Phase A — Understand the business
Lens 1 · Company Overview
What it is. A vertically-integrated operator of high-power data center infrastructure across nine sites in five U.S. states (Georgia, Kentucky, North Carolina, Oklahoma, Texas), totaling 920 MW of total leasable customer power capacity, ~1.4 GW on a gross-utility basis. It owns the land, the power interconnects, the buildings, and the operating expertise.
Three reported segments:
- Colocation — leasing power-dense capacity to HPC/AI customers. License fees + pass-through power. FY2025 revenue $65.4M (gross margin 30%). One customer — CoreWeave — is 100% of this segment.
- Digital Asset Self-Mining — mining bitcoin for its own account. FY2025 revenue $229.2M (gross margin 5%). The legacy cash cow, now shrinking and low-margin.
- Digital Asset Hosted Mining — hosting third-party miners. FY2025 revenue $24.4M (gross margin 32%). Small and declining.
Consolidated FY2025 revenue: $319.0M (down from $510.7M FY2024 and $502.4M FY2023). The revenue decline is the headline that misleads: it reflects bitcoin self-mining shrinking faster than colocation has yet scaled. The Q1 2026 print (Lens 5) shows the crossover beginning.
Business model in plain terms. Today it is a hybrid: a declining bitcoin-mining cash engine funding a capital-intensive build-out of AI-datacenter shells that it leases to CoreWeave on take-or-pay-style contracts ("the customer is obligated to pay for leased customer power capacity regardless of utilization, providing operators with revenue visibility" ). The stated end-state, per CEO Adam Sullivan: every megawatt converted to colocation within ~3 years. So the destination is a pure-play AI-infrastructure REIT-like landlord — without the REIT tax structure.
Key payment terms. Colocation contracts are long-dated (CoreWeave 12-year terms) with monthly fees on billable capacity plus pass-through power and usage charges. Take-or-pay economics on contracted capacity = high revenue visibility if the counterparty performs. That conditional is the whole risk (Lens 13).
Lens 2 · Supply Chain
Map: upstream power & hardware → Core Scientific shells → CoreWeave (AI hyperscaler) → CoreWeave's own customers (Microsoft, OpenAI, etc.). Named stakeholders:
Upstream — electric utilities (the binding input). Each facility is tied to a named utility:
- Denton Municipal Electric (Denton, TX) — 394 MW gross (the largest; the primary CoreWeave HDC site).
- Dalton Utilities (Dalton, GA) — 195 MW gross.
- Oklahoma Gas & Electric (Muskogee, OK) — 100 MW gross (the 1.5 GW expansion campus).
- Austin Energy (Austin, TX) — 20 MW (the original 16 MW CoreWeave pilot site).
- Plus sites in Kentucky, North Carolina (Marble), and additional Texas (Pecos).
Upstream — equipment. Mining ASICs (legacy, in runoff). For the HDC build, the binding inputs are power transformers, switchgear, cooling (liquid/direct-to-chip for GPU density), and the GPUs themselves — but note CoreWeave, not Core Scientific, supplies the GPUs. Core Scientific delivers the shell + power + cooling ("critical IT load"). The filing flags extended equipment lead times (transformers, switchgear) as a bottleneck alongside power.
The company (midstream). Converts gross utility power → "billable customer power capacity" (the gross-to-net conversion is the core operating skill and the source of execution risk). Q1 2026: 225 MW billable, up from 120 MW.
Downstream — the single chokepoint. CoreWeave = 100% of Colocation revenue. CoreWeave in turn is heavily concentrated: ~71% of its revenue from Microsoft in Q2 2025, ~86% from four clients by Q3 2025. So Core Scientific's colocation revenue is, two hops down, effectively a leveraged bet on Microsoft's and OpenAI's AI capex.
Chokepoints / single-source dependencies: (1) power interconnect availability at each utility — the #1 industry bottleneck; (2) the single tenant, CoreWeave; (3) transformer/switchgear lead times. Names present — this lens passes.
Lens 3 · Competitive Advantages (moats)
What's actually defensible:
- Secured power + interconnects at scale. The genuine moat. "The ability to secure large amounts of power capacity is often viewed as a primary bottleneck". Core Scientific controls ~1.4 GW gross today and a 4.5 GW total pipeline — including planned 1.5 GW expansions at Muskogee and Pecos. In a power-constrained AI build cycle, pre-secured megawatts behind the meter are the scarce asset. This is real and durable on a 3–5 year view.
- Speed-to-energize. Management claims it energized more MW in 2025 than TeraWulf, Cipher, Galaxy, Applied Digital, Hut 8, and IREN. Converting brownfield mining sites (which already have power + land + permits) into AI colocation is faster than greenfield — a real time-to-market edge over hyperscaler new-builds.
- Switching costs (contractual). 12-year take-or-pay leases lock CoreWeave in. But this cuts both ways — concentration is the lock-in.
What's weak:
- No technology/IP moat. Core Scientific does not own the GPUs, the chips, or the AI software. It is a landlord with electrical and construction expertise. Replicable by any well-capitalized data-center developer (Vantage, QTS, Switch, the hyperscalers themselves).
- Bargaining power is inverted. With one customer at 100% of the segment, CoreWeave needs Core Scientific less than Core Scientific needs CoreWeave — until the capacity is delivered and locked. The Q1 2026 renegotiation that raised Core Scientific's target cash gross margin to 80–85% (from 75–80%) suggests the company has held its ground so far — but that is a relationship, not a moat.
Verdict on moat: the moat is megawatts behind the meter, not the company. Medium-strength, real but commoditizing, and entirely undermined on the demand side by single-customer concentration. Ground: +.
Lens 4 · Segments
Revenue, gross profit, and margin by segment — every figure `` (FY2025 / FY2024 / FY2023, $000s):
| Segment | FY2025 rev | FY2024 rev | FY2023 rev | FY2025 GM | FY2024 GM | Trend |
|---|
| Colocation | 65,424 | 24,571 | — | 30% | 11% | Accelerating hard (new, scaling, margin expanding) |
| Digital Asset Self-Mining | 229,207 | 408,740 | 390,333 | 5% | 23% | Decelerating + margin-crushed (post-halving + difficulty) |
| Digital Asset Hosted Mining | 24,388 | 77,554 | 112,067 | 32% | 31% | Declining (wound down as sites convert) |
| Consolidated | 319,019 | 510,672 | 502,400 | 11.9% | 23.7% | Total down, but mix shifting to colocation |
The story the table tells: Self-mining revenue collapsed from $408.7M (FY24) to $229.2M (FY25) and its gross margin fell from 23% to 5% — the April 2024 halving plus rising network difficulty gutted mining economics. That is exactly why the BTC business is being put in runoff. Meanwhile colocation grew 2.7x (24.6M → 65.4M) and its margin nearly tripled (11% → 30%) as capacity ramped. The consolidated decline masks a favorable mix shift toward a higher-margin, contracted revenue stream.
Geography: all U.S. (TX, GA, OK, KY, NC). No meaningful geographic diversification — concentrated in a few utility service territories (Lens 2). Filing does not break revenue by state; the concentration is in capacity, dominated by Denton, TX.
Q1 2026 segment crossover — the inflection in real time:
- Colocation revenue $77.5M (vs $8.6M Q1 2025 — ~9x).
- Self-mining revenue $30.1M (vs $67.2M Q1 2025 — ~halved).
- Consolidated $115.2M (vs $79.5M — +45% YoY).
Colocation became the largest segment in a single year. This is the most important number in the dossier.
Phase B — Measure performance
Lens 5 · Earnings Result (Q1 2026, filed 2026-05-06)
Top line — strong. Total revenue $115.2M, +45% YoY (from $79.5M), driven by colocation revenue surging to $77.5M from $8.6M. Consensus beat per analyst commentary.
Bottom line — large GAAP loss, but read it carefully. Net loss $(347.2M) vs net income of +$576.3M in Q1 2025. The swing is almost entirely non-cash and counterintuitive:
- $266.5M non-cash impairment charge — write-down tied to legacy mining assets / site conversions.
- ~$30.8M non-cash loss on warrants + contingent value rights. The warrant liability (Tranche 2 Warrants) is marked to market: change in fair value of warrants swung ~$127M against the company because the stock went up. A rising share price creates a GAAP "loss" here — a classic SPAC/restructuring-warrant artifact, not an operating problem.
- The prior-year gain was itself a non-cash bitcoin/warrant fair-value mark in the other direction. Comparing GAAP net income across these quarters is nearly meaningless — the signal is in revenue, segment mix, and adjusted EBITDA.
The real operating signal — Adjusted EBITDA turned positive: +$4.4M vs −$6.1M a year ago. Underlying operations are crossing breakeven as colocation ramps. Still thin, still pre-profitability ("analysts do not anticipate profitability this year" ).
Margin / guidance moves — the genuinely bullish item: management raised the target cash gross profit on the CoreWeave contract to 80–85% (from 75–80%) as billing ramped and cost visibility improved. Higher-margin guidance on the anchor contract is the best news in the print.
Balance-sheet flags:
- Cash & equivalents ballooned to ~$1.005B at Q1 2026 (from $311.4M at YE2025) — funded by a $1.0B Term Loan Facility (drawn via the Core Scientific Finance subsidiary; see Lens 10). Liquidity is not the near-term worry; the maturity is (March 2027).
- Operational delivery: 243 MW fully billing to CoreWeave, with >450 MW expected by end of summer 2026 and the full 590 MW on track for early 2027.
Market reaction: the stock has rerated up through 2026 (52-wk range $10.93→$30.46; trading ~$27–29 in June 2026 ) — the market is paying for the AI-landlord transition and looking through the GAAP noise. Notably, shares fell ~7% on the original $8.7B CoreWeave deal expansion announcement — the market briefly worried about concentration/capex before resuming the uptrend.
Lens 6 · Earnings Calls (sentiment trend)
No transcripts on the shelf (transcripts/ empty) — this lens is ``.
Management focus (Q1 2026 call): the pivot narrative is now total. CEO Adam Sullivan frames bitcoin mining as "essentially in runoff," kept running only to satisfy minimum power commitments while sites convert. The recurring themes: (1) speed of energization vs peers; (2) the 4.5 GW pipeline and new campuses (Pecos, Muskogee, Polaris); (3) monetizing "substantially all" bitcoin in 2026 to self-fund the build; (4) raised CoreWeave contract margins.
Tone shift over time (inferred from the narrative arc):
- Through mid-2025: defensive/strategic — fielding (and ultimately recommending against, then losing the vote on) the CoreWeave buyout.
- Post-Oct 2025 (deal rejected): vindicated and offensive — "we're worth more independent," followed by the 590 MW / $10.2B expansion that proved the point.
- 2026: execution-and-scale — the language is now a data-center developer's (gigawatts, energization cadence, gross-to-net conversion), not a miner's. The words they stopped saying: hashrate, exahash, mining fleet efficiency — all receding. The word they say constantly now: megawatts.
This is a genuine, observable identity change, and the sentiment trend is consistently more confident through 2026. ``.
Lens 7 · Comps
Peer set: the bitcoin-miners-turning-AI-landlords cohort (the only true comps for the transition) — IREN, TeraWulf (WULF), Cipher Mining (CIFR), Applied Digital (APLD), Hut 8 (HUT), Galaxy (GLXY). Pure neocloud (CoreWeave itself) and traditional data-center REITs (Equinix, Digital Realty) are reference points but different animals.
| Company | Ticker | Mkt cap | NTM EV/EBITDA | P/S | Note |
|---|
| Core Scientific | CORZ | ~$8.1–8.7B | ~23.6x | ~19.9x | Anchor: CoreWeave 12-yr / $10.2B |
| TeraWulf | WULF | n/a | ~53.3x | n/a | Higher EV/EBITDA multiple |
| Hut 8 | HUT | n/a | ~52.5x | n/a | Higher EV/EBITDA multiple |
| Cipher Mining | CIFR | n/a | n/a | n/a | Same AI-conversion playbook |
| Applied Digital | APLD | n/a | n/a | n/a | Same AI-conversion playbook |
| IREN | IREN | n/a | n/a | n/a | Building its own AI cloud |
| Peer-group avg (Simply Wall St) | — | — | — | ~4.3x P/S | vs CORZ ~19.9x |
The provenance-critical conflict — surfaced, not resolved: the two sourced multiples point opposite directions.
- On EV/EBITDA, CORZ (~23.6x) trades at a steep discount to WULF (~53x) and HUT (~53x). Bullish framing: cheaper than peers on forward cash earnings.
- On P/S, CORZ (~19.9x) trades at a huge premium to the peer average (~4.3x) and the US-software fair ratio (~5.8x). Bearish framing: priced like software on a landlord's revenue.
Both can be true: CORZ has a near-term EBITDA inflection (raising the denominator on EV/EBITDA fast) while its current revenue is still small relative to its market cap (inflating P/S). The honest read: the stock is priced for the contracted future, not the trailing present. I will not fabricate the missing peer market caps or P/E figures — most of this cohort is loss-making, so P/E and ROE are not meaningfully sourceable. Dividend yield: zero across the cohort. n/a is used wherever a live multiple wasn't in the search returns.
Lens 8 · Stock-Price Catalysts (events that moved CORZ >5%, ~2024–2026)
Mostly ``; the company only re-listed in early 2024 post-bankruptcy, so the window is ~2.5 years.
- Jan 2024 — Emergence from Chapter 11 / Nasdaq relisting. The reset to zero.
- Feb 2024 / Jun 2024 — First CoreWeave contracts (16 MW, then 200 MW). Re-rating from "miner" to "AI-infra" began.
- Feb 26, 2025 — $1.2B CoreWeave data-center expansion → stock surged.
- Jul 7, 2025 — CoreWeave agrees to acquire CORZ (~$9B all-stock, 0.1235 CRWV/share). The stock initially plunged on the all-stock structure (it pinned CORZ to CRWV's volatile equity at a perceived-low ratio).
- Oct 30, 2025 — Stockholders REJECT the merger (Two Seas Capital, Gullane, ISS, Glass Lewis all said it undervalued the company). The deal died. CORZ traded on its standalone merits thereafter.
- Feb 2026 — CoreWeave relationship expanded to ~590 MW / ~$10.2B / 12-yr; $8.7B incremental deal. Briefly down ~7% on concentration/capex fears, then resumed higher.
- May 2026 — Polaris DS acquisition (440 MW) + 4.5 GW pipeline + Muskogee/Pecos 1.5 GW each + $500M/$1B term loan. Scale-up catalysts.
- 2026 YTD — rerate to ~$27–30, near all-time high.
Pattern — what the market actually reacts to: (1) CoreWeave contract news (up) and CoreWeave structure news (the all-stock buyout, down); (2) capacity/MW milestones; (3) capital raises (mixed — funds the build but adds leverage). The market does not react much to bitcoin price anymore — confirming the identity shift. The single most important price driver going forward is CoreWeave's health and the energization cadence, not BTC.
Phase C — Judge people & books
Lens 9 · Management
CEO — Adam Sullivan (CEO since 2023; led the company through bankruptcy emergence and the pivot).
- Track record: Steered Core Scientific out of Chapter 11 (emerged Jan 2024) and architected the conversion from the largest public BTC self-miner into an AI-colocation landlord — landing the anchor CoreWeave relationship and growing it to $10.2B / 590 MW. Energized more MW in 2025 than the named peer cohort. That is a genuine, quantified operational turnaround.
- Skin in the game / tenure: Multi-year tenure through the hardest period. Insider ownership not sourced (
insider-transactions.csv absent) — n/a; the 10-K notes routine Rule 10b5-1 plan adoptions by directors/officers but quantum of ownership wasn't extracted.
- Capital allocation: The defining call — recommending the company NOT take the CoreWeave all-stock deal (and the board/holders rejecting it) looks vindicated: the stock has roughly doubled off the lows and the commercial relationship deepened on better terms. Using a $1.0B term loan + bitcoin liquidation to self-fund the AI build rather than dilute equity at the lows is defensible. Counter: the balance sheet is now deeply levered with negative book equity (Lens 10), and the term loan matures in <12 months — aggressive, refinancing-dependent capital allocation.
- Red flags (governance): Standard SPAC/restructuring-warrant overhang (Tranche 1/2 Warrants) creating GAAP noise and dilution. $21.6M in merger advisory/legal fees burned on the failed CoreWeave deal. No related-party or promotional-behavior flags surfaced.
- Archetype: Professional turnaround operator / dealmaker, not founder. Appropriate for a post-bankruptcy asset-conversion story — the job is financing + construction + contract execution, which suits an operator over a visionary founder. The risk this archetype carries: a dealmaker may keep doing deals (Polaris, 4.5 GW pipeline) and out-run the balance sheet.
Lens 10 · Forensic Red Flags
Forensic stance. The accounting is noisy by nature (bitcoin fair-value marks under ASU 2023-08, warrant liabilities, impairments, fresh-start-adjacent restructuring). Key risks:
- Negative stockholders' equity — the headline forensic fact. Total Stockholders' Deficit of $(962.7M) at YE2025; Total Liabilities $3.31B vs Total Assets $2.35B. Book equity is deeply negative. Common for a heavily-leveraged post-restructuring company with warrant liabilities, but it means the equity value is entirely a forward-option on the AI build — there is no asset-backing cushion under the stock.
- Leverage & refinancing wall. Notes/loans outstanding:
- 2027 Term Loan — $1.0B (drawn 2026; the source of the $1B cash balance) — matures March 3, 2027. A 364-day-style facility. This is the single biggest near-term risk: ~$1B due inside 12 months.
- 2029 Convertible Notes — $460M (3.0% coupon; conversion ~$11.00/share — deep in the money at ~$28).
- 2031 Convertible Notes — $625M (0.00% coupon).
- Total notes/loans ≈ $2.085B. The two convertibles are in-the-money (BTC-era $11 strike vs ~$28 stock), so they likely convert to equity (dilution) rather than cash repayment — that helps the cash picture but dilutes. The $1B term loan is the one that needs refinancing or repayment.
- Impairment cadence. $266.5M non-cash impairment in Q1 2026 and a massive FY2024 net loss of $(1.44B). Recurring large impairments as mining assets are written down through the conversion — expected, but it tells you reported book values are soft.
- Cash flow vs earnings divergence — favorable direction. The GAAP losses are dominated by non-cash items (impairments, warrant marks). Adjusted EBITDA is positive (+$4.4M Q1 2026). The risk is the opposite of the usual forensic worry: don't be scared by the GAAP loss; do watch whether contracted colocation converts to real operating cash flow as capacity energizes.
- Single-customer revenue recognition. With CoreWeave = 100% of colocation, any dispute, renegotiation, or CoreWeave distress flows straight to reported colocation revenue. Take-or-pay terms mitigate utilization risk but not counterparty-credit risk.
- NOL carryforwards of $159.4M (FY2025) — a real but contingent asset (valuation allowance applied), only useful once profitable.
Regulatory findings (required sub-section):
- SEC enforcement: Zero. No SEC Litigation Releases and no AAERs naming Core Scientific in 2021-06-24→2026-06-24, verified via EDGAR EFTS.
- 10-K Item 3 (Legal Proceedings): The material legal matter disclosed relates to pre-bankruptcy / merger litigation — claims tied to the July 2021 SPAC merger/de-SPAC and bankruptcy-era class treatment motions; one such matter was settled during the period, and a bankruptcy motion-to-dismiss was administratively closed (March 2025). Routine post-restructuring litigation tail; nothing that reads as a going-concern or fraud risk.
- Non-SEC enforcement (FTC/DOJ/etc.): Web search returned no material FTC/DOJ/CFPB enforcement actions, consent decrees, or fines against Core Scientific. Crypto-mining environmental/noise complaints exist at the industry level but no company-specific material penalty surfaced.
- Conclusion: No material regulatory or fraud findings — verified via SEC EDGAR EFTS (LR, AAER), 10-K Item 3, and web search as of 2026-06-24. The forensic concern here is leverage and a single customer, not accounting integrity.
Phase D — Project & stress-test
Lens 11 · Forward Projection
The company is at a GAAP loss and not expected to be profitable in FY2026. A clean EPS projection is low-confidence; the more honest forward frame is revenue + adjusted EBITDA driven by billable-MW energization, then EPS-implied. All `` with arithmetic shown; inputs labeled.
Driver — billable colocation MW (the master variable):
- Q1 2026: 243 MW billing.
- End-summer 2026: >450 MW.
- Early 2027: 590 MW (full CoreWeave).
- Beyond: 4.5 GW gross pipeline (Polaris 440 MW + Muskogee/Pecos 1.5 GW each) — but this requires new contracts beyond CoreWeave and years of build.
Revenue path:
- FY2025 actual: $319.0M consolidated; colocation $65.4M.
- FY2026 base: colocation scales from ~243→~450 avg billable MW over the year; self-mining shrinks toward runoff. Estimate consolidated FY2026 ≈ $550–700M. Colocation likely >2x FY2025.
- FY2027 base: full 590 MW CoreWeave billing all year at raised 80–85% cash gross margin. Estimate colocation revenue ≈ $700M–1.0B. Consolidated FY2027 ≈ $800M–1.1B.
- FY2028 base: depends entirely on signing new tenants for the 4.5 GW pipeline. If CoreWeave-only, revenue plateaus near full-590-MW run-rate (~$0.9–1.0B). If new contracts land, materially higher. The pipeline beyond 590 MW is an option, not a base case.
EPS: With ~318M+ diluted shares (FY2025) growing via convertible dilution toward ~360–410M, and the first GAAP-profitable year likely FY2027 or later once depreciation on the built-out capacity is absorbed by 80–85%-margin contracted revenue:
- FY2026 EPS: negative (GAAP loss; impairments + ramp).
- FY2027 EPS: roughly breakeven to modestly positive.
- FY2028 EPS: first clearly positive year if energization stays on schedule and no new equity overhang — call it a wide $0.20–$0.80 range, entirely conditional on the build.
Bull / Base / Bear EPS framing (FY2027, illustrative):
- Bull: new pipeline tenants signed, 590 MW + incremental billing, refinancing clean → revenue ~$1.1B, GAAP positive, stock rerates on a "multi-GW AI landlord" multiple.
- Base: 590 MW CoreWeave delivered on time, pipeline still being contracted → ~$0.9B revenue, near-breakeven GAAP, EBITDA strongly positive.
- Bear: CoreWeave stumbles / capex slips / $1B term loan refinanced at punitive rates or via dilutive equity → revenue disappoints 20–30%, GAAP loss persists, multiple compresses.
Forecast logged (Brier-tracked): per --watchlist rules, the forecast.ts create step is skipped in the unattended sweep (only log when genuinely committed to a base case). Recommend Connor log on promotion: "CORZ FY2027 Adjusted EBITDA positive and >$300M" — p≈0.65, resolves 2028-03-31.
Lens 12 · Bull vs Bear
Bull case. Core Scientific owns the single scarcest input in the AI build-out — secured power behind the meter at gigawatt scale — and is converting brownfield mining sites to AI colocation faster than anyone in its cohort. It has a 12-year, $10.2B anchor contract with the most aggressive AI-cloud builder, at a freshly raised 80–85% cash gross margin. The BTC drag is being deliberately removed (runoff + asset sales self-fund the pivot). Adjusted EBITDA just turned positive; the 590 MW ramp is on schedule for early 2027; and there's a 4.5 GW pipeline of optionality on top. The shareholders who rejected CoreWeave's $9B were proven right within months — the standalone equity is worth more, and the market agrees (near all-time high, Strong Buy, 15/0 buy/sell). On EV/EBITDA it's cheaper than WULF/HUT.
Bear case (2–3 permanent-impairment risks):
- Single-customer, two-hop concentration. 100% of colocation is CoreWeave, which is itself ~71–86% concentrated in Microsoft + 3 clients. If Microsoft pulls back AI capex (it has already passed on a CoreWeave expansion in favor of Nebius ) or CoreWeave hits financial stress (it is burning $30–35B/yr in capex and is a recurring "AI bubble" poster child ), Core Scientific's revenue engine seizes. Take-or-pay helps against utilization but not against counterparty insolvency.
- Negative book equity + a $1B refinancing wall in March 2027. With $(963)M stockholders' deficit and ~$2.1B of debt, the equity is a pure forward option. The term loan matures inside 12 months; refinancing it depends on capital-markets access and continued AI optimism. A risk-off AI repricing closes that window at the worst time.
- Valuation priced for perfection. ~20x P/S vs ~4x peers. The stock already capitalizes the contracted future. Any slip in the energization cadence, a CoreWeave renegotiation, or an AI-capex air-pocket de-rates a richly-priced, loss-making, levered name fast.
Pre-mortem (18 months out, thesis broke): Most likely cause — the AI-capex cycle cooled and CoreWeave slowed take-up of new capacity, so the 4.5 GW pipeline became stranded option value, the $1B term loan refinanced at a high rate or via dilutive equity, and a 20x-sales stock re-rated toward its asset value (which, with negative book equity, is grim). Second cause — a CoreWeave-specific stumble (its Microsoft concentration unwinds) cascades down the chain.
Are multiples too high? On P/S, plainly yes for the current business; on EV/EBITDA, defensible if the EBITDA ramp lands. The market is making a forward bet, not a value call.
Contrarian view (what the market refuses to see): The bears fixate on "bitcoin miner" and "concentration"; the subtler truth the market under-weights is that the failed CoreWeave buyout was the best thing that happened to CORZ shareholders — it kept 100% of the AI-landlord upside in CORZ stock instead of swapping it into CoreWeave's more-concentrated, more-levered equity at a low ratio. CORZ is now the cleaner way to own the CoreWeave/AI-power thesis than CoreWeave itself — owning the power and the buildings, not the GPU-depreciation treadmill.
Lens 13 · Devil's Advocate (short-seller)
Skeptical short stance.
- Where revenue is concentrated: 100% of the growth segment (colocation) is one customer. That is not "concentration risk," that is being a captive vendor to a single counterparty that has explicitly said it would rather acquire you than depend on you — and whose own revenue is ~71% one client (Microsoft) that is building competing capacity and steering deals to rivals (Nebius). Two hops up the chain, CORZ is a leveraged Microsoft-AI-capex derivative with extra steps.
- Why the moat is weaker than bulls think: "Secured power" is replicable by any developer with capital and patience, and the hyperscalers are the deepest-pocketed power-buyers on earth — they can disintermediate the landlord. Core Scientific owns no chips, no GPUs, no AI IP. It is a construction-and-electrical contractor with a 12-year lease, dressed as a tech growth stock at 20x sales.
- Most dangerous competitor bulls underestimate: the customer itself. CoreWeave (and Microsoft/OpenAI) can build or buy their own shells; the whole point of CoreWeave's failed acquisition was to internalize this capacity. The next CoreWeave expansion could just as easily be a CoreWeave self-build.
- Worst capital-allocation / balance-sheet facts: $(963)M negative equity; a $1B term loan due March 2027; recurring nine- and ten-figure impairments; $21.6M torched on a deal the shareholders then voted down. The convertibles are in-the-money → guaranteed dilution.
- Assumptions that must hold for today's price (~$28, ~$8B cap): (1) CoreWeave stays healthy and keeps paying for 12 years; (2) the 590 MW energizes on schedule; (3) the $1B term loan refinances cheaply; (4) the AI-capex super-cycle does not air-pocket; (5) the 4.5 GW pipeline finds new tenants. That's five things, several outside the company's control.
- If growth disappoints 20–30%: at 20x sales a revenue miss is brutal — a re-rate to even 8–10x sales on a lower revenue base could halve the equity, and with negative book value there is no floor.
- Single scenario that permanently impairs the business: CoreWeave financial distress or a sharp AI-capex contraction during 2026–2027 while the $1B term loan is up for refinancing. Plausibility: not the base case (AI capex is still surging in mid-2026), but distinctly non-trivial given CoreWeave's own leverage and concentration — call it a real tail, not a remote one.
Lens 14 · Management Questions (ordered by information value)
- The $1.0B term loan matures March 2027 — what is the refinancing plan, and what happens to it if AI-infrastructure credit markets tighten before then?
- CoreWeave is 100% of colocation revenue. What concrete steps (and timeline) are you taking to diversify the tenant base across the 4.5 GW pipeline, and do you have signed non-CoreWeave LOIs today?
- What contractual protections do you have if CoreWeave defaults, restructures, or is itself acquired — termination rights, security, parent guarantees, standby letters of credit?
- CoreWeave's revenue is ~71% Microsoft, and Microsoft has steered an expansion to Nebius. How do you underwrite CoreWeave's 12-year payment ability against its customer concentration?
- You raised the CoreWeave cash gross margin target to 80–85%. What drove that, and is it durable across the full 590 MW or front-loaded on the easiest sites?
- Walk through the gross-to-net (1.4 GW gross → 920 MW leasable → billable) conversion economics and the capex per MW to energize — what is the all-in cost to build a billable megawatt?
- With negative book equity of ~$(963)M and ~$2.1B of debt, what is your target capital structure once the build matures, and at what point do you expect positive GAAP equity?
- The two convertibles ($460M / $625M) are in-the-money near $11 — what dilution should shareholders model, and will you cash-settle or share-settle?
- Bitcoin is "in runoff." What is the timeline and expected proceeds from monetizing the remaining BTC, and what is the cash-flow bridge if BTC falls 40% before you sell?
- The 4.5 GW pipeline (Muskogee/Pecos 1.5 GW each, Polaris 440 MW) — how much is contracted vs aspirational, and what's the capital required to build it?
- How do you compete for the next AI tenant against hyperscalers self-building and against better-capitalized data-center developers (Vantage, QTS, Switch)?
- What is your power-cost structure across utilities, and how exposed are you to grid-tariff changes or curtailment as utilities impose phased load-ramping requirements?
- After the rejected CoreWeave merger, what is the board's stated M&A posture — would you entertain a revised bid, and at what premium?
- What are the equipment lead-time risks (transformers, switchgear, cooling) for hitting >450 MW by end-summer and 590 MW by early 2027, and where is the schedule most fragile?
- What does steady-state, fully-built, single-tenant-diversified Core Scientific look like in 2029 — revenue, EBITDA margin, leverage, and free cash flow?