Core Scientific
Power-monetization / colocation landlord: repurpose energized, interconnect-queued MW originally secured for Bitcoin mining into long-dated (12-year) contracted AI/HPC hosting leases. The economic engine is the spread between contracted, credit-backstopped colo revenue (~$1.4M/MW/yr at ~243MW billing => ~$350M annualized) and the combined power, build-out and debt-service cost. A capital-intensive real-asset landlord whose value is the site's power, not its ASICs; still carries a shrinking, hash-price-linked self-mining tail as a cash-flow hedge being retired.
The thesis on this name
State of the AI Cloud
The largest, earliest miner-turned-AI-colo landlord — ~590MW/>$10B of energized power leased to a SINGLE anchor tenant (neocloud CoreWeave) on 12-year contracts, and independent again after the CoreWeave buyout was rejected.
Earnings, margins, COGS & capex
Core Scientific is mid-transformation from a Bitcoin miner into an AI/HPC colocation landlord. Q1 FY26 revenue was $115.2M with colocation ($77.5M, +~800% YoY) now well above self-mining ($30.1M, -55% YoY), and adjusted EBITDA turned marginally positive (+$4.4M). But GAAP is deeply loss-making (-$347.2M net, driven by a $266.5M non-cash impairment), capex ($389.2M in Q1) vastly outruns revenue while sites energize, and the company restated several 2024-2025 periods after PP&E accounting errors and disclosed a material weakness in internal control — a governance flag. The whole model rests on one anchor tenant, CoreWeave, under 12-year contracts scaled to ~590MW / ~$10.2B; ~243MW is billing today (~$350M annualized colo run-rate) with the remainder delivering through 2026-2027.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~57¢ is cost of goods and ~43¢ operating expense, and the remainder is an operating loss.
Revenue trend
Margins
improving - first positive print (vs -$6.1M a year ago) as colo scales
negative - dominated by a $266.5M non-cash impairment plus build-out depreciation/interest
rising - colo overtook self-mining as the primary revenue line
declining/retiring - hash-price-linked, being wound down
COGS structure
COGS is dominated by data-center operating cost: power/electricity (the core input), site operations and staffing, plus — on the legacy book — mining fleet energy and hosting cost. As the mix shifts from volatile, hash-price-linked self-mining toward contracted 12-year colocation, revenue quality and margin stability improve. The larger economic costs sit below the line as depreciation on the build-out and interest on the ~$2B note stack; the $266.5M Q1 impairment reflects write-downs on legacy/mining-related assets as the pivot completes [fact/estimate].
Capex
Q1 FY26 capex $389.2M — of which ~$129.9M was funded by CoreWeave under the hosting service agreements. Capital funds energizing and retrofitting sites (e.g. Denton TX +70MW to >260MW critical IT load) to deliver the ~590MW CoreWeave contract, plus new power (Hunt County TX ~430MW acquired; total pipeline expanded to ~4.5GW). Funded by a $3.3B 7.75% senior-secured-notes offering due 2031 plus the CoreWeave capex contribution [fact].
Latest earnings
Mixed-but-improving operationally; the GAAP headline is misleading. Revenue more than doubled YoY (+109%) and consolidated gross margin went 6% -> 43%; adjusted EBITDA rose to $41.1M from $28.5M a year ago and $8.9M in Q1 2026. The $(1,155.3)M net loss is NOT an operating deterioration -- the company states it was 'primarily driven by the change in fair value of warrants, reflecting appreciation in the Company's stock price during the period' ($1,045.5M non-cash). The $266.5M property, plant and equipment impairment belongs to Q1 2026 (three months ended 31 Mar 2026); the Q2 2026 impairment line is $0. Q2's own non-recurring drags were a $41.9M loss on contract termination and a $19.5M loss on remeasurement of assets held for sale.
The company issued NO revenue, EBITDA or earnings guidance in the Q2 2026 release -- confirmed by reading the full release, which contains no guidance section. It instead frames the ramp through operating metrics: billing for 437 MW as of mid-July representing approximately $635 million in average annualized colocation GAAP revenue, against ~1.1 GW of leased customer power capacity and >$24 billion of potential contracted revenue. AMD deployments are stated to begin in 2027. Hosted mining operations are expected to conclude by 31 Dec 2026 and self-mining continues to wind down.
- Q2 2026 total revenue
- $164.2M (+109% YoY vs $78.6M; +43% QoQ vs $115.2M)
- Colocation revenue
- $136.7M, ~83% of total revenue (Q2 2025: $10.6M)
- Gross profit / margin
- $70.0M; 43% consolidated, 59% colocation, (56)% self-mining, 37% hosted mining
- Adjusted EBITDA
- +$41.1M (Q1 2026 $8.9M; Q2 2025 $28.5M)
- Net loss
- $(1,155.3)M / $(3.32) per share -- $1,045.5M is a NON-CASH warrant fair-value charge from the share price rising; Q2 impairment was $0
- AMD agreement (signed 27 Jul 2026)
- AMD Leases 377 MW + Neocloud Leases 152 MW with AMD credit support = ~530 MW across five sites; 15-year terms with three five-year options; >$14B potential base contracted revenue; AMD reservation right on a further 1,925 MW through 28 Dec 2028 (the path to 'up to 2.5 GW')
- AMD warrant
- Up to 30M shares at $23.47 (5-day VWAP pre-signing), vesting 12,222 shares per MW; ~6.5M vested on the executed leases; terminates 27 Jul 2031. ~9.3% potential dilution on 321.3M shares
- Leased vs billing capacity
- ~1.1 GW leased / >$24B potential contracted revenue; 590 MW CoreWeave with 395 MW billing at 30 Jun 2026 (225 MW at 31 Mar); 437 MW billing by mid-July = ~$635M annualized colocation GAAP revenue
- Power footprint
- ~2.1 GW gross utility power capacity and ~1.3 GW total leasable customer power capacity across 11 data centers in 7 states, at 30 Jun 2026
- Customer concentration
- One Colocation customer (CoreWeave) = 83% of Q2 2026 and 77% of 1H 2026 total revenue; colocation revenue is derived entirely from CoreWeave
- Debt / liquidity
- $4.385B principal ($3.30B 7.75% SSN 2031 + $460M 3.00% converts 2029 + $625M 0.00% converts 2031); liquidity $1,819.4M; restricted cash $781.7M; stockholders' deficit $(2,419.4)M
- Internal control
- Material weakness in internal control over financial reporting NOT remediated as of 30 Jun 2026; disclosure controls concluded 'not effective'; remediation targeted at the 31 Dec 2026 annual assessment
Growth drivers
- CoreWeave 12-year HPC hosting — scaled from ~200MW (Jun 2024) via option exercises (+112MW, +120MW, +70MW Denton) to ~590MW across ~6 sites, ~$10.2B contracted over the 12-year terms [fact]
- ~243MW currently billing (~$350M annualized colo GAAP run-rate), with the remaining contracted MW delivering through 2026-2027 as sites energize [fact]
- ~4.5GW total power pipeline (incl. ~430MW Hunt County TX land/power acquired in Q1 FY26) creates expansion runway beyond the current CoreWeave book [fact]
- Mix shift to contracted colo (+~800% YoY) away from volatile self-mining (-55% YoY) improves revenue quality and durability [fact]
- Staying independent (CoreWeave buyout rejected Oct 2025) lets CORZ holders capture the datacenter re-rating directly rather than converting into CRWV at a fixed ratio [fact]
- $3.3B 7.75% senior-secured notes (due 2031) fund the build-out without a near-term equity raise [fact]
Reported financials — SEC EDGAR
Audited GAAP figures pulled from SEC filings · latest filing 2026-03-02. The audited primary-source spine — not financial advice.
Revenue — annual (GAAP)
Margins & balance sheet — FY’25
Bull & bear
The single-tenant bear case was directly answered this quarter. AMD signed ~530 MW across five sites on 15-year terms (>$14B potential base contracted revenue) with a reservation right on a further 1,925 MW, taking leased capacity to ~1.1 GW and >$24B of potential contracted revenue -- a second anchor at near-parity with CoreWeave's 590 MW. Underneath it, the colocation engine is now visibly profitable: 59% colocation gross margin, 43% consolidated, adjusted EBITDA $41.1M, and billing capacity up 225 MW -> 437 MW in roughly a quarter.
- A second anchor tenant of comparable scale, from a very different credit than a neocloud: AMD Leases for 377 MW directly plus 152 MW to a Neocloud under AMD credit-support agreements that let AMD cure defaults -- ~530 MW on 15-year terms with three five-year options
- Margin quality is now demonstrated, not asserted: colocation gross margin 59% and consolidated gross margin 43% in Q2 2026, against 11% and 6% respectively a year earlier
- Energisation is executing: billable capacity 225 MW (31 Mar) -> 395 MW (30 Jun) -> 437 MW (mid-July), the last representing ~$635M of average annualized colocation GAAP revenue vs $164.2M of total quarterly revenue today
- Adjusted EBITDA inflected to +$41.1M from +$8.9M in Q1 2026, and the loss-making self-mining tail (gross margin (56)%) is being retired -- hosted mining ends by 31 Dec 2026
- Fully funded through the near-term build: $1,819.4M of liquidity plus $781.7M restricted (incl. a $344.8M debt-service reserve) against ~$1.0B of committed spend, $264M of which passes through to the customer
- Optionality is real and dated, not vague: AMD's 1,925 MW reservation right runs to 28 Dec 2028, and the Polaris DS acquisition (~$421M cash, up to 440 MW gross utility, expected Q3 2026 close) plus Hunt County (~430 MW) extend the developable footprint
- The $1.16B net loss is an accounting artifact of success -- $1,045.5M is a non-cash warrant remeasurement caused by the share price appreciating
The de-risking is contractual and future-dated; the risk is on the balance sheet today. AMD contributes no revenue until 2027, CoreWeave still supplies 83% of revenue, and the funding gap between them is a $4.385B debt stack against a $(2.42)B stockholders' deficit with an uncapped completion guarantee. The governance flag did not clear: the material weakness is still unremediated and disclosure controls are 'not effective'. And the ~1.1 GW now leased nearly exhausts the ~1.3 GW currently leasable -- further growth requires new capex, not just new signatures.
- Material weakness in internal control over financial reporting remains UNREMEDIATED as of 30 Jun 2026 and disclosure controls were concluded 'not effective'; remediation is only targeted at the 31 Dec 2026 annual assessment, so reporting reliability is an open risk through year-end
- Leverage stepped up sharply: $4.385B principal debt vs a $(2,419.4)M stockholders' deficit, with the $3.30B senior secured notes amortising at an initial 11.50%/yr of original principal once rent commences and revenue credits abate -- and an UNCAPPED completion guarantee on the specified development projects
- AMD delivers nothing yet: deployments begin in 2027, the '2.5 GW' headline is ~530 MW contracted plus a 1,925 MW reservation RIGHT (not a commitment), and the $14B / $24B figures are 'potential' base contracted revenue over 15-year terms, not backlog under GAAP
- Concentration is still extreme in the reported period: one Colocation customer (CoreWeave) = 83% of Q2 2026 revenue and 77% of 1H; colocation revenue derives entirely from CoreWeave, and the 10-Q notes its licenses were assigned to CW SPV -- CoreWeave remains primary obligor, but this adds a counterparty layer whose enforceability depends on CoreWeave's condition at any default
- Capacity headroom is thinner than the pipeline implies: ~1.1 GW is now leased against ~1.3 GW of total leasable customer power capacity at 30 Jun 2026, so incremental tenants depend on Hunt County (~430 MW gross utility) and the unclosed Polaris DS / Muskogee acquisition (~$421M cash plus up to $40M contingent, ~440 MW, expected Q3 2026) coming through
- Cash burn is structural while building: ~-$956M of 1H free cash flow (operating $230.9M less $1,186.7M capex), with operating cash flow itself flattered by customer prepayments recorded as deferred revenue ($654.4M current + non-current at 30 Jun 2026)
- Q2 carried its own real, non-warrant charges: a $41.9M loss on contract termination and a $19.5M loss on remeasurement of assets held for sale, on top of the $266.5M impairment taken in Q1 2026
- Tariff exposure is asymmetric on the new sites: the CoreWeave projects are customer-funded with the company's contribution capped, but Hunt County and Muskogee are not, so equipment and materials cost inflation lands on CORZ's own capital
- AMD warrants dilute up to 30M shares (~9.3%) at $23.47 as capacity is signed -- the second tenant is partly paid for in equity
What it is worth
Sum-of-the-parts / contracted-backlog framing rather than near-term earnings multiple: value the ~590MW / ~$10.2B 12-year CoreWeave colo book (~243MW billing at ~$350M annualized) plus the ~4.5GW power-pipeline option, net of ~$2B+ debt, the $3.3B 7.75% notes, and a discount for single-tenant concentration + the controls/restatement overhang. GAAP earnings are negative and not a usable near-term anchor; adjusted EBITDA is only marginally positive (+$4.4M Q1).
De-rating case: a CoreWeave stress/renegotiation, a refinancing scare, an AI-capex air-pocket, or unresolved controls issues impair the backlog value and compress the equity materially [estimate]
Range-bound near current levels (~$20): the CoreWeave ramp delivers roughly on plan but single-tenant concentration, GAAP losses, leverage and the controls overhang cap the multiple [estimate]
Re-rating case: tenant book diversifies beyond CoreWeave, the ~4.5GW pipeline converts to contracted leases, controls are remediated, and CORZ trades toward a datacenter-landlord multiple on ~$10.2B+ of contracted revenue [estimate]
Equity value is a leveraged call on (a) CoreWeave paying and renewing, (b) the pipeline converting into new contracted leases, and (c) clean remediation of the material weakness. A datacenter-landlord/REIT-style re-rating is the upside case; a CoreWeave stumble or refinancing scare is the downside. Research conviction is LOW (watch), not a core long. [estimate]
SWOT
Strengths
- Largest contracted HPC backlog of the pure miner-pivots — ~590MW / ~$10.2B with CoreWeave over 12-year terms, giving long-dated revenue visibility [fact]
- Sites already energized and delivering (~243MW billing, ~$350M annualized colo) de-risks execution versus greenfield developers [fact]
- ~4.5GW power pipeline (incl. ~430MW Hunt County) is the scarce asset in the AI build-out — energized, interconnect-queued MW others can't quickly replicate [fact]
- Independent path (CoreWeave buyout rejected Oct 2025) lets equity holders capture the DC re-rating directly rather than swapping into CRWV at a fixed 0.1235 ratio [fact]
- Adjusted EBITDA turned positive (+$4.4M Q1 FY26) as the mix shifts to contracted colocation [fact]
Weaknesses
- Extreme single-tenant concentration — CoreWeave is effectively the entire HPC book — CORZ's revenue quality is only as good as CoreWeave's own creditworthiness and demand [fact]
- Deeply GAAP loss-making: -$347.2M net in Q1 FY26 incl. a $266.5M non-cash impairment [fact]
- Governance/controls flag — restated several 2024-2025 periods after PP&E accounting errors and disclosed a material weakness in internal control over financial reporting [fact]
- Capital-intensive — Q1 FY26 capex $389.2M vs $115.2M revenue; the build-out is debt-funded (~$2B notes + a new $3.3B 7.75% issue) [fact]
- Legacy BTC/self-mining exposure remains a hash-price-linked, declining tail (-55% YoY) that adds noise while being retired [fact]
Opportunities
- Convert the ~4.5GW pipeline into additional contracted colo leases with new hyperscaler/neocloud tenants, diversifying away from CoreWeave [estimate]
- Continued AI training/inference demand keeps energized power scarce and colo rates firm through multiple GPU generations [estimate]
- A datacenter-landlord (or REIT-style) re-rating if the tenant book diversifies and contracted revenue proves durable — the peer-group re-rating option [estimate]
- Full retirement of self-mining removes a volatile, low-margin distraction and simplifies the equity story [estimate]
Threats
- AI-capex air-pocket — two hyperscaler capex guide-downs or an AI-capex-digestion regime would reprice the whole miner-pivot cluster at once [estimate]
- CoreWeave counterparty risk — CoreWeave is itself debt-financed and concentration-exposed; if it stumbles or renegotiates, CORZ's backlog is impaired [estimate]
- Power/grid + interconnect delays gate how fast contracted MW convert to billing revenue [fact/estimate]
- Refinancing/leverage risk — the debt-funded build-out relies on continued capital-market access; a risk-off window makes rolling the notes expensive [estimate]
- Rejecting the CoreWeave buyout forfeited a certain exit premium — the market now prices standalone build/financing/execution risk [fact]
Moats, dependencies & bottlenecks
Moats
interconnect-queued power at scale (~4.5GW pipeline; sites already delivering) real and scarce today, but power siting is replicable by well-capitalized rivals over time The genuine edge is the site's energized power, not its ASICs — the scarce asset in the AI build-out [fact]
12-yr with CoreWeave) contractual switching cost, not technological lock-in; single-counterparty renewal is the risk Backlog is only as good as CoreWeave's willingness/ability to pay and renew [fact]
earliest miner-pivot) operational lead, capital-replicable by funded peers (APLD, WULF, CIFR, IREN) Advantage is execution and delivered MW, not a structural barrier [estimate]
a liability, not a moat; debt-funded build vs cash-rich hyperscalers Relies on ~$2B notes + a new $3.3B 7.75% issue; leverage is a vulnerability if AI-capex sentiment turns [fact]
Dependencies
~590MW / ~$10.2B 12-yr book is effectively the entire HPC thesis; CoreWeave's own creditworthiness and demand pipeline are the existential dependency [fact]
Energizing contracted MW is gated by electricity, grid interconnect and site build-out; also the scarce asset that underpins the whole model [fact]
~$2B notes + a new $3.3B 7.75% senior-secured issue due 2031; the build-out cannot outrun capital-market access, and a risk-off window raises refinancing cost [fact]
CORZ hosts the shells, but demand and lease durability track CoreWeave's Nvidia GPU deployment and the broader AI-capex cycle [estimate]
Declining (-55% YoY) and being retired, but still injects hash-price volatility into revenue while it winds down [fact]
Advantages
- Largest and earliest miner-to-HPC pivot with the biggest contracted backlog of the pure pivots (~590MW / ~$10.2B) [fact]
- Sites already energized and delivering (~243MW billing, ~$350M annualized colo) - execution de-risked vs greenfield [fact]
- ~4.5GW power pipeline (incl. ~430MW Hunt County) provides expansion runway [fact]
- Independent post-buyout, so equity holders capture the DC re-rating directly rather than a fixed CRWV swap [fact]
- Adjusted EBITDA turned positive (+$4.4M) as contracted colocation overtook volatile self-mining [fact]
Weaknesses
- Extreme single-tenant concentration in CoreWeave - the counterparty's credit IS the thesis [fact]
- Deeply GAAP loss-making (-$347.2M Q1 FY26) driven by a $266.5M non-cash impairment [fact]
- Restated 2024-2025 results and a disclosed material weakness in internal controls - a governance/reporting flag [fact]
- Capital-intensive, debt-funded build (~$2B notes + a new $3.3B 7.75% issue) with capex ~3.4x revenue [fact]
- Residual legacy-BTC self-mining exposure adds hash-price volatility while it is wound down [fact]
Bottlenecks
- Power and grid interconnect — energizing the remaining contracted MW (toward ~590MW) is gated by electricity, interconnect queues and site retrofits [fact]
- Capital access — the debt-funded build cannot grow faster than it can raise notes; rate/risk-off shocks throttle the ramp [fact/estimate]
- Single-counterparty concentration — converting the ~$10.2B backlog to cash depends entirely on CoreWeave taking and paying for capacity on schedule [fact]
- Internal controls / financial-reporting reliability — the disclosed material weakness and prior-period restatement must be remediated to restore reporting confidence [fact]
- Skilled data-center operations talent to stand up and run high-density GPU colocation at contracted SLAs [estimate]
Top signals & trends
Top signals
A new anchor tenant would de-risk the single-counterparty thesis and support a re-rating; continued CoreWeave-only reliance keeps the bear case intact [estimate]
CORZ is a leveraged proxy on CoreWeave's creditworthiness; any CoreWeave stress transmits directly to CORZ's backlog value [estimate]
On-schedule energization converts backlog to cash; interconnect/grid delays are the bear trigger [fact]
Until controls are remediated and restated periods are clean, reporting-reliability risk overhangs the equity [fact]
CORZ is a high-beta expression of the miner-pivot trade; a hyperscaler capex cut reprices the whole Layer-3 cluster [estimate]
Trends
Converting volatile ~$3-5M/MW/yr hash-price revenue into long-dated contracted AI/HPC leases at higher, more stable margins and richer multiples - CORZ is the anchor of the CoreWeave-counterparty cluster [fact]
Keeps energized power scarce and colo rates firm; the entire bull thesis rests on this persisting [fact]
Scarcity supports the value of CORZ's energized MW, but interconnect/grid delays gate how fast contracted capacity converts to billing revenue [fact]
CORZ (CoreWeave), APLD (CoreWeave), WULF/CIFR (Fluidstack) all rest on one neocloud - a cluster-wide fragility [estimate]
Removes a volatile low-margin book (positive for revenue quality) but also removes a cash-flow hedge and triggers impairments on legacy assets (the $266.5M Q1 charge) [fact]
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Electric utilities / grid operators (ERCOT and others) Power and interconnect are the core input and the scarce asset underpinning the colo model
Supplies (via CoreWeave) the GPUs hosted in CORZ facilities; the AI-capex/GPU cadence drives lease demand
Vertiv) Power, cooling and thermal-management infrastructure for high-density GPU colocation retrofits
Construction / EPC and power-equipment contractors Build out and energize contracted MW at Denton, Hunt County and other TX/US sites
Anchor and effectively sole HPC tenant: ~590MW / ~$10.2B over 12-year terms across ~6 sites; funds part of the build capex (~$129.9M in Q1 FY26)
Declining hash-price-linked self-mining revenue ($30.1M Q1 FY26, -55% YoY), being wound down
The other CoreWeave-counterparty miner-pivot: ~400MW at Ellendale ND + a ~$5B IG-hyperscaler lease at Polaris Forge 2, pursuing a REIT conversion - the closest structural comp [fact]
Cleanest full pivot on the Fluidstack/Google-backstopped side (~360MW Lake Mariner, Google ~14% equity, zero-carbon power); competes for the same AI-colo leases and capital [fact]
TeraWulf's structural twin - entire 300MW Barber Lake TX leased to Google-backstopped Fluidstack; hybrid still-mining model [fact]
Best-anchored miner-pivot (direct Microsoft ~$9.7B + Nvidia ~$3.4B, owned-GPU AI Cloud model); non-US (Australia) domicile, US-listed and US-asset-heavy (Texas) [fact]
Both CORZ's effectively-sole HPC customer AND the neocloud whose credit underpins the whole backlog - the dependency cuts both ways [fact]