28 July 2026; financial period ended 30 June · Statements; notes 6–8 and 13–15; liquidity, commitments, controls, and later July common count. Primary financial source.
https://www.sec.gov/Archives/edgar/data/1839341/000183934126000014/core-20260630.htmWatchlist · delivery and funding must work together
The investment decision
At $13.10, the funded base returns 11.3% a year over five years. A conditional 20% entry is $9.01. The decisive question is how much rental value remains after construction, debt and dilution.
Edition update. 26 October Sydney initiation. Reference: 23 October US regular close. Financial actuals end in June; July–September developments are identified separately. Independent Capital Thesis assumptions, with dated broker material used as context.
01. The AI opportunity is real; the equity still needs a funding test
Research view: watchlist, with a conditional construction-and-financing case. At the $13.10 reference, our base produces 11.3% a year over five years. A 20% hurdle requires an entry of $9.01, holding the operating, funding, dilution and exit assumptions fixed. The stock is not automatically cheap because management describes more than $24bn of potential contracted revenue. That revenue spans long contracts, arrives after delivery, and sits above substantial construction costs and financing claims. [S2] [S3]
Core Scientific owns and operates the powered infrastructure used by AI customers. Its existing sites and utility relationships provide a starting advantage over a developer assembling every input from scratch. The proof is tangible: billing capacity rose from 120 MW at December 2025 to 395 MW at June and 437 MW by mid-July. The investment debate has moved from whether conversion can work to whether the next, much more capital-intensive wave produces attractive cash for each share. [S1] [S2]
The July agreements improve customer diversification, but they do not make every dollar of the pipeline equivalent. Of the new 529 MW, 377 MW is leased directly to AMD and 152 MW to an unnamed neocloud with conditional AMD support. The original 590 MW CoreWeave portfolio has a particularly favourable customer-funded conversion structure. The new build programme requires a different financing underwrite. [S2] [S4]
Evidence confidence is high for the filed historical numbers and disclosed contract structure, but forward capital costs, financing availability, customer performance and terminal valuation remain assumptions. This is an absolute-return research watchlist, not a personalised position recommendation. A stronger ownership case needs delivery, rent collection, secured funding and disciplined share issuance to be demonstrated together.
02. Dates and definitions prevent a misleading capacity story
Reference price: USD 13.10 on 2026-10-23, the regular-session close at 4 pm EDT (7 am on 24 October in Sydney). Research date: 26 October 2026, Sydney. The latest full financial statements reviewed end on 30 June; July operating and contract updates, the August acquisition and financing, and the September Texas power update are later disclosures. The report does not invent September-quarter results before their scheduled release on 5 November. [S1] [S3] [S5] [S6] [S7] [S8] [S9]
| Capacity measure | June 2026 filed | Later disclosed update | What it means |
|---|---|---|---|
| Gross utility power | 2,115 MW | Polaris added about 440 MW gross in August | Utility power is not customer IT load |
| Total leasable IT capacity | 1,275 MW | Expansion opportunities remain conditional | Potential customer load, signed or unsigned |
| Signed customer IT capacity | 590 MW | 1,119 MW after July agreements; rounded to about 1.1 GW | 590 CoreWeave + 377 AMD + 152 neocloud |
| Billable customer IT capacity | 395 MW | 437 MW as of mid-July | Service has commenced and billing has started |
The 437 MW billing snapshot predates the 27 July agreements. It is not an October measurement and should not be silently updated to the larger signed total. Likewise, the approximately 1,925 MW reserved expansion and more than 2 GW of prospective new sites are not signed, operating revenue. Nameplate power, utility approval, completed infrastructure, leased IT load and billable IT load are separate milestones. [S1] [S2] [S4]
The model uses the exact 529 MW sum from the filing; management often rounds it to 530 MW. The existing 590 MW plus 529 MW equals 1,119 MW, rather than a different economic portfolio. No interpolation is used to invent missing monthly billing data.
03. An AI landlord, with a mining business in transition
The product is access to reliable power, cooling, space, networking and operating support for dense computing equipment. Customers supply the computing hardware; Core Scientific builds and operates the supporting facilities. Its advantages are site access, power relationships, conversion experience and construction execution. Those are valuable in a constrained market, but they are not substitutes for completed buildings, creditworthy customers or reasonably priced capital. [S1] [S2]
Colocation was 77% of first-half revenue and 83% of the second quarter, so treating the company as only a Bitcoin miner is stale. Mining still contributes revenue and asset-sale liquidity during the transition, while mining-related impairments show the cost of changing the asset base. The model assigns no continuing mining value after the transition. That is deliberately more cautious than capitalising both the old mining capacity and its replacement AI use. [S1]
| USD million unless stated | FY2024 restated | FY2025 | H1 2026 | Q2 2026 |
|---|---|---|---|---|
| Total revenue | 510.7 | 319.0 | 279.4 | 164.2 |
| Colocation revenue | 24.4 | 65.4 | 214.2 | 136.7 |
| Net loss | (1,437.9) | (288.6) | (1,502.5) | (1,155.3) |
| Operating cash flow | 42.9 | 278.3 | 230.9 | Not separately used |
| Cash PP&E purchases | 95.0 | 729.0 | 954.2 | Quarterly headline includes land |
Annual and half-year periods are not directly interchangeable. The old mining-heavy revenue base fell before colocation scaled, so a generic five-year revenue CAGR would obscure the business change. The more useful operating series is colocation revenue of $31.3m, $77.5m and $136.7m in Q4 2025, Q1 2026 and Q2 2026 alongside billing capacity of 120, 225 and 395 MW. [S1] [S2] [S10]
04. Two contract structures, different cash economics
| Portfolio | Signed IT load | Revenue context | Funding and credit distinction |
|---|---|---|---|
| CoreWeave | 590 MW | More than $10bn potential; about $850m lifetime-average annual GAAP colocation revenue | Customer funds much of conversion; credits reduce early cash bills |
| AMD direct | 377 MW | Part of new agreement group with more than $14bn potential base revenue over 15 years | Direct AMD counterparty; developer funding required |
| Neocloud | 152 MW | Included with AMD agreement group, not additional to it | Conditional AMD support; not an unconditional full guarantee |
| Reserved AMD expansion | 1,925 MW | Not part of the signed 529 MW | Rights and conditions through December 2028; no base-case revenue credit |
Management describes the CoreWeave arrangement as take-or-pay, with an annual escalator and no unilateral tenant termination. The anticipated 80–85% site profit margin is a management claim, not a consolidated net margin. Most contracts span 12 years with extension options; Austin is a shorter seven-year contract with elective extensions. Contract rights reduce utilisation risk, but customer default, enforcement and joint delivery risk remain. [S2]
Up to approximately $750m of CoreWeave-funded build costs is credited against hosting payments at no more than 50% of monthly fees until repaid. Other modification equipment is purchased directly by the customer and contributed for facility use. The additional 70 MW expansion requires approximately $104m from Core Scientific without that credit. The headline is therefore not that every project can be built for almost no landlord capital. [S2]
The July filing describes AMD rights to cure certain neocloud defaults, rather than an obligation to cure every failure. Support can terminate in specified circumstances, including certain insolvency/default events and a company representation breach after a cure period. We therefore distinguish direct AMD leases from supported neocloud exposure. Counterparty diversification improves, but the current signed portfolio remains concentrated. [S4]
The approximately $1.8bn combined average annualised GAAP revenue cited in the presentation is a lifetime-average illustration. It is not current annual cash rent, a quarterly run rate or a forecast of free cash flow. Our model sets its own starting rents and escalators, then subtracts credit offsets and costs explicitly. [S2]
05. Revenue and accounting losses each need a bridge
Q2 colocation revenue of $136.669m comprised $98.812m of licence fees, $35.073m of power passed through at no markup, and $2.784m of maintenance and other revenue. A higher power bill can increase reported revenue and cost by the same amount without improving gross profit. Colocation gross profit was $79.983m: about 58.5% of gross reported revenue, or 78.7% after excluding pass-through power from the denominator. This adjusted denominator is an analytical comparison, not a new GAAP margin. [S1]
The $1.155bn Q2 net loss was dominated by a $1.046bn non-cash fair-value change in warrants and contingent value rights. Such charges can become larger when the share price rises; they are not equivalent to the same amount of cash leaving the business. The claims themselves still matter because exercise or settlement affects equity ownership. Adjusted EBITDA of $41.1m strips out those items and other costs, but it does not pay construction invoices, debt principal or interest. [S1] [S3]
Reported EPS is also easy to misread. The Q2 loss available for the EPS calculation is adjusted by $74.234m for Tranche 2 penny warrants, which are included in the basic denominator. Thus ($1,155.310m loss less $74.234m adjustment) divided by 325.329m weighted shares gives approximately $3.32 loss per share. Simply dividing the full net loss by weighted shares gives the wrong result. The model avoids P/E and PEG for this transition. [S1]
An unresolved material weakness concerned accounting controls for intended demolition of mining assets during conversion, including impairment evaluation and insufficient trained personnel. It led to restatements of earlier periods. At June 2026, controls were still ineffective; management expected its remediation programme to be sufficient for the year-end assessment, conditional on operating and testing the controls. Remediation is a monitoring item, not an achieved fact. [S1] [S10]
06. Positive operating cash flow is not free construction funding
| H1 2026 cash measure | USD million | Interpretation |
|---|---|---|
| Reported operating cash flow | 230.949 | Includes asset monetisation and customer prepayments |
| Included Bitcoin sale proceeds | 208.249 | Drawdown of mining-generated holdings, not all current rental cash |
| Cash PP&E purchases | (954.244) | Construction and equipment purchases |
| Land and development-rights acquisitions | (232.500) | Hunt County; separate from cash PP&E |
| Net investing cash flow | (1,182.891) | Includes other purchases and disposal proceeds |
| CFO plus net investing | (951.942) | Funding gap before financing |
| Net financing cash flow | 3,191.955 | Debt issuance dominated; not operating profit |
Removing Bitcoin sale proceeds from CFO leaves $22.7m, but that remainder is not a clean recurring rental-cash measure either. Customer prepayments, funding receivables, construction accruals and working-capital timing also affect CFO. We do not present it as sustainable owner earnings. CFO less cash PP&E was negative $723.3m before the land purchase. [S1]
Colocation deferred revenue was $654.4m at June. Customer advances fund construction now but are recognised later as service revenue or credits against bills. The model carries $555m of future cash-credit offsets across 2027–2029 after its H2 bridge; that is a house approximation anchored to the deferred balance, not an exact disclosed payment schedule. It prevents prepaid cash and future rent from being counted twice. [S1]
The CoreWeave revenues feed a lockbox associated with the $3.3bn secured notes. Cash inside a project is therefore not necessarily freely distributable to the parent. Completion guarantees, reserves, lender restrictions and principal amortisation can stand between attractive site economics and cash available to common shareholders. [S1] [S2]
07. The balance sheet is larger than the basic market value
| June 2026 capital item | USD bn | Treatment |
|---|---|---|
| Unrestricted cash | 1.769735 | Dated balance, not October bank cash |
| Restricted cash | 0.781656 | Includes debt-service reserve, Polaris escrow and construction escrow |
| Bitcoin at reported fair value | 0.049675 | Liquidation at this value is a model assumption |
| 7.75% senior secured notes | 3.300000 | Face amount; 2031 maturity must be refinanced or repaid |
| 3% 2029 convertible notes | 0.460000 | About $11 conversion price; debt or dilution |
| 0% 2031 convertible notes | 0.625000 | About $22.49 conversion price; debt or dilution |
| Total funded debt face | 4.385000 | Versus $4.297967bn carrying value after discounts |
| Operating / finance lease liabilities | 0.108499 / 0.000876 | Operating rent stays in margins; finance lease treated as a debt claim |
The August financing is $100m of revolving capacity and $500m of letters of credit. It is not $600m of cash received. Both facilities were undrawn at closing; management expected approximately $300m of restricted cash to be released. That release is not independently verified as an October cash balance and is not added as extra money to the model. The revolver can help working capital; the letters of credit support obligations and collateral efficiency. [S5]
The credit agreement requires at least $150m of liquidity, including unused revolver availability, at quarter-end. Each revolving borrowing also requires at least $3bn market capitalisation on the preceding trading day. A falling share price can therefore constrain access to funding. The agreement states 1.75% annual LC fees, a fronting charge and an unused-commitment fee; the report does not assume an unused facility is costless or freely available through every stress. [S5]
Polaris closed in August for approximately $444m, rather than the earlier $421m estimate, adding about 440 MW of gross grid-connected power. Gross power acquired is not immediately leasable, completed AI capacity. The planned Muskogee expansion to 1.5 GW gross and roughly 1.0 GW leasable remains a development plan. [S6]
The filed July 23 common count was 321.340m, implying about $4.21bn basic equity value at $13.10. Our schedule starts with the same-date June balance-sheet count of 319.587m and adds all June unvested awards and options; subsequent vesting must not be added a second time. June also had about 96.361m Tranche 1 warrants, 6.6m penny warrants and 69.611m shares underlying convertible notes. Management’s July presentation used approximately 508m fully diluted shares. Basic market capitalisation alone misses substantial claims. [S1] [S2]
The 96.361m legacy warrants are assumed exercised in 2027 at $6.81, bringing about $656m of cash and the same number of new shares. The penny warrants add shares with almost no cash. The base assumes conversion of the $460m notes in 2029 into approximately 41.8m shares and cash repayment of the $625m notes in 2031. Those settlement choices depend on conditions at maturity; the model does not claim to know the future share price on the exercise dates. [S1]
AMD received warrants for up to 30m shares at $23.47. Approximately 6.5m vested on the initial agreements; further vesting is linked to additional leased MW, and expiry is 27 July 2031. Base assumes no exercise of that block, while bull includes warrants earned by its extra 384 MW and exercise proceeds. Future employee grants add 2% to economic shares each year; they are not then charged again as a cash compensation reserve in EBITDA. [S4]
08. Building the new capacity is the critical capital-allocation decision
Management’s July average build-cost estimate was $11m–$12m per MW, excluding $1m–$2m per MW of tenant fit-out. The base uses $11.5m per MW for 529 MW, or $6.0835bn, with tenant fit-out excluded from landlord spending. The customer-funded CoreWeave conversion economics are not transplanted onto this new programme. A tenant-fit-out cost falling on Core Scientific would require additional funding and lower returns. [S2]
The model assumes 80% of AMD build cost can be financed with project debt at an 8.5% coupon and a 2% issuance fee; the remaining cost comes from existing cash, warrant exercise proceeds and operating cash. These loans are not claimed to be committed. They are assumed interest-only beyond the 2031 exit, with future refinancing risk retained. The ability to arrange them on these terms is a central condition, not a footnote.
| Opening 2027 bridge; USD bn | Amount | Basis |
|---|---|---|
| June cash including restricted cash | 2.551391 | Filed; not all available for discretionary use |
| Assumed Bitcoin monetisation | +0.049675 | June fair value, no price gain assumed |
| Polaris cash cost | −0.444000 | Includes use of acquisition escrow |
| Reserve remaining restricted cash | −0.350000 | House buffer; not counted as distributable cash |
| H2 cash operations before interest | +0.100000 | House assumption, net of operating effects and credit timing |
| H2 AMD / other net capex | −0.600000 / −0.300000 | House spending bridge |
| H2 project debt | +0.480000 | Assumed 80% funding of AMD spending |
| H2 interest / finance fees | −0.128000 / −0.009600 | House assumptions |
| Opening model cash | 1.349466 | Calculated, not a reported year-end balance |
This bridge starts with total cash because the Polaris purchase consumes acquisition escrow. It then spends the acquisition cost and retains a $350m restricted buffer. It does not subtract the acquisition once from unrestricted cash and again from escrow, nor add the expected $300m release on top of already-counted cash. The allocation is a conservative approximation of restricted uses, not a claim that every restricted dollar can be released. [S1] [S5] [S6]
June purchase and construction commitments were approximately $1bn, of which $264m was customer-funded. We reserve the full $736m net remainder: $300m in H2 and $436m in 2027, in addition to the AMD unit-build budget. The exact project overlap was not disclosed. This can double-count part of a commitment already within AMD construction; it is deliberately a contingency reserve rather than an unsupported omission. The $264m reimbursement is assumed to match its spending timing; delays would need more liquidity. [S1]
The senior notes are not left outstanding past maturity without explanation. The model amortises $189.75m in 2029 and $379.5m annually thereafter, then refinances approximately $2.541bn at the 2031 maturity after the first semiannual principal payment. Base applies an 8.5% new coupon and a 2% refinancing fee, using a half-year interest approximation. The June contractual schedule depends on commencement and credit abatement; a different first-payment date changes the cash path. [S1]
Base minimum model cash is approximately $640m in 2028, above the $300m house floor, so it needs no additional marketed common-equity issue on these assumptions. That does not mean no dilution: warrants, conversion and employee grants increase economic shares. Bear cannot self-fund its delays and cost overruns; the model raises $2.91bn gross equity at a fixed assumed $8 execution price. Such funding may be unavailable. If lenders or equity investors refuse financing, default, asset sales, restructuring or a zero-equity outcome can be worse than the numerical bear case.
09. The base requires a completed portfolio, not the whole pipeline
| Capital Thesis base | 2027 | 2028 | 2029 | 2030 | 2031 |
|---|---|---|---|---|---|
| Average CoreWeave billable MW | 590 | 590 | 590 | 590 | 590 |
| Average AMD/neocloud billable MW | 120 | 390 | 529 | 529 | 529 |
| Rent before credits, USD bn | 0.994 | 1.483 | 1.771 | 1.824 | 1.879 |
| Cash EBITDA before award dilution, USD bn | 0.720 | 1.166 | 1.426 | 1.469 | 1.513 |
| Cash interest, USD bn | 0.407 | 0.594 | 0.683 | 0.655 | 0.627 |
| End gross debt, USD bn | 7.147 | 9.253 | 8.603 | 8.223 | 7.219 |
| End cash, USD bn | 0.938 | 0.640 | 0.978 | 1.264 | 0.931 |
| End economic shares, million | 448.1 | 457.0 | 508.8 | 519.0 | 529.4 |
The model annual periods are FY2027–FY2031 as a practical approximation for a five-year 23 October 2026–23 October 2031 investment. It is not an exact monthly discount model. The separate H2 bridge precedes the five forecast years. End-2031 terminal operating levels stand in for the exit run rate; two months of fiscal-calendar approximation remain.
Base CoreWeave starting rent is $1.35m per MW per year, with 82.5% site margin. AMD/neocloud starts at $1.65m per MW, with a 92.5% blended site margin. Both rents grow 3% annually; these are house cash-rent assumptions, not the company’s lifetime-average GAAP revenue. The blended AMD margin reflects direct and neocloud terms but is not a filed contract margin. Cash overhead begins at $120m and grows 3%.
Maintenance capex is 3% of rent. Cash taxes use a 21% rate on positive EBITDA less an assumed 20%-of-rent depreciation allowance and interest. This is a normalisation rather than a tax-return forecast; NOL benefits and interest-deduction limits are not modelled in detail. Lease operating costs stay in site margins. EBITDA excludes cash interest, credit offsets, maintenance and tax, all of which are charged later in the cash bridge.
No additional uncontracted capacity, new land-value premium or mining terminal value is included in base. The bull adds only 384 MW from the identified grid-linked expansion opportunity, assumes $4.224bn additional construction spend and a stronger 15x exit. It does not capitalise all 1,925 MW of reserved capacity or the larger load-study and behind-the-meter pipeline.
10. Five-year equity returns after debt and dilution
| House case | 2031 EBITDA / multiple | 2031 net debt | Economic shares | Terminal value | Annual return |
|---|---|---|---|---|---|
| Bear | $1.080bn / 9x | $6.926bn | 867.7m | $3.22 | -24.5% |
| Base | $1.513bn / 12x | $6.288bn | 529.4m | $22.42 | +11.3% |
| Bull | $2.300bn / 15x | $8.152bn | 569.2m | $46.30 | +28.7% |
Bear five-year total-wealth CAGR: -24.5%. Base five-year total-wealth CAGR: 11.3%. Bull five-year total-wealth CAGR: 28.7%.
Bear combines slower deployment, $13m/MW AMD build cost, 90% rent collection, lower site margins, 70% debt funding at 11%, equity issued at $8 and a 9x exit. It repays both convertible notes in cash. Base uses only signed capacity, $11.5m/MW, 80% funding at 8.5% and 12x. Bull needs added grid-linked capacity, lower unit construction cost, 7.5% new debt and a 15x multiple. The assumptions move together because execution and access to capital are related risks.
Base terminal enterprise value is $18.159bn. Subtract $7.219bn gross debt and add $0.931bn cash to obtain $11.871bn equity. Divide by 529.389m economic shares for $22.42. Restricted reserves are not added again. Warrant liabilities are not subtracted as cash debt after their exercise/dilution has already been modelled.
Total-wealth CAGR = ((terminal share value + cumulative cash distributions) / $13.10)^(1/5) − 1. Common dividends and buybacks are zero in every case, so wealth equals terminal share value. Returns are nominal USD before investor tax, fees and foreign-exchange effects. Cases are conditional illustrations without probability weights; they are not broker targets or guaranteed outcomes.
Base five-year total-wealth CAGR without rerating: 7.6%. This uses a 10.764x reference economic-EV / 2027 house EBITDA benchmark, held at exit instead of 12x. It is a forward-model comparison, not an unchanged historical P/E or reported EBITDA multiple.
The comparison uses 481.108m economic shares at $13.10 (June common, awards, options, legacy warrants and assumed conversion of the $460m notes), adds remaining funded debt and finance leases, and subtracts June unrestricted cash, Bitcoin and legacy warrant proceeds. That gives $7.753bn economic EV, divided by $0.720263bn 2027 house EBITDA. The $625m notes remain debt and restricted cash is not offset. Forecast spending and the terminal cash/share path remain identical to base. This dated pro forma bridge is explicitly not the company’s observed October capitalisation.
11. The exit multiple and cash-flow cross-check constrain the story
A 12x cash-EBITDA exit is a Capital Thesis judgement about a completed, contracted data-centre portfolio. It is not an observed peer average or a promise that a finite lease is worth a perpetuity. By 2031, some original contracts will have materially less remaining term; re-leasing, extension economics, physical asset value and continued customer demand therefore matter. The multiple already gives credit for value beyond the first five forecast years. A terminal haircut is appropriate if those residual economics weaken.
| Financing-neutral enterprise DCF | 10x exit | 12x exit | 14x exit |
|---|---|---|---|
| 10% discount rate | $9.54 | $13.22 | $16.79 |
| 12% discount rate | $7.75 | $11.27 | $14.53 |
| 14% discount rate | $6.15 | $9.39 | $12.50 |
The 12% discount-rate / 12x exit cross-check is $11.27 per share. It discounts unlevered operating cash after credits, maintenance, normalised tax, construction and the commitment reserve, then uses the same exit EBITDA basis. It is an exit-multiple DCF, not a Gordon-growth perpetuity. The terminal component is substantial, so changing the discount rate does not remove residual-value risk.
The DCF is a financing-neutral check rather than the same funded equity path. It excludes future debt fees and leverage tax shields, assumes no new marketed equity raise, applies a 2% annual future-award haircut to the common base, and allocates convertible claims at the greater of face or conversion value. Legacy warrant exercise cash is credited once. It does not pretend out-of-money notes disappear into shares. The funded scenarios remain the controlling shareholder-return model.
| Base exit multiple only | Terminal price | Conditional 20% entry |
|---|---|---|
| 8x | $10.99 | $4.42 |
| 10x | $16.71 | $6.71 |
| 12x | $22.42 | $9.01 |
| 14x | $28.14 | $11.31 |
| 16x | $33.86 | $13.61 |
This sensitivity fixes the base funding, prior convertible settlement, awards and cash path, then changes the final exit multiple. It does not reoptimise conversion or funding under every share-price outcome. A full downside re-underwrite would change those jointly, as bear does. No current comprehensive peer-multiple, institutional ownership, float, borrow or short-interest dataset was independently verified; the report does not invent those inputs.
The supplied September initiation used a $28 broker target combining contracted value and pipeline optionality; the supplied July quick take used $37. These are dated third-party opinions with different assumptions and target horizons, not consensus and not our five-year terminal values. They informed the diligence questions but do not substitute for the after-financing model.
12. About $9 is a conditional hurdle, not a standing buy instruction
The conditional base entry for a 20% five-year total-wealth CAGR is $9.01 per share, with no common dividends and all operating, funding, refinancing, cash, dilution and exit-multiple assumptions held fixed.
The calculation is $22.42 terminal wealth / 1.20^5 = $9.01, using unrounded values. At the $13.10 reference, the same operations and cash/share bridge would require approximately 15.56x terminal cash EBITDA to deliver 20%, above the chosen 12x.
A lower share price can coincide with worse customer credit, construction delays or more expensive capital. It may also impair revolver access because of the market-cap condition. Entry sensitivity changes the investor’s purchase price only; it does not silently cheapen construction, improve rent collection, change assumed future equity-issue prices or remove refinancing risk. Re-underwrite the business if the evidence changes.
The cash-flow cross-check around $11 is a valuation under a 12% discount rate, while the approximately $9 entry asks for a 20% shareholder hurdle. Different required returns and claim-allocation methods produce different answers. Neither number is a guaranteed floor.
13. What would prove or break the investment case
| Evidence window | What to monitor | Effect on the thesis |
|---|---|---|
| 5 November results, before US market open | Billable MW, licence fees excluding power, cash-credit use and capex | Proves whether construction is becoming collected rent |
| Next financing disclosures | Committed project debt, coupon, covenants, equity contribution and completion support | Determines whether the $6bn new-build programme is fundable |
| Late 2026 / early 2027 | Remaining CoreWeave delivery, especially Dalton; legacy warrant exercise | Changes rental ramp, cash proceeds and common shares |
| 2027–2028 | AMD/neocloud deliveries against the announced phasing | Tests the new customer model and construction budget |
| Year-end control assessment | Demolition/impairment controls tested and remediated | Improves confidence in reported capitalisation |
| 2029 / 2031 | Convertible settlements and secured debt refinancing | Determines what terminal value remains for common owners |
The company announced the earnings call for 5 November at 8:30 am Eastern Time. That is 6 November at 12:30 am in Sydney after US daylight saving ends. The website calendar’s static EDT label should not override the announced Eastern local time. No earnings outcome is assumed. [S8]
The September power update is encouraging but conditional. Denton’s established load was outside Batch Zero; Pecos had conditional approval for existing and studied load, with final allocation and ramp still pending the process. Hunt had conditional approval and completed milestones. Conditional utility progress is not the same as fully energised IT capacity. The model excludes the large load-study and behind-the-meter option from base. [S7]
The positive case strengthens with completed billable capacity, collection after credits, committed funding near the model’s terms and limited equity issuance. It weakens if a high headline MW number is not accompanied by service commencement, if tenant support is overstated, if build cost rises without funding, or if debt service consumes the rental cash. These are observable tests of the thesis, not just changes in market sentiment.
Source reconciliation: the three supplied broker PDFs were retained with page-level extraction and hashes. We corrected the initiation’s isolated $1.4bn AMD revenue typo against its own $14.1bn table and the primary release; distinguished direct AMD credit from conditional neocloud support; used the $444m closed Polaris cost instead of its earlier estimate; and treated cap-rate/NAV forecasts as dated opinions. The EPS numerator difference is explained by the filed penny-warrant adjustment, not an unresolved financial discrepancy.
Current source limitations are the absence of a September balance sheet, an independently verified October cash balance, fully committed AMD project financing and detailed public lease cash schedules. The underwriting therefore remains conditional even though the historical evidence is well sourced. A successful business buildout does not by itself establish an attractive entry price.
Financial exhibits
Exact observations accompany every chart. Actuals, management guidance and Capital Thesis assumptions are labelled separately; the chart shapes do not imply probabilities.
Signed capacity still has to become billable
The disclosed billing ramp is substantial, but it is earlier than the research date.
| Observation | Billable capacity |
|---|---|
| Dec 2025 actual | 120 |
| Mar 2026 actual | 225 |
| Jun 2026 actual | 395 |
| Mid-July 2026update | 437 |
Point-in-time observations, not average quarterly billing. The 1,119 MW signed total includes agreements dated 27 July and is not inserted into this billing series. Sources: [S1] [S2] [S3]
Three counterparty exposures inside two agreement groups
Direct AMD credit and conditional neocloud support are different.
| Observation | Signed capacity after 27 July |
|---|---|
| CoreWeave | 590 |
| AMD direct | 377 |
| Neocloud | 152 |
Total 1,119 MW. Excludes 1,925 MW reserved expansion and prospective sites. No claim all signed capacity is operational. Sources: [S1] [S4]
Power reimbursement inflates revenue without adding gross profit
Licence fees are more informative than the gross revenue headline.
| Observation | Revenue |
|---|---|
| Licence fees | 98.812 |
| Power pass-through | 35.073 |
| Other services | 2.784 |
| Total colocation | 136.669 |
The first three bars sum to the final bar. Power costs equal the power-fee revenue; do not add the total a second time. Sources: [S1]
Construction spending exceeded operating cash
Positive CFO did not eliminate the need for outside financing.
| Observation | Signed cash flow |
|---|---|
| Operating cash | 230.949 |
| Net investing | -1,182.891 |
| Net financing | 3,191.955 |
CFO includes $208.249m Bitcoin sale proceeds and customer funding. CFO plus net investing was −$951.942m before financing. Sources: [S1]
Rent growth has to carry a larger funding bill
The base grows EBITDA, but substantial interest remains.
| Observation | Cash EBITDA before award dilution | Cash interest |
|---|---|---|
| 2027 house | 0.7203 | 0.4073 |
| 2028 house | 1.1663 | 0.5937 |
| 2029 house | 1.4264 | 0.6832 |
| 2030 house | 1.4692 | 0.6547 |
| 2031 house | 1.5132 | 0.6275 |
After site operating costs and cash overhead, before credit offsets, interest, maintenance, tax and growth capex. Employee economics enter via 2% annual share dilution. Sources: [S1] [S2] [S4]
The base does not meet a 20% hurdle at $13.10
Financing, delivery and the exit multiple shape equity outcomes.
| Observation | Shareholder return |
|---|---|
| Bear house | -24.4775 |
| Base house | 11.3495 |
| Bull house | 28.7257 |
No common dividends. Explicit funded cash/share paths; base only signed capacity, bull adds 384 MW. These cases have no probability weights; financing failure may produce zero equity. Sources: [S1] [S2] [S4] [S9]
Source register and research basis
Evidence checked for this edition on 26 October 2026. Dates below identify source vintages. Company guidance and house scenarios are conditional; source links and the downloadable model allow the assumptions to be inspected.
28 July 2026 · Pages 3, 5–6, 9–10, 12 and 14: capacity, delivery schedule, contract economics, construction-cost estimate and billing progression. Management claims distinguished from actuals.
https://d1io3yog0oux5.cloudfront.net/_b76ddaa46c59e7a64138659274ff5f32/corescientific/db/1066/11282/pdf/Q2+FY26+earnings+deck+7.28+AM.pdf28 July 2026 · Mid-July 437 MW and new contract context; GAAP revenue, reported cash and adjusted EBITDA. Power pass-through and footnotes retained.
https://investors.corescientific.com/news-events/press-releases/detail/139/core-scientific-announces-second-quarter-2026-results28 July 2026; agreements dated 27 July · 377 MW direct / 152 MW neocloud, conditional support, reservation rights and AMD warrant terms.
https://www.sec.gov/Archives/edgar/data/1839341/000183934126000012/core-20260727.htm27 August 2026; agreement dated 25 August · $100m revolver and $500m LC facility; undrawn at closing, fees, market-cap borrowing condition and liquidity covenant. Related issuer release describes expected restricted-cash release.
https://www.sec.gov/Archives/edgar/data/1839341/000183934126000021/core-20260825.htm14 August 2026 · $444m cash purchase; 440 MW gross grid-connected power; Muskogee expansion and next 82 MW delivery expectation.
https://investors.corescientific.com/news-events/press-releases/detail/142/core-scientific-completes-acquisition-of-polaris-ds-advancing-planned-muskogee-expansion-to-1-5-gw-of-gross-power10 September 2026 · Conditional ERCOT status of Denton, Pecos and Hunt. Final allocation and load ramp are not universally complete.
https://www.sec.gov/Archives/edgar/data/1839341/000183934126000023/core-20260910.htm15 October 2026 · Results before market open 5 November; call 8:30 am Eastern Time. IANA time zone conversion used for Sydney date.
https://investors.corescientific.com/news-events/press-releases/detail/145/core-scientific-inc-schedules-third-quarter-fiscal-year-2026-earnings-release-conference-call-and-webcast23 October 2026, 4 pm EDT regular close; accessed 26 October · $13.10 reference. Yahoo Finance corroborated; one alternate history showed $13.11. The selected regular-session close is frozen; overnight quotes excluded.
https://stockanalysis.com/stocks/corz/history/2 March 2026 · Audited annual financials and restated FY2024 comparative. The 18 March amendment only supplied a missing exhibit; it did not change financial statements.
https://www.sec.gov/Archives/edgar/data/1839341/000162828026013305/core-20251231.htmAccessed 26 October 2026 · Scoped filing-vintage check; latest reviewed financials remain June 2026. User broker PDFs are separate dated evidence, not issuer forecasts.
https://data.sec.gov/submissions/CIK0001839341.json