Phase Space AI

Financial Model Notes

Core Scientific [CORZ]

Core Scientific, Inc. [CORZ] — Financial Model Notes

As of: 2026-07-29 · Framework: v1.5.1 / criteria.md 2026-07-29 · All figures traced to primary EDGAR filings.

Per the brief, the screen's inputs were not trusted. Share count, TTM revenue and net cash were independently verified against primary filings, and net income ÷ shares ≈ filed EPS was cross-checked for scale. Every discrepancy is reported; none is silently adopted.

---

Verified inputs

EPS cross-check (net income ÷ shares ≈ filed EPS)

−$1,155,310,000 ÷ 325,329,000 = −$3.552 against a filed −$3.32. The gap is explained: the EPS numerator excludes the portion of the warrant-liability change attributable to Tranche 2 (penny) warrants, which are treated as participating. Order of magnitude and sign reconcile; no scale error. This is the closest EPS tie achieved on any name in the cluster.

Net cash reconstruction — the one name the screen got right on debt

CORZ is the only name in this cluster whose net cash the screen computed correctly — because it happens to tag its debt as LongTermDebtNoncurrent, which is the one tag the resolver looks for. The error is elsewhere: the $1,811,587,000 warrant liability was carried at zero, understating EV by 21.7%. On the screen's own spot of $20.74 its EV of $9,192.8m should have been $11,004.4m.

Modelling notes and hazards

1. Model the three segments separately — CORZ's own disclosure makes a blended model indefensible. Colocation earns a 59% gross margin; self-mining earns −56%; hosted mining ends 2026-12-31. A blended model applies an average of +59% and −56% to a mix that is changing by ~18pp a quarter.
2. Strip the power pass-through before applying any revenue multiple. $35.073m of Q2 2026 revenue is zero-margin power recovery. Colocation revenue ex pass-through is $101.596m, and the gross margin on it is 78.7%. A revenue multiple applied to the gross figure values a cost recovery.
3. The warrant liability and the share count are two views of the same claim. Add the $1,811.6m liability to EV or dilute by 133.0m shares — never both. This analysis adds the liability.
4. Reported net income is unusable. 90.5% of the Q2 2026 loss is a warrant mark that moves with the share price. Use the operating line, then strip the $41.9m contract termination and $19.5m held-for-sale remeasurement.
5. Five restated periodic filings. FY2024, FY2025 and all three 2025 quarters were amended on 2026-03-02 and 2026-03-18. Use post-amendment figures only. companyfacts contains both, distinguishable by the form field (10-Q versus 10-Q/A) — e.g. H1 2025 net loss appears as both −$356.106m and −$360.548m.
6. Terminal value exceeds 60% of EV, so the reverse DCF is the mandatory primary long-horizon output. Here it is runnable, because the exit multiple is IDENTIFIED — CORZ is the only name in the cluster where the required-path analysis rests on a growth-matched anchor rather than substituting for a missing one.
7. No RPO is tagged. Forward revenue must be built from the management-stated 590MW / 395MW capacity split and observed revenue per MW, which is an operational estimate rather than an audited contractual schedule. The revenue-per-MW figure derived here ($546.7m annualised ÷ 395MW ≈ $1.38m/MW/year) is an inference from two disclosed numbers, not a disclosed rate.

---

Reproduction

Commit: e67ab6405055a342b54a5ef03613a3650d59ab05