawatt against the $1.08M earned in five weeks on 0.6 MW, and the fleet to reach it would cost several times the $48.3M deployed so far (C14, C15).
- The company guided $40M of capital spending for fiscal 2026 in March and reported about $110M deployed by August (C28).
The two documents were filed the same day and give different Alabama figures; the 10-K is the signed one and is what this thesis carries.
The company
AI chips draw so much power and heat that ordinary data halls struggle to cope. Digi Power X sells the building around them: its own substation, two independent power paths, backup supply and liquid cooling, so any part can be serviced without switching the hall off.
The tenant brings its computers. Cerebras chose this over converting other people's halls, which its own prospectus calls slow, costly and less efficient.
The evidence under this - 4 claims
How it makes money
Three small businesses pay the bills today, and all three lost money in the June quarter: hosting other people's bitcoin miners, selling power from the gas plant, and renting out GPU servers.
The one that matters starts in December. Cerebras pays $195 a kilowatt each month, rising 3% a year, whether it uses the space or not. Power comes out of that fee, and in Alabama it costs at least $27.5M a year for the full hall.
The evidence under this - 4 claims
- A large-load contract with Alabama Power may price below or above the state industrial average; the delivery agreement's rate is not disclosed, and the Alabama PSC tariff was not fetched (open question).
- The company's 10-K reports 89% of its New York electricity as zero-carbon and a 60 MW hydro allocation, but the Cerebras hall is in Alabama, served by Alabama Power, so the New York power does not help this contract.
Where the money comes from, stream by stream
| Stream | Share of sales (Q2 2026) | How it earns it |
|---|---|---|
| Colocation (hosting bitcoin miners) | 51.7% | Charges miners' owners to house and power their machines in New York; a legacy business losing money and winding down. |
| Energy sales | 29.5% | Sells electricity from the 60 MW North Tonawanda gas plant into the New York grid. |
| AI data centers (GPU rental) | 16.3% | Rents NVIDIA GPU servers to one customer, SubQ AI, billed on usage; Cerebras's fixed colocation fee joins this segment from December 2026. |
| Cryptocurrency mining | 2.4% | Mines bitcoin on its own machines; nearly wound down. |
Who runs it
Michel Amar has run the company since 2020, with his son Alec as president. The pair turned a bitcoin miner into an AI landlord and signed Cerebras, but spent about $110M on building in 2026 against the $40M they planned. Insiders owned about 12% in April, before many more shares were sold. Their bonuses reward a positive adjusted profit figure and a bigger stock-market value, which selling new shares also raises.
The evidence under this - 4 claims
The Cerebras agreement was signed after the March guidance, so the larger programme has a contracted reason; what did not change was that it was funded by the ATM rather than the balance sheet the release described.
- The percentages are struck on 72.8 million shares; the count was 98.5 million at 30 June 2026 with 2.85 million more sold after (C19), so unless insiders added shares the group's stake has since been diluted to well under its April figure.
- Most of the chief executive's position is held through three companies he controls (Bit Mining International, Bit.Management and NYAM), and 1.415 million of it is options rather than shares.
- Adjusted EBITDA as the company defines it kept a $7.69M currency gain in Q2 2026 while the FY2025 table added back a currency loss (C16, C17), so the measure the bonus floor turns on is one the company's own definition flatters.
- Market value rises with share issuance as well as with the share price: the share count tripled from 2024 to mid-2026 (C19), so the milestones can be crossed by selling stock rather than by raising its value per share.
Why it wins, and who could beat it
Why customers come back
The scarce thing is power already delivered. The company holds 196.7 megawatts of connected power across Alabama and New York, when North America's main markets have 1.4% of space empty and New York has paused permits for new large data centres.
The proof so far is one contract: Cerebras, which leases every hall it uses, signed for the whole Alabama building for ten years before it existed.
The evidence under this - 5 claims
The two documents were filed the same day and give different Alabama figures; the 10-K is the signed one and is what this thesis carries.
Whether converting an existing permitted mining site to an AI hall needs a new discretionary permit is not addressed by the order's announcement or by the company; the conversion is targeted only for late 2027 (C31).
The threat
Converted miners with a head start compete for the same tenants. Core Scientific has leased 590 megawatts, mostly to CoreWeave. TeraWulf has 522 contracted, with Google standing behind part of the rent. Applied Digital delivers standard 150-megawatt buildings in 14 to 18 months. Digi Power X is still building its first hall.
The evidence under this - 3 claims
The rivals: CORZ, WULF, APLD, IREN, CRWV, RIOT, HUT, KEEL, HIVE, CIFR.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| CORZ | Core Scientific, Inc.Priced against it: One segment only | AI and HPC colocation | On sale | One segment only |
| WULF | TeraWulf Inc.Priced against it: One segment only | AI and HPC colocation | On sale | One segment only |
| APLD | Applied Digital Corp.Priced against it: Same product | AI and HPC colocation | On sale | Same product |
| IREN · CRWV | IREN Ltd (NVIDIA Cloud Partner) and CoreWeave, Inc.Priced against it: Different product | NeoCloudz bare-metal GPU rental | On sale | Different product |
| RIOT · HUT · KEEL · HIVE · CIFR | Riot Platforms, Hut 8, Bitfarms, HIVE and Cipher - the hosts Digi Power X's own 10-K namesPriced against it: One segment only | Legacy bitcoin-miner hosting and the North Tonawanda gas peaker | On sale | One segment only |
How big it could get
How big this could get
North America's main data-centre markets had 7,481 megawatts under construction in the first half of 2026, 80.4% already spoken for. None of those markets includes Alabama, so Digi Power X's own market cannot be sized. The clearest bound is its one buyer: Cerebras has promised OpenAI 750 megawatts of computing, against 40 contracted here.
The evidence under this - 3 claims
Who has signed
Cerebras has signed firmly for the first 15 megawatts, due 15 December 2026. The other 25 megawatts depend on finance both sides accept. The GPU customer's $19.6M contract is only expected value: after a $2.95M upfront payment, it pays for what it uses.
The evidence under this - 4 claims
The results
The numbers
Sales fell 7.6% in the year to Dec 2025.
It lost 83% of sales.
It keeps -9¢ of each sales dollar after making what it sells, and spends 48¢ of it on admin, its biggest running cost.
Shareholders' money lost 39% a year, up from −56% in the year to Dec 2024.
It used $42.8M of cash in the latest year, so the $78.5M it holds lasts about 1.8 years at that pace.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-10-04. Fiscal years end in December.
| Dec 2024 | Dec 2025 | |
|---|---|---|
| Size and growth | ||
| Sales | $37.0M | $34.2M |
| Sales growth | - | −7.6% |
| Profit | ||
| Gross margin | −31% | −9.4% |
| Net margin | −33% | −83% |
| What growth costs, per sales dollar | ||
| General and admin | 20% | 48% |
| Pay in shares | 6.9% | 23% |
| Capital spending | 10% | 51% |
| Cash | ||
| Cash from operations | −$17.5M | −$25.5M |
| Free cash flow | −$21.3M | −$42.8M |
| Balance sheet | ||
| Cash and short-term investments | $1.70M | $78.5M |
| Returns and shares | ||
| Shares (diluted) | 30.7M | 44.5M |
What growth costs
So far growing has made the company less profitable: office costs, pay in shares and building spending have all risen much faster than sales, which fell. That is spending ahead of a fixed rent that starts in December; whether it pays off depends on that rent arriving on time.
The evidence under this - 3 claims
The spending is for a building that bills from December 2026 at a fixed fee (C04, C05); overhead of about $15M a year in cash (C48) set against a contract worth $35M to $94M a year would read very differently once it bills.
The last report (14 Aug)
Quarter ended June 2026, filed figures.
The June 2026 quarter caught the old business shrinking before the new one had started. Revenue fell 18.3% to $6.63M, with the first GPU rental adding $1.08M from five weeks, and the company lost $14.36M.
The release led with a positive adjusted profit of $3.3M. That figure kept a $7.69M currency gain, although a currency loss was taken out the year before. Without the gain it was a loss of about $4.3M.
What it did to the thesisIt confirmed the Unattractive view: the old business shrank before the new rental income arrived, and the headline adjusted profit only looked positive because it kept a currency gain. Nothing in it eased the main worry, which is how the building gets funded without more shares.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $-0.17 against $-0.05 expected; $-0.17 as filed, with no adjustments.
The evidence under this - 4 claims
The company's stated definition excludes 'changes in fair value of financial instruments' and does not mention foreign exchange in either direction, so the asymmetry may be inconsistency rather than design; the FY2025 add-back is nonetheless in its own table.
What management did with the money
It sold shares, fast: 27.95 million for $163.2M in the first half of 2026. The chief executive told the August call none had been sold below $7.25; the filing that day showed 2.85 million sold after June at about $3.95. Building cost about $110M against a $40M plan.
Insiders in 2026Q2: 2 purchases and 3 sales, from their Form 4 filings.
The evidence under this - 3 claims
The Cerebras agreement was signed after the March guidance, so the larger programme has a contracted reason; what did not change was that it was funded by the ATM rather than the balance sheet the release described.
The price, and where we disagree
In short. Full. The price pays for the contracted building finished with borrowed money, and for a second building that has no tenant. Each new share the company sells shrinks what a holder owns of both.
The first phase is due to open on 15 December 2026, and the September-quarter results, due around 12 November, come first. Management promised revenue more than double June's and a loan for the second phase, with Goldman Sachs arranging it. The report will show whether the loan was signed, or more shares sold instead.
The evidence under this - 5 claims
- The contracted piece is one agreement whose second phase is conditioned on financing not yet raised (C27), and the 50 MW and 10 MW beyond it have no signed customer.
- The GPU run-rate of $100M on 10 MW implies about $10M of revenue for each megawatt against the $1.08M earned in five weeks on 0.6 MW, and the fleet to reach it would cost several times the $48.3M deployed so far (C14, C15).
- The company guided $40M of capital spending for fiscal 2026 in March and reported about $110M deployed by August (C28).
- The same 10-Q states that financing 'may not be available on acceptable terms or at all' and that without it the company may be unable to satisfy the Cerebras agreement (C27).
- The company's delivery record for a Tier III hall is one 1 MW modular pod live since May 2026; the 10-K lists 17 employees.
How we read the multiple
At $4.09 on 18 September the market valued the company at about $417M, about $275M more than its cash and crypto. The contracted Alabama building, fully let, earns at most about $66M a year after power, before cooling, staff and upkeep, and half of it still needs a loan.
So the price pays for that building finished and debt-funded, plus roughly one more like it that nobody has signed. Meanwhile the share count more than doubled in a year.
Against its own history. Far dearer than usual per dollar of sales, because sales shrank as the shares rose. Against what it owns it looks cheaper: the new shares brought cash.
The evidence under this - 4 claims
- If the Cerebras non-recurring charges and prepayments are large, the building costs the company less than the $110M spent so far suggests, and the equity's share of the fee is higher than this reading allows (C10).
- The New York sites' 141.7 MW, the North Carolina land and the West Virginia letter of intent are carried here at nothing; if any is contracted the second building is not hypothetical.
- A project loan on Phase 2, if signed at a rate the fee carries, would leave the $66M to the equity net of interest, which at the market value above cash is a plausible return rather than a full price.
- A large-load contract with Alabama Power may price below or above the state industrial average; the delivery agreement's rate is not disclosed, and the Alabama PSC tariff was not fetched (open question).
- The company's 10-K reports 89% of its New York electricity as zero-carbon and a 60 MW hydro allocation, but the Cerebras hall is in Alabama, served by Alabama Power, so the New York power does not help this contract.
Priced against its rivals
Only Applied Digital sells the same thing to the same buyers as a large part of what it does. A price comparison needs at least two, so the price is judged against its own history instead.
Also checked and set aside: Digital Realty Trust (a substitute, not the same product); Equinix (a substitute, not the same product); MARA Holdings (a substitute, not the same product).
Private or foreign rivals that cannot be priced: Bitfarms Ltd., Argo Blockchain Plc.
The bet: what it pays if it works
This is a bet on a landlord filling a second building. If the Cerebras hall opens on time, the loan is signed and a second tenant takes a hall of the same size, the business could be worth about four times today's market value in two to three years, on a generous reading of the leases.
If it fails, little is left: the cash is going into the building, a loan would be paid before shareholders, and Cerebras may walk away if the hall is late.
| Revenue in two to three years | Contracted megawatts needed | At the stated multiple |
|---|---|---|
| $200.00M (the big win) | about 80 at 100% kept | 5.1× |
| $300.00M | about 123 at 100% kept | 7.6× |
| $415.00M (market forecast) | about 172 at 100% kept | 10.6× |
JLL's outlook implies about 19 GW built for lease in the US over 2027-29; at 0.75%, the power Digi Power X holds or plans, it leases 140 MW plus 10 MW of GPUs, about $415 million. On that forecast, $415.00M pays 10.6× from today's value: a Cheap bet on the market case.
To pay 2× (a Fair bet) it would need about $78.59M of revenue (about 28 contracted megawatts); to pay 5× (a Cheap bet), about $196.47M of revenue (about 79 contracted megawatts), at the same multiple.
The steps it has to take
- Done Ten-year take-or-pay agreement with Cerebras for about 40 MW at Columbiana
- Done First modular pod live and GPU rental billing
- In progress Q3 2026 revenue at least double Q2's, as guided due 2026-11-12
- In progress Phase 2 project loan signed as debt, amount and rate disclosed due 2026-12
- In progress Phase 1's 15 MW ready for service for Cerebras due 2026-12-15
- Slipped Share count stops rising: no at-the-market draw while the build is funded due 2027-03
- In progress Phase 2's 25 MW ready for service due 2027-03
- Not started A second AI colocation tenant signed for any site due 2027-12
- Not started Electricity below a third of the fee through a full year of operation due 2027-12
The detail behind the label
The evidence under this - 7 claims
- The contracted piece is one agreement whose second phase is conditioned on financing not yet raised (C27), and the 50 MW and 10 MW beyond it have no signed customer.
- The GPU run-rate of $100M on 10 MW implies about $10M of revenue for each megawatt against the $1.08M earned in five weeks on 0.6 MW, and the fleet to reach it would cost several times the $48.3M deployed so far (C14, C15).
- The company guided $40M of capital spending for fiscal 2026 in March and reported about $110M deployed by August (C28).
The two documents were filed the same day and give different Alabama figures; the 10-K is the signed one and is what this thesis carries.
Where we disagree with the market
The $1.1B contract makes this an AI landlord worth paying up for; the one analyst covering it says buy.
The contract is real and the building will probably be finished. The shareholder's slice of it is what keeps shrinking.
Twice in a year the same pattern: an announcement lifted the shares, the company sold stock into the rally, and the price fell back. After the Cerebras deal the shares hit $9.20 in May and closed August at $3.60, the contract intact and more shares out.
Against us: the chief executive and the president bought shares in August, and a signed project loan would change the funding story.
What analysts say, and the evidence
The evidence under this - 7 claims
- The same 10-Q states that financing 'may not be available on acceptable terms or at all' and that without it the company may be unable to satisfy the Cerebras agreement (C27).
- The company's delivery record for a Tier III hall is one 1 MW modular pod live since May 2026; the 10-K lists 17 employees.
What could go wrong, and how we would know
The risks
- One tenant, one building. Cerebras is all the contracted AI revenue, and Cerebras itself gets most of its own revenue from two customers.
- A late opening. Miss the December date and Digi Power X owes delay credits; miss the final deadline and Cerebras can walk away.
- Paying for the second phase. The filing says more financing is needed, and without it the company may not meet its obligations to Cerebras.
- GPUs that do not pay. The GPU fleet loses about $16.1M a year to depreciation against the roughly $9.8M its contract is expected to bill.
The evidence under this - 5 claims
- Management targets 10 MW and up to a $100M annualised run-rate by end-2027 and says NVIDIA and several banks now finance GPUs against residual value, so a second customer or higher utilisation could close the gap (C25).
- Depreciation over three years is the company's own choice; a longer economic life would lower the annual charge, though the useful-life policy is what the filing states.
What we still do not know
- How large Cerebras's upfront charges and prepayments are. They are redacted, and could make the shareholder's share much better than this reading.
- Whether the New York sites, which cannot add power under the state's pause, find an AI tenant in 2027.
The evidence under this - 3 claims
Whether converting an existing permitted mining site to an AI hall needs a new discretionary permit is not addressed by the order's announcement or by the company; the conversion is targeted only for late 2027 (C31).
No New York customer is named or signed; the FY2025 release's '60 MW hydro' allocation could not be found in the New York Power Authority's award releases and is carried as the company's statement.
What would change our view
- Unattractive→NeutralPhase 2 is financed with debt, not with shares
It needs all of:
- ○The first phase opens on time and the second phase's loan closes as project debt, with no share sales paying for the build.watched as it happens
- ○The share count stops rising: no more shares sold while the build is being paid for.checked at the next results, due by 12 Nov
- Unattractive→AttractiveA second tenant, and the economics hold at the meter
It needs all of:
- ○A second AI tenant signs for any site, so the case stops being one customer in one hall.watched as it happens
- ○Electricity for the Cerebras hall stays below a third of its fee through a full year, with the fee still covering power.checked at the next results, due by 12 Nov
We would also be wrong if
- The first phase is not open by 31 March 2027, or Cerebras ends the agreement.
- GPU rental still earns less than the fleet's depreciation through 2027.
What is coming
- 12 November 2026 Q3 2026 results: Whether revenue at least doubles Q2's as guided, the Alabama project loan is signed with terms, and the quarter was funded without more share sales.
Every dated event ahead
- 12 November 2026 Q3 2026 results: Whether revenue at least doubles Q2's as guided, the Alabama project loan is signed with terms, and the quarter was funded without more share sales.
- 15 December 2026 Alabama Phase 1 ready for service (15 MW for Cerebras): Whether the first contracted hall is delivered on date and starts billing, or delay credits and the termination clock begin.
- March 2027 NVIDIA Vera Rubin systems delivered: Whether the $35M GPU order arrives and finds a renter on terms that cover depreciation, unlike the first contract.
- March 2027 Alabama Phase 2 (25 MW) targeted complete: Whether financing closed and the larger part of the Cerebras fee bills, or the phase slips for want of a loan.
- July 2027 New York data-centre permit pause lapses at the latest: Whether the New York sites can add power for the planned AI conversion or stay capped at today's load.
Since we wrote this
- 13 OctNewsDigi Power X has a $200M Alabama loan commitment, not a signed loan. View held$200M Alabama debt financing commitment letter.
- 9 OctBet moveBet re-struck: Fair bet to Cheap betRe-priced at the 8 October close.
- 5 OctThesis reviewThe bet now names what it ridesThe owner asked that every bet name the macro trend behind it, after BWXT's bet was missing its chief executive's case for nuclear over gas.
- 4 OctThesis reviewOwner's checklist: revenue streams, cost of growth, management, catalystsThe owner's checklist found the method silent on who runs the company, what growth costs, where the money comes from and what is dated ahead; this pass fills those four blocks from the 10-K/A and filings.
- 19 SepThesis writtenUnattractive, medium convictionWritten a month after the June-quarter results, three months before the Alabama hall is due to open, with the shares 55% below May's high.
Also checked 3 times with nothing new to act on.
Sources 35
Sources - 35
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-03-31Item 1 business and competition, Item 2 power available (196.7 MW), risk factors, the going-concern and functional-currency notes, the FY2025 revenue split, share counts at 2024 and 2025 year-ends, share-based pay, the remediated material weaknessesbacksC01C18C19C29C32C41C44C46C47C53C64C66
- S02primarycompany filing2026-04-30directors and officers, the father-son relationship, Ajay Gupta's background, beneficial ownership, salaries partly paid in bitcoinbacksC43C62C63C64C65
- S03primarycompany filing2026-08-14the statements, the four-segment note, the Cerebras commitment note with the $195 rate and 3% escalator, the lessor note on the usage-based GPU contract, the $48.3M of leased hardware and its three-year life, the USDC deconsolidation, the ATM issuance and the post-quarter sales, the warrant table, the financing risk factor, the New York executive orderbacksC02C03C04C05C10C13C14C16C18C19C20C21C22C23C24C27C30C42C44C47C53C66
- S04primarycompany release2026-08-14the adjusted EBITDA reconciliation and definition, the '$142.4 million cash' description, the 2027 outlook, the Alabama, New York, North Carolina and West Virginia updates, the USDC valuation statementbacksC02C16C23C25C26C28C31C42C51
- S05primarytranscript2026-08-14the chief executive's ATM price statement, the New York grandfathering and power figures, the Gupta relationship on the West Virginia letter of intent, the USDC raise, Goldman Sachs's engagement; the questions were shareholder-submitted and read by the operator, a different origin from the answers
we are grandfathered in New York. We cannot expand our current footprint of power. So I think we are using 60 megawatts in North Tonawanda and we are using 18 megawatts in Buffalo. So we are allowed to use the same power. We are not allowed for now to expand power.
backsC21C25C26C30C31C42C43C45 - S06primarycompany filing2026-05-08the summary terms: 40 MW, ten years, exclusive licence, the two phases and their dates, the financing condition, the one-time payments and prepaid fee, the $1.1B and $2.5B figuresbacksC04C10
- S07primarycompany filing2026-05-08the fee basis 'per kW per month (all-in, including delivered power, excluding applicable taxes)', the take-or-pay clause, the daily delay credits, the final-outside-date termination right with refund of unapplied prepaid fees and NRCs, the liquidated-damages clause, the letter of credit, the Phase 2 financing approval mechanism, the lender recognition provisionsbacksC04C05C09C10
- S08primarycompany release2026-05-05the substation-complete and Alabama Power delivery-agreement statements, the self-funding of Phase 1, Hans Vestberg as senior advisorread, but no claim rests on it
- S09primarycompany release2026-04-20the 24-month term, the $19.6M expected value, the $2.95M upfront payment, the facility description (Rated 3, N+1 UPS, CDU and chiller two-loop cooling), the 4,000-GPU capacity claimbacksC13C40
- S10primarycompany release2026-06-03the $35M Vera Rubin commitment, cash of about $150M and $65M deployed at 3 June, the substation and interconnection statementsbacksC24C26
- S11primarycompany release2026-07-07the 2027 outlook by segment, cash of about $155M and $95M deployed at 3 July, the ARMS 200 pod live since 15 May 2026backsC25C26C51
- S12primarycompany release2026-05-15the first statement of the 2027 outlook, the May 2026 option and RSU grants, Q1 adjusted EBITDA and share-based paybacksC25C44
- S13primarycompany release2026-03-31the $40M fiscal 2026 capex guide, the '70MW approved' Alabama and '400MW pipeline' descriptions, the 60 MW hydro claim, the FY2025 adjusted EBITDA table with its foreign-exchange add-backbacksC28C29
- S14primarycompany filing2026-05-08the $175M programme size, $72.4M sold by 8 May under the first $75M, the 3% commission, the $6.19 share price that daybacksC20
- S15secondaryvendor data2026-09-19Research pack, 19 September 2026 (FMP market tier; FINRA short interest; price series from FMP historical-price-eod)price, the 52-week range and its dates, the 252-day decomposition, the analyst panel and estimates, short interest, the measured news reactions, revenue per employeebacksC22C41C45C50C53
- S16primarycompany filing2026-08-21two discretionary purchases in August 2026 and six planned sales by the president from January to May 2026backsC43
- S17primarycompany filing2026-05-14the buyer's own account of leasing third-party halls in California, Oklahoma and Canada, why retrofits are worse than purpose-built space, and the 750 MW OpenAI deploymentbacksC12
- S18primarycompany filing2026-08-12$690.3M of fixed lease obligations, further leases signed in Q2 and Q3 2026 not yet commenced, customer concentration at 34% and 32%; no mention of this landlordbacksC11
- S19primarycompany filing2026-03-02the capacity ladder (1,426 / 920 / 590 / 120 MW), colocation revenue of $65.4M, the conversion planbacksC33
- S20primarycompany filing2026-02-27522 MW contracted, the Fluidstack and Core42 leases, Google credit support, ten- to twenty-five-year termsbacksC34
- S21primarycompany filing2026-07-29the 150 MW standard design in 14 to 18 months, hyperscaler qualification, long-term lease and hosting arrangementsbacksC35
- S22secondaryindustry data2026-09-19vacancy, under-construction capacity, the pre-committed share and the available inventory, measured outside the company
primary market vacancy fell to a record low of 1.4% in H1 from 1.6% in H1 2025 … Total under-construction capacity across primary markets increased by 24.8% in H1 to a record high of 7,481.1 MW … Preleasing activity accelerated with commitments made on 80.4% of all under-construction capacity, compared with 74.3% a year ago … Less than 1,500 MW of future capacity across all primary markets remains available, representing approximately six months of demand
backsC36 - S23secondaryindustry data2026-04-07Synergy Research Group - Hyperscale operators to account for 67% of all data center capacity by 2031 Synergy Research Groupthe own-built share of the buyer class, and colocation's growth rate
they now account for 48% of the worldwide capacity of all data centers. Almost 60% of that hyperscale capacity is now in own-built, owned data centers with the balance being in leased facilities … Looking ahead to 2031, hyperscale operators will account for 67% of all capacity … colocation capacity will continue to increase each year at near double-digit rates
backsC37 - S24primarygovernment data2026-09-19Alabama's average industrial price, 7.84 cents a kilowatt-hour year-to-date June 2026 against 7.69 a year earlier, the basis for netting power out of the all-in fee
Table 5.6.B. Average Price of Electricity to Ultimate Customers by End-Use Sector, by State, Year-to-Date through June 2026 and 2025 (Cents per Kilowatthour) … Alabama 16.58 15.98 14.79 14.46 7.84 7.69
backsC07 - S25primarygovernment data2026-09-19the rate of the driver, measured by a statistical agency
Across all cases, servers alone accounted for an estimated 7% of commercial sector electricity consumption in 2025. Data center server electricity use grows to 22%-33% of commercial building electricity use by 2050 across our cases.
backsC38 - S26primaryindustry data2026-09-19Uptime Institute - Tier Classification System Uptime Institutewhat the Tier III label in the agreement actually requires
A Tier III data center is concurrently maintainable with redundant components as a key differentiator, with redundant distribution paths to serve the critical environment. Unlike Tier I and Tier II, these facilities require no shutdowns when equipment needs maintenance or replacement.
backsC39 - S27primaryother2026-09-19Cerebras - CS-3 vs. Nvidia B200: 2024 AI accelerators compared (vendor technical page) Cerebras Systemsthe per-system power draw the hall's cooling is sized for
The CS-3 is a 15U server consuming up to 23kW of power. … The CS-3 consumes 23kW peak while the DGX B200 consumes 14.3 kW.
backsC40 - S28primaryregulator2026-07-14Governor of New York - First statewide moratorium on new hyperscale data centers launched by Governor Kathy Hochul (Executive Order, 14 July 2026) New York State, Office of the Governorthe state action that freezes new large data centres in New York, where 141.7 MW of the company's power sits
The Governor is temporarily pausing State environmental permits for up to one year in order to build a nation-leading regulatory framework that protects ratepayers, the environment, the energy grid and communities across the state. … During the development of this GEIS, which will take up to a year, a moratorium will be in place and the Department of Environmental Conservation (DEC) will not issue any discretionary permits not already deemed complete.
backsC30 - S29primaryindustry data2026-08-11a broker's North American absorption, the capacity under construction and its owner-occupied part, the tenant mix, the frontier markets' share of the pipeline and the size of an enterprise requirement; read through ir_fetch
With vacancy sustained at 1% for the third consecutive year despite unprecedented construction activity, North America now has 66 GW under construction, 95% of which is pre-committed. … Hyperscalers remain the dominant force, accounting for 59% of 2026 data center tenant demand while meeting capacity needs through both leasing and self-building, with 28 GW of owner-occupied capacity currently under construction. … Individual enterprise requirements typically range from 500 kW to 3 MW
backsC54 - S30primaryindustry data2026-01-05a broker's forecast of global capacity to 2030, the Americas' share and growth rate, the US share of the Americas, and the construction cost a megawatt; read through ir_fetch
The data center sector is projected to increase by 97 GW between 2025 and 2030, effectively doubling in size over a five-year period. By 2030, global data center capacity could reach 200 GW. … The Americas is the largest data center region , representing about 50% of global capacity. The Americas also has the fastest growth rate of the three global regions, with a projected 17% supply CAGR through to 2030 … Between 2020 and 2025, the average global data center construction cost increased from $7.7 to $10.7 million per MW
backsC55 - S31primaryindustry data2026-04-02Synergy Research Group press release - Neocloud Market Forecast to Approach $400B by 2031, Driven by Surging AI Infrastructure Demand Synergy Research Groupan independent market tracker's 2025 GPU-rental (neocloud) revenue and its forecast to 2031, naming transitioning crypto infrastructure providers in the long tail; read through ir_fetch
New data from Synergy Research Group reveals that neocloud revenues are scaling at an unprecedented pace, reaching $9 billion in Q4-up 223% year-over-year-and exceeding $25 billion for full-year 2025. Driven by surging demand for AI infrastructure, Synergy forecasts the market will approach $400 billion by 2031, representing a sustained 58% compound annual growth rate.
backsC56 - S32primaryindustry data2025-12a second analyst firm's 2025 base and 2030 forecast for GPU rental by neoclouds, on a different definition from Synergy's; read through ir_fetch
ABI Research expects that by 2030, global revenue from GPUaaS offerings by neocloud providers are projected to reach nearly US$250 billion, at a Compound Annual Growth Rate (CAGR) of 43% (from US$42 in 2025).
backsC57 - S33primarycompany filing2026-08-11the largest listed GPU-rental company's quarterly revenue, active and contracted power and revenue backlog: the revenue a megawatt of rented GPUs earns; read through filing_fetchbacksC58
- S34primarycompany filing2026-07-28590 MW leased to CoreWeave with 395 MW billing, and the 27 July 2026 leases of 377 MW to AMD and 152 MW to a neocloud, fifteen-year terms, with AMD's right to reserve up to 1,925 MW more; read through filing_fetchbacksC59
- S35primarytranscript2026-08-14the chief executive's timing and size for North Carolina, the West Virginia growth path and the capital intensity of GPUs; read through transcript_fetch (FMP), the shareholder-submitted questions read by the operator
So we own 2 lands over there, about 40 acres. We acquired within the last 12 months, an additional adjacent land so we can properly design and build an adequate data center there. We are in the process with Duke Energy and the zoning to get all the necessary load studies, permits for a major data center. And we believe that we will be able to utilize that asset towards 2029, 2030 with an additional 150 to 200 megawatts of power.
backsC60