Core Scientific
This is the archived full research. Read the story
A former bitcoin miner in Texas and the South-East that now converts its mining sites into data centres and rents them, whole, to CoreWeave and AMD on long leases.
The scores
Growth3/4 ↑
Colocation revenue rose from $31.3M to $215.4M a quarter in three quarters; mining is ending and AMD's halls start in 2027.
Business quality3/4 →
Energised mining sites won twelve- and fifteen-year leases from CoreWeave and AMD, but rivals court the same tenants and buyers mostly build their own.
Financial position2/4 →
$4.4B of debt and $4.4B of commitments against $1.6B of cash; the AMD project bonds are still to raise, though warrant cash arrives in January.
Execution record3/4 →
485 MW delivered to CoreWeave within two years of signing, on the stated early-2027 path; the AMD schedule is untested.
Per share capture2/4 →
About 509M diluted shares against 322M outstanding, $62.4M of stock pay in nine months, and insiders owning 1.7%.
Potential3/4 →
Bounded and committed: about 1.1 GW signed with $24B of contracted revenue. Not a 4: AMD's reservation commits AMD to nothing.
Risk profile2/4 →
CoreWeave pays 93% of revenue with its bonds near 15%, and the AMD halls carry most of the growth before they are funded.
Valuation3/4 →
About 7.7 times stated NOI with build spending counted, under a landlord's usual ten. Not a 4: the main tenant's credit explains the gap.
Evidence quality3/4 →
Leases, debt, warrants and ERCOT's approvals are filed facts; the NOI and contracted totals are presentation figures.
If it works 1.4× · if it fails you keep about 49%.
The Verdict
We rate Core Scientific Neutral. A former bitcoin miner, it now rents converted mining sites to CoreWeave and AMD as AI data centres on long leases. The business is mixed: most of CoreWeave's halls already pay rent and the company now earns an operating profit, but almost all of today's money comes from one tenant whose own bonds trade as junk, and AMD's halls are unbuilt and not yet financed. The price is fair rather than cheap: it pays a little less for the signed rent than landlords usually fetch, which is about right for that tenant. As a bet it is fragile: the upside is mostly priced.
The long caseAI needs buildings with power already connected, and Core Scientific holds old mining sites the grid serves; long leases to CoreWeave and AMD turn that power into rent.
How we weighed it
The price against our value
Cuts both ways
No fair value is drawn: the filed years are a miner's. The price pays a little under a landlord's usual multiple for the signed rent, about right given who pays it.
The business
Cuts both ways
Mixed
Energised power on converted sites and a record of delivering, but rivals sign the same tenants and the buyers build most capacity themselves.
Growth against what the price assumes
In its favour
Rent has risen fast as CoreWeave's halls opened, and AMD's leases double the signed capacity over the next two years if built on time.
What could go wrong
Against it
Most of today's revenue comes from one tenant whose own bonds trade as junk, and the AMD build still needs its debt raised.
Against competitors
Cuts both ways
No filed rival compares cleanly; Applied Digital and Cipher offer the same product to the same kind of tenant.
How its group is behaving
Cuts both ways
AI hosting names move together on AI spending and credit news, and the group has fallen hard since early summer.
The latest news and report
In its favour
The AMD leases brought a second, stronger tenant, and the third quarter beat revenue estimates as CoreWeave's halls billed.
The evidence under this - 9 claims
The contracts are summarised in a presentation rather than filed in full, and the build credits mean cash rent runs below the straight-line GAAP revenue in the early years.
The March release targeted about 590 MW by early 2027 and the date still holds, but Dalton's second phase is the one building still under construction.
The net income is a warrant mark: the second quarter's net loss of $1,155.3M came from the same line moving the other way.
The same week's selling hit other AI-infrastructure names, and CoreWeave's contract-backed secured debt prices far tighter than its unsecured notes.
No bond has priced; the 10-Q says the company expects project-level bonds within twelve months, and analysts on the call asked whether lenders had tightened terms.
The commitments include greenfield work beyond the 1.1 GW and pass-through items, and the NOI is a stabilised figure reached only in 2028 (C07), before about $120M a year of cash overhead.
The company
Core Scientific takes sites that already have big connections to the power grid, rebuilds them as data centres and rents each building to one tenant for twelve years or more. The tenant brings its own computers; Core Scientific supplies the power, the cooling and the shell.
Tenants come because powered space is scarce: under 1,500 megawatts was free across the main U.S. markets in mid-2026, about six months of demand. The alternative is building it themselves, which the largest buyers do for almost 60% of their space.
The evidence under this - 5 claims
The contracts are summarised in a presentation rather than filed in full, and the build credits mean cash rent runs below the straight-line GAAP revenue in the early years.
How it makes money
Rent is now almost all of the money. Third-quarter 2026 revenue was $232.9 million: $215.4 million from renting data centres to CoreWeave, of which $57.6 million was electricity passed on at cost, and $17.5 million from bitcoin mining, which closes this quarter.
Management says the signed leases total more than $24 billion, about $1.6 billion a year of profit after running the buildings once all are built. CoreWeave paid 93% of the quarter's revenue.
The evidence under this - 5 claims
The net income is a warrant mark: the second quarter's net loss of $1,155.3M came from the same line moving the other way.
These are presentation figures; NOI is a non-GAAP measure, and it leaves out corporate overhead of about $30M a quarter in cash (C46).
Where the money comes from, stream by stream
| Stream | Share of sales (nine months to 30 September 2026) | How it earns it |
|---|---|---|
| Colocation | 84% | Licence fees for powered, cooled halls leased to CoreWeave and power passed through at cost; licence fees were $157.3M of the third quarter's $215.4M. |
| Bitcoin self-mining | about 12% | Mining bitcoin on Core Scientific's own machines and selling it; ending in the fourth quarter. |
| Hosted mining | about 4% | Running other miners' machines for a fee and a share of power; ending with the self-mining. |
Who runs it
Adam Sullivan has run Core Scientific since 2023 and led it out of bankruptcy in early 2024; the finance chief joined in 2025. Insiders own about 1.7%, and large funds such as Vanguard and BlackRock hold much more. Most long-term pay rides on the share price against small companies, megawatts energised and customers won, not on value per share.
The evidence under this - 3 claims
Why it wins, and who could beat it
Why customers come back
What Core Scientific holds is power already connected. At the end of 2025 it had 1,426 megawatts of grid capacity across ten sites. In Texas, where requests to connect large users far exceed the state's record demand, its Pecos and Hunt sites are conditionally approved in the grid operator's first batch.
That is why CoreWeave and then AMD signed long leases with a company that started renting halls only in 2024. It buys a place in line, not pricing power: nobody publishes what each megawatt rents for.
The evidence under this - 7 claims
Rivals hold powered sites and sign the same kind of tenant (C39, C40), and the buyers build most capacity themselves (C36).
Conditional approval is not allocation: ERCOT sets the final load ramp only when Batch Zero ends.
The CoreWeave figure is lowered by the build credits it paid for, and both totals are presentation numbers (C03).
The threat
The biggest buyers mostly build their own data centres, and rivals sell the same thing: Applied Digital signs fifteen-year leases with large cloud companies, Cipher rents halls to Amazon, and private developers such as Aligned, QTS and Vantage build to order. Core Scientific's answer is speed on sites it already runs, not lower rent.
The evidence under this - 4 claims
The rivals: AMZN, GOOGL, MSFT, META, APLD, CIFR, MARA, RIOT, WULF, and 1 more that is private, in-house or listed abroad.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| AMZN · GOOGL · MSFT · META | Amazon, Google, Microsoft and Meta building their own capacityPriced against it: Not checked | CoreWeave colocation | On sale | Not checked |
| APLD | Applied Digital CorporationPriced against it: Same product | AMD-contracted colocation | On sale | Same product |
| CIFR | Cipher Digital Inc.Priced against it: One segment only | AMD-contracted colocation | On sale | One segment only |
| Private | Aligned, Compass, QTS, Vantage and the other large-hall developersPriced against it: Private | CoreWeave colocation | On sale | Private |
| MARA · RIOT | MARA Holdings, Riot Platforms and the other listed bitcoin minersPriced against it: Not checked | Bitcoin mining | On sale | Not checked |
| WULF | TeraWulfPriced against it: One segment only | - | On sale | One segment only |
How big it could get
How big this could get
Demand is running hot. North American data centres absorbed 25 gigawatts in the first half of 2026, double a year earlier, and Texas alone has far more requests to connect than power to give.
The government's energy statisticians expect servers to take a growing share of the power used by commercial buildings for decades. The limit on Core Scientific is power it can deliver and money to build, not tenants.
The evidence under this - 3 claims
Who has signed
CoreWeave signed for about 590 megawatts across five sites for twelve years, and pays for its own power; 485 megawatts already bill. AMD and a cloud company it backs signed for about 530 megawatts for fifteen years, with AMD's credit behind the cloud company's leases. How the 530 is split between them is not disclosed.
The evidence under this - 5 claims
The March release targeted about 590 MW by early 2027 and the date still holds, but Dalton's second phase is the one building still under construction.
The contracts are summarised in a presentation rather than filed in full, and the build credits mean cash rent runs below the straight-line GAAP revenue in the early years.
The neocloud is unnamed and its own credit is undisclosed; AMD's support terms are not filed.
The results
The numbers
Sales fell 38% in the year to Dec 2025, slower than the 1.6% in the year before.
Operating margin is −77%, up from −329% in the year to Dec 2022.
Running costs take 89¢ of each sales dollar, up from 20¢ in the year to Dec 2021.
Yearly sales growth has ranged from −38% to +803% over the last 5 years, with sales falling in 2 of them.
It used $451M of cash in the latest year, so the $311M it holds lasts about 0.7 years at that pace.
It owes $749M more than the cash it holds, about 2.7 years of the cash its operations bring in.
The share count rose 36% in 4.0 years, and pay in shares costs 31% of sales.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-11-17. Fiscal years end in December.
| Dec 2021 | Dec 2022 | Dec 2023 | Dec 2024 | Dec 2025 | |
|---|---|---|---|---|---|
| Size and growth | |||||
| Sales | $544M | $640M | $502M | $511M | $319M |
| Sales growth | +803% | +18% | −22% | +1.6% | −38% |
| Profit | |||||
| Gross margin | 44% | 1.3% | 25% | 24% | 12% |
| Operating margin | 24% | −329% | 1.8% | −28% | −77% |
| Net margin | 8.7% | −335% | −49% | −282% | −90% |
| What growth costs, per sales dollar | |||||
| Research and development | 1.4% | 4.2% | 1.4% | 2.3% | - |
| Sales and marketing | 0.7% | 2.0% | 1.4% | 2.0% | - |
| General and admin | 11% | 33% | 19% | 22% | - |
| Pay in shares | 7.2% | 29% | 12% | 10% | 31% |
| Capital spending | 0.0% | 60% | 3.2% | 19% | 229% |
| Cash | |||||
| Cash from operations | −$56.7M | $205M | $65.1M | $42.9M | $278M |
| Free cash flow | −$56.7M | −$179M | $49.0M | −$52.1M | −$451M |
| Balance sheet | |||||
| Cash and short-term investments | $118M | $15.9M | $50.4M | $836M | $311M |
| Debt | $728M | $36.2M | $808M | $1.09B | $1.06B |
| Net cash | −$610M | −$20.4M | −$758M | −$254M | −$749M |
| Returns and shares | |||||
| Return on capital | 9.1% | - | - | - | - |
| Shares (diluted) | 233M | 341M | 380M | 256M | 318M |
What growth costs
It is too early to say whether growth will make Core Scientific more profitable. Building costs billions while the AMD rent has not started, and write-offs from leaving mining fill this year's accounts. Management expects about $30 million a quarter of cash overhead against about $1.6 billion a year of rent profit once everything is built.
The evidence under this - 3 claims
These are presentation figures; NOI is a non-GAAP measure, and it leaves out corporate overhead of about $30M a quarter in cash (C46).
The last report (5 Nov)
Quarter ended September 2026, filed figures.
The third-quarter 2026 report, out on 5 November, showed the switch nearly done. Revenue was $232.9 million against $81.1 million a year earlier, and the business made $72.4 million of operating profit. Net income of $1,178 million came from an accounting gain on the value of its warrants, not from operations.
On the call management said the AMD halls cost $11 to $12 million a megawatt, that its own share is already paid from cash, and that bonds raised against the projects, not new shares, will fund the rest. Mining ends this quarter.
What it did to the thesisThe rent is arriving on schedule and the business now earns an operating profit, but the AMD financing is still a promise and CoreWeave's weight grew, so the case holds without getting stronger.
The quarter in detail
Sales growth from the previous quarter, filed figures. The company has restated its own history, so a quarter either side of the restatement may describe a different business; n/c marks one that does not compare.
Earnings per share $0.47 against $-0.05 expected; $0.47 as filed, with no adjustments.
The evidence under this - 6 claims
The net income is a warrant mark: the second quarter's net loss of $1,155.3M came from the same line moving the other way.
The March release targeted about 590 MW by early 2027 and the date still holds, but Dalton's second phase is the one building still under construction.
No bond has priced; the 10-Q says the company expects project-level bonds within twelve months, and analysts on the call asked whether lenders had tightened terms.
What management did with the money
Insiders in 2026Q3: 4 purchases and 23 sales, from their Form 4 filings.
The price, and where we disagree
In short. Fair. The price pays a little less for each dollar of signed rent than landlords usually fetch, about right while the main tenant's credit is weak and the AMD halls are unfunded.
The shares have fallen from $30.46 in June to $13.58, half their value in six months, as AI data-centre stocks sold off and doubts grew about CoreWeave's borrowing. They fell 10.8% on the day of a third-quarter report that beat revenue estimates. In January about 95 million warrants expire, bringing about $650 million of cash and more shares.
The next test is the fourth-quarter report, due about 1 March 2027: whether all of CoreWeave's halls pay rent, mining is gone and the AMD debt is raised without new shares.
The evidence under this - 7 claims
The same week's selling hit other AI-infrastructure names, and CoreWeave's contract-backed secured debt prices far tighter than its unsecured notes.
The warrant cash depends on the shares staying above $6.81 into January, which at $13.58 they are.
The March release targeted about 590 MW by early 2027 and the date still holds, but Dalton's second phase is the one building still under construction.
No bond has priced; the 10-Q says the company expects project-level bonds within twelve months, and analysts on the call asked whether lenders had tightened terms.
How we value it
No fair value: the filed years describe a bitcoin miner being shut, nine months of 2026 lost money after an impairment and stock pay, and the leasing business has billed for only a few quarters, so a value built on earnings would rest on a guess; the bet prices it instead.
How we read the multiple
Count what a buyer of the whole company takes on: about 509 million shares once warrants and convertible notes are counted, worth about $6.9 billion, plus the secured debt less cash, plus about $4.4 billion still to spend on building. About $12.3 billion.
Against about $1.6 billion of yearly rent profit once everything is built, that is about 7.7 times. Landlords with strong tenants often fetch about ten; CoreWeave's 15% bond yields say its rent deserves less.
Nothing is paid for AMD's right to reserve up to 1,925 megawatts more.
Against its own history. Its own past ratios describe a bitcoin miner, not a landlord, so this comparison carries little weight.
The evidence under this - 8 claims
The commitments include greenfield work beyond the 1.1 GW and pass-through items, and the NOI is a stabilised figure reached only in 2028 (C07), before about $120M a year of cash overhead.
These are presentation figures; NOI is a non-GAAP measure, and it leaves out corporate overhead of about $30M a quarter in cash (C46).
The same week's selling hit other AI-infrastructure names, and CoreWeave's contract-backed secured debt prices far tighter than its unsecured notes.
The Tranche 1 warrants date from the 2024 bankruptcy exit and are not new issuance; their cash funds the next builds.
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).
Private or foreign rivals that cannot be priced: Aligned Data Centers, Compass Datacenters, NTT, QTS, Switch, Inc., Vantage Data Centers, CyrusOne Inc., Equinix, Inc.
The bet: what it pays if it works
This is a bet that AMD's halls get built and that AMD also leases the grid power Core Scientific has secured at Hunt and Muskogee. Even then the company would be worth only a little more than today's price once the warrants are counted, a modest reward for the risk.
If leasing fails, CoreWeave stops paying or the AMD build stalls unfunded, little is left for shareholders: the secured notes and the convertibles are paid first.
| AMD-contracted colocation revenue in two to three years | Leased critical IT megawatts needed | At the stated multiple |
|---|---|---|
| $930.00M | about 528 at 100% kept | 1.0× |
| $1.61B (the big win) | about 915 at 100% kept | 1.4× |
| $2.67B | about 1.5K at 100% kept | 2.0× |
To pay 2× (a Fair bet) it would need about $2.61B of AMD-contracted colocation revenue (about 1.5K leased critical IT megawatts); to pay 5× (a Cheap bet), about $7.79B of AMD-contracted colocation revenue (about 4.4K leased critical IT megawatts), at the same multiple.
Each row is AMD-contracted colocation alone, plus the rest of the company (CoreWeave colocation) counted at about $3.40B.
The steps it has to take
- Done 590 MW leased to CoreWeave and 485 MW billing
- Done 530 MW leased to AMD and an AMD-backed neocloud
- Done Polaris's 440 MW bought beside Muskogee
- In progress AMD project bonds raised without new shares due 2027-06
- In progress First AMD halls at Pecos and Auburn delivered due 2027-06
- In progress ERCOT's final Batch Zero allocation for Pecos and Hunt due 2027-06
- Not started AMD exercises a reservation tranche due 2027-12
- Not started All 530 MW billing due 2028-12
The detail behind the label
The evidence under this - 6 claims
The behind-the-meter power needs a pipeline at Pecos and air permits that have not been filed.
The deadlines are undisclosed by agreement with AMD, and the reservation commits AMD to nothing.
The same week's selling hit other AI-infrastructure names, and CoreWeave's contract-backed secured debt prices far tighter than its unsecured notes.
No bond has priced; the 10-Q says the company expects project-level bonds within twelve months, and analysts on the call asked whether lenders had tightened terms.
Where we disagree with the market
Nineteen of twenty-two analysts rate it a buy, with targets about twice the price: they see a large AI landlord with AMD's reservation still to come.
The signed rent is worth about today's price; AMD's reservation is real but commits AMD to nothing, and its halls need new debt.
The difference is in who pays: CoreWeave pays 93% of revenue and its own bonds trade like junk, while the AMD halls that dilute that exposure are not yet built or financed.
We would be wrong if the AMD debt is raised with no new shares and the first AMD halls pay rent in 2027, or if CoreWeave's credit improves sharply.
What analysts say, and the evidence
The evidence under this - 7 claims
The commitments include greenfield work beyond the 1.1 GW and pass-through items, and the NOI is a stabilised figure reached only in 2028 (C07), before about $120M a year of cash overhead.
The deadlines are undisclosed by agreement with AMD, and the reservation commits AMD to nothing.
The same week's selling hit other AI-infrastructure names, and CoreWeave's contract-backed secured debt prices far tighter than its unsecured notes.
No bond has priced; the 10-Q says the company expects project-level bonds within twelve months, and analysts on the call asked whether lenders had tightened terms.
Pecos and Hunt still await ERCOT's final Batch Zero allocation and air permits at Texas's environment regulator (C20, C50).
What could go wrong, and how we would know
The risks
- One tenant carries the money. CoreWeave paid 93% of third-quarter revenue, and its bonds yielded near 15% in October.
- The AMD build is unfunded. About $4.4 billion of building commitments fall due within two years, and the project bonds have not priced.
- Your slice is smaller. About 322 million shares are out, and warrants and convertible notes take that to about 509 million.
- Texas has not finished deciding. Pecos and Hunt are approved only conditionally by the grid operator, and air permits for backup generators are pending.
- The profit line swings wildly. Warrant values moved net income by more than a billion dollars in each of the last two quarters.
The evidence under this - 9 claims
The same week's selling hit other AI-infrastructure names, and CoreWeave's contract-backed secured debt prices far tighter than its unsecured notes.
No bond has priced; the 10-Q says the company expects project-level bonds within twelve months, and analysts on the call asked whether lenders had tightened terms.
The Tranche 1 warrants date from the 2024 bankruptcy exit and are not new issuance; their cash funds the next builds.
Conditional approval is not allocation: ERCOT sets the final load ramp only when Batch Zero ends.
Analysts asked three questions on Texas politics and permits, and a new governor takes office in January.
The net income is a warrant mark: the second quarter's net loss of $1,155.3M came from the same line moving the other way.
What we still do not know
- What rent each lease pays per unit of power; only the totals are public.
- Who the AMD-backed cloud company is, and how much of the 530 megawatts it holds.
- What the AMD project bonds will cost and how much they will fund.
- When AMD must decide on its reservation; the dates are kept private.
The evidence under this - 4 claims
The CoreWeave figure is lowered by the build credits it paid for, and both totals are presentation numbers (C03).
The neocloud is unnamed and its own credit is undisclosed; AMD's support terms are not filed.
No bond has priced; the 10-Q says the company expects project-level bonds within twelve months, and analysts on the call asked whether lenders had tightened terms.
The deadlines are undisclosed by agreement with AMD, and the reservation commits AMD to nothing.
What would change our view
- Neutral→AttractiveAMD build financed cleanly and delivering
It needs all of:
- ○The AMD project bonds raised with no new shares or convertible notes.watched as it happens
- ○The first AMD halls at Pecos and Auburn paying rent by 30 September 2027.watched as it happens
- Neutral→UnattractiveThe funding, the build or CoreWeave gives way
It needs any of:
- ○New shares or convertible notes sold to pay for the AMD build.watched as it happens
- ○The first AMD halls slip past the end of 2027.watched as it happens
- ○CoreWeave misses rent, renegotiates or ends a lease.watched as it happens
We would also be wrong if
- AMD lets its reservation lapse through 2027 without leasing more.
What is coming
- 23 January 2027 Tranche 1 warrants expire and are exercised: Whether about $650M of cash arrives for the AMD build, and the share count settles near 509M.
Every dated event ahead
- 23 January 2027 Tranche 1 warrants expire and are exercised: Whether about $650M of cash arrives for the AMD build, and the share count settles near 509M.
- 1 March 2027 Q4 2026 results and the 2026 annual report: The double-digit revenue rise, mining ended, Dalton's last 100 MW billing, and whether AMD project bonds have priced.
- 30 June 2027 First AMD halls at Pecos and Auburn: Whether the AMD schedule holds at $11-12M per MW, and ERCOT's final Batch Zero allocations.
- 31 December 2027 AMD decides on the larger reservation tranches: Whether AMD exercises more of its 1,925 MW at returns like the first 530 MW.
Since we wrote this
- 17 NovThesis writtenFirst thesis: fair for signed rent, AMD's build unfundedNeutral with medium conviction: about 1.1 GW is leased to CoreWeave and AMD, but CoreWeave's credit is junk-rated and AMD's halls still need their debt raised.
Sources 21
Sources - 21
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 filing2026-03-02Item 1: the shift to high-density colocation, the power table of 1,426 MW gross utility capacity across ten sites, the Competition section naming nine data-centre landlords and five converted miners, 325 employees, and the 2025 statementsbacksC34C41C47
- S02primarycompany filing2026-11-05the AMD Leases and the 1,925 MW reservation right, the AMD warrant, the Polaris acquisition, the debt note (secured notes, two convertibles, the credit facilities), the warrants, the $4.4B of capital commitments and the liquidity plan, revenue concentration, and 322,236,200 shares outstanding on 2 NovemberbacksC01C05C06C09C10C11C12C17C18C19
- S03primarycompany release2026-11-05the statements of operations and cash flows, the balance sheet, the segment results, stock pay and the adjusted EBITDA reconciliationbacksC02C10C14C15C16C46
- S04primarycompany release2026-11-05about 1.1 GW contracted (590 MW CoreWeave by campus, 530 MW AMD and an AMD-backed neocloud by campus), $24B+ of contracted revenue, $1.8B of average annual revenue and $1.6B of NOI, NOI margins by counterparty, the CoreWeave contract summary, the AMD reservation power pathways, the AMD delivery schedule and the fully diluted count of about 509MbacksC01C02C03C04C07C13C22C26C27
- S05primarytranscript2026-11-05Core Scientific third-quarter 2026 earnings call transcript485 MW billing and about 100 MW in commissioning, the AMD build cost of $11-12M per MW, the equity already invested, project bonds as the expected financing, about $650M from Tranche 1 warrants in January, the mining exit by year end, cash SG&A of about $30M a quarter, the ERCOT and Texas permitting status, the AMD reservation deadlines, and analyst turns from Needham, Lucid, Cantor, Jefferies, B. Riley, BTIG, Craig-Hallum, Compass Point, KBW, JonesTrading, UBS and Clear Street
We do not expect to issue additional equity or equity-linked securities to fund the first phase of the AMD build-out.
backsC02C03C04C07C08C13C18C22C23C26C27C46C50C51 - S06primarycompany release2026-07-28the AMD partnership: about 530 MW on 15-year agreements across five sites, more than $14B of potential base contracted revenue and up to 2.5 GW of leasable capacity; leased capacity raised to about 1.1 GWbacksC03C05
- S07primarycompany filing2026-09-10Denton's 297 MW as longstanding existing load outside Batch Zero, Pecos's 300 MW conditionally approved as base load and 300 MW as studied load, and Hunt's 431 MW conditionally approved under pathway (e)backsC20
- S08primarycompany release2026-08-14about 440 MW of in-service grid power at Muskogee bought for about $444M in cash, and the plan to scale Muskogee to about 1.5 GW grossbacksC17
- S09primarycompany filing2026-03-31the executive officers and their tenure, the beneficial ownership table at 23 March 2026, the 2025 bonus metrics and the 2025 long-term awardsbacksC43C44C45
- S10primaryvendor data2026-11-17Research pack, frozen at content/CORZ/research-pack.jsonthe 16 November close of $13.58, the 52-week range and returns, the price-move decomposition, the analyst rating series and target drift, FINRA consolidated short interest, the insider base rate and the analysts' revenue pathbacksC28C29C30C52
- S11primarycompany filing2026-11-1750 open-market sales of 320,000 shares, every one under a Rule 10b5-1 plan, and 4 discretionary director purchases of 26,000 shares, read off each Form 4's own checkboxbacksC31
- S12primaryregulator2026-06-18the grid operator's own measure of how scarce a Texas connection is, and the batch process Core Scientific's Texas sites sit in; read with curl and pdftotext
As of mid-2026, large users totaling more than 438,000 MW of proposed demand had entered ERCOT's connection queue. Nearly 90% of those requests were from data centers. To put that in perspective, ERCOT's all-time peak demand record is 85,508 MW
backsC21 - S13primarycompany filing2025-10-30the counterparty's own account: the all-stock takeover ended when Core Scientific's holders voted it down, and the colocation partnership continues
We respect the views of Core Scientific stockholders and look forward to continuing our commercial partnership
backsC25 - S14primarygovernment data2026-09-15the underlying rate of change, measured by a statistical agency rather than the industry; quote as frozen by the Applied Digital and Cipher theses' read of the same page
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 - S15secondaryindustry data2026-09-15Synergy Research Group - hyperscale operators to account for 67% of all data center capacity by 2031 Synergy Research Groupthe size of the substitute: how much of their own capacity the buyers build rather than lease
Almost 60% of that hyperscale capacity is now in own-built, owned data centers with the balance being in leased facilities … they now account for 48% of the worldwide capacity of all data centers … by 2031 hyperscale operators will account for 67% of all capacity
backsC36 - S16primaryindustry data2026-08-11a broker's North American absorption, capacity under construction and the owner-occupied part of it
North America data center demand reached an all-time high in the first half of 2026 with 25 GW of absorption, double the level from a year ago and five times that of two years prior. … 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.
backsC35 - S17secondaryindustry data2026-09-15vacancy at 1.4%, 80.4% of capacity under construction pre-committed, under 1,500 MW available: the scarcity the moat rests on, measured outside the company
primary market vacancy fell to a record low of 1.4% in H1 … commitments made on 80.4% of all under-construction capacity … Total under-construction capacity across primary markets increased by 24.8% in H1 to a record high of 7,481.1 MW … Less than 1,500 MW of future capacity across all primary markets remains available, representing approximately six months of demand
backsC37 - S18primarycompany release2026-11-02a competitor selling the same product to the same tenants: 700 MW leased to Amazon, a Google-backed Fluidstack and a hyperscaler, about $16.6bn of contracted revenuebacksC39
- S19primarycompany filing2026-07-29a named competitor selling the same product to the same buyer: approximately 15-year leases to investment-grade hyperscalersbacksC40
- S20primarycompany release2026-03-02fourth-quarter 2025 revenue of $79.8M with colocation at $31.3M, about 350 MW energized for CoreWeave and the target of about 590 MW by early 2027, and the Hunt County sitebacksC47
- S21secondarynews2026-10-14investingLive, 'CoreWeave bonds trade like distressed debt as AI spending outpaces balance sheets' investingLivebond yields on October 14, the S&P B+ rating, the 16% share fall that day; it reports CNBC's story, which refused this repository's fetcher
Its 9.625% 2032 notes yielded about 14.8%, briefly touching 15%, and its 9.75% 2031 notes yielded around 14.9%
backsC24