Core Scientific (CORZ) Q2 FY2026: Bitcoin Miner to AI Data Center Pivot Explained
The Hook
Core Scientific $CORZ reported a net loss of $1,155.3 million for the three months ended June 30, 2026, on revenue of $164.2 million. The loss is seven times the revenue. In the same quarter, gross margin went from 6.4% to 42.7% and gross profit went from $5.0 million to $70.0 million. Both of those things are true, and reconciling them is the whole story.
The bridge is one line. Operating loss for the quarter was $78.5 million. Below the operating line sits a single non-cash item, "change in fair value of warrants and contingent value rights," of $1,045.5 million. That one line is 90.5% of the net loss. It is not an operating result, not an impairment, and not a cash cost. It is a mark-to-market on warrants issued under the company's 2024 plan of reorganization, remeasured every quarter against the share price. The filing states the cause plainly: the change "was driven by changes in our stock price during the respective periods." Because those warrants are a liability, a rising share price produces a larger accounting loss.
Company Snapshot
Core Scientific operates large-scale data centers. Its historical business was mining bitcoin for its own account. Its current business is high-density colocation, which is leasing power and data center space to AI and high-performance computing operators. This analysis covers the Form 10-Q for the quarter ended June 30, 2026, filed July 28, 2026. Every figure below is labelled with its period, because a Q2 10-Q carries three-month, six-month and prior-year facts side by side.
At June 30, 2026 the company held 2,115 MW of gross utility power capacity, of which 1,275 MW is leasable to customers, 590 MW is leased, and 395 MW was actually billing. That last number, billable customer power capacity, is the revenue driver, and it went 120 MW to 225 MW to 395 MW across the last three quarter-ends.
The Financial Story
The operating-to-net bridge. Here is the full reconciliation for the three months ended June 30, in thousands of dollars, exactly as the filing presents it.
| Line item | Q2 2025 | Q2 2026 |
|---|---|---|
| Reportable segment gross profit | 5,025 | 70,044 |
| Loss (gain) on fair value of digital assets | (29,797) | 9,368 |
| Loss on disposal of property, plant and equipment | 4,166 | 1,273 |
| Loss on remeasurement of assets held for sale | 0 | 19,495 |
| Loss on contract termination | 0 | 41,948 |
| Colocation organizational and site startup costs | 11,655 | 27,039 |
| Selling, general and administrative | 45,285 | 49,389 |
| Operating loss | (26,284) | (78,468) |
| Loss on debt extinguishment | 1,377 | 5,435 |
| Interest expense (income), net | (1,185) | 23,833 |
| Change in fair value of warrants and CVRs | 909,958 | 1,045,515 |
| Other non-operating expense, net | 207 | 152 |
| Loss before income taxes | (936,641) | (1,153,403) |
| Income tax expense | 158 | 1,907 |
| Net loss | (936,799) | (1,155,310) |
Note the prior-year column. Q2 2025 carried a $910.0 million warrant charge of its own. This is not a one-quarter anomaly, it is the recurring behaviour of the capital structure. Run the series and it inverts: in Q1 2025 the warrants moved the other way, a fair value decrease of $634.3 million, and Core Scientific reported net income of positive $580.7 million. Nothing about the underlying business changed by $1.5 billion between those two quarters. The share price did.
The company's own earnings-per-share disclosure makes the point better than any commentary. Because the penny-strike warrants sit inside the basic share count, the filing adds part of the warrant charge back to the numerator. Q2 2025's $936.8 million net loss became a loss available to common of $13.3 million, or four cents a share. This quarter, with a smaller add-back of $74.2 million, the same treatment gives a loss per share of $3.32. The GAAP headline and the per-share line are telling two different stories, and neither is the business.
Where the margin actually came from. The gross margin move from 6.4% to 42.7% is real, and it is not the mining business improving. It is one segment displacing another. Three-month figures, in thousands:
| Segment | Q2 2025 revenue | Q2 2026 revenue | Q2 2025 gross profit | Q2 2026 gross profit | Q2 2025 margin | Q2 2026 margin |
|---|---|---|---|---|---|---|
| Colocation | 10,560 | 136,669 | 1,130 | 79,983 | 11% | 59% |
| Digital asset self-mining | 62,424 | 21,535 | 2,835 | (12,165) | 5% | (56)% |
| Digital asset hosted mining | 5,644 | 5,997 | 1,060 | 2,226 | 19% | 37% |
| Total | 78,628 | 164,201 | 5,025 | 70,044 | 6.4% | 42.7% |
Colocation revenue grew 12.9 times and colocation gross profit grew 70.8 times. Colocation went from 13% of revenue to 83%. Meanwhile self-mining revenue fell 65% and its gross profit went from positive $2.8 million to negative $12.2 million. Mining is now a gross-margin-negative business inside this company, and management is winding it down: hosted mining is expected to conclude by December 31, 2026. Colocation contributed $80.0 million of the $70.0 million of consolidated gross profit, which is to say more than all of it.
The colocation economics are cleaner than the headline margin suggests. Of the $136.7 million of colocation revenue, $35.1 million is power fees, and the filing shows cost of power fees at exactly $35.1 million, because power is passed through to the customer without markup. Strip both sides out and the margin-bearing revenue is $101.6 million of license and maintenance fees against $21.6 million of non-power cost, a 78.7% margin. That also means reported colocation revenue moves with power prices in ways that say nothing about the underlying business, a point the filing makes itself.
The cash and the balance sheet do not match the income statement either. For the six months ended June 30, 2026, against a net loss of $1,502.5 million, net cash provided by operating activities was positive $230.9 million. Adjusted EBITDA, the company's own non-GAAP measure, was $41.1 million for the quarter against $28.5 million a year earlier. One caveat worth stating: $208.2 million of that operating cash was proceeds from selling mined bitcoin, and the digital asset balance fell from $222.0 million at December 31, 2025 to $49.7 million at June 30, 2026. The treasury is largely gone.
The balance sheet is where the warrant math lands. Total stockholders' equity at June 30, 2026 is negative $2,419.4 million. That is not driven by operating losses alone. Warrant liabilities on the balance sheet total $1,975.3 million, up from $936.1 million at December 31, 2025. Against 96.4 million Tranche 1 warrants excluded from diluted EPS as antidilutive, that liability works out to roughly $20.50 per warrant, which is approximately where the stock traded at quarter end. The liability is the share price. It settles in shares, not cash, which is why negative book equity here does not carry the meaning it would at an ordinary company, and why the filing contains no going-concern language.
The financing. In March 2026 the company drew a $1.0 billion 364-day term loan. On May 6, 2026 an indirect subsidiary issued $3.30 billion of 7.75% Senior Secured Notes due 2031 at 99.25, repaid the term loan in full (a $5.4 million extinguishment loss), and funded a debt service reserve of which $344.8 million remained restricted at quarter end. Principal debt is now $4.385 billion: $3.30 billion secured, $460 million of 3.00% converts due 2029, $625 million of 0.00% converts due 2031. Cash and equivalents stand at $1,769.7 million. The 2029 converts have been in the money since Q4 2025 and remain convertible into Q3 2026; no holder has converted yet. Capital spending was $954.2 million in the half plus $232.5 million for land in Hunt County, Texas, and the company is contractually committed to roughly $1.0 billion more, of which about $264 million passes through to the customer.
Valuation
Where it trades. At $18.12 per share, market capitalization is $5.82 billion and enterprise value is $8.46 billion, against a price-to-sales ratio of 13.2 and EV/sales of 19.2 (stockanalysis.com, July 29, 2026; the same source shows a consensus of Strong Buy across 17 analysts and an average target of $35.86, and notes the stock fell 12.65% on the July 28 earnings release). Enterprise value computed from the filing agrees: $5.82 billion of equity plus $4.385 billion of principal debt less $1.770 billion of unrestricted cash is $8.44 billion. There is no meaningful P/E to quote, because there is no meaningful E.
A capacity-based scenario range. The honest way to value this is per megawatt, because that is how the revenue arrives. Q2 2026 colocation gross profit was $79.98 million on 395 MW of billable capacity, which is $0.81 million of annualized gross profit per MW. Holding those per-MW economics flat, holding SG&A at the current $197.6 million annualized run rate, treating the $1.085 billion of converts as converted into the share count, and using net debt of $1.53 billion ($3.30 billion secured notes less $1.77 billion cash) on roughly 509 million fully diluted shares:
| Scenario | Billing capacity | Annualized colocation gross profit | EBITDA proxy | EV/EBITDA | Implied EV | Implied equity | Per diluted share |
|---|---|---|---|---|---|---|---|
| Bear: CoreWeave contract only | 590 MW | $478M | $280M | 10x | $2.8B | $1.3B | ~$2.50 |
| Base: plus AMD and Neocloud leases | 1,119 MW | $906M | $709M | 13x | $9.2B | $7.7B | ~$15 |
| Bull: all leasable capacity leased | 1,275 MW | $1,033M | $835M | 16x | $13.4B | $11.8B | ~$23 |
These are implied values under stated assumptions, not price targets, and the arithmetic is ours, not the company's. Two things fall out of it. First, the CoreWeave contract on its own, fully commissioned at 590 MW, does not obviously support the capital structure: $280 million of EBITDA proxy against $270 million of annual coupon interest on $4.385 billion of debt is roughly a wash. Second, at $18.12 the market is sitting between the base and bull cases, which means the price already assumes the AMD and Neocloud capacity gets built and commissioned on schedule and that a good share of the 685 MW of unleased capacity finds a tenant.
The largest weakness in this arithmetic is depreciation. Total D&A was only $15.5 million in the quarter against $1.77 billion of net property, plant and equipment, because most of the $954 million of half-year capital spending is not yet in service. As that capacity commissions, GAAP depreciation rises sharply. An EBITDA-based frame flatters a business this capital-intensive, and returns on invested capital, not EBITDA multiples, are the eventual test.
Risks
Customer concentration is the risk, and it is absolute. The MD&A states it without hedging: "Our colocation revenue is currently derived entirely from a single customer, CoreWeave, a provider of HPC services." That is 100% of the segment that is 83% of revenue and more than 100% of gross profit. The company added a new risk factor this quarter specifically on the point, warning that failure to win additional customers leaves "our revenue dependent on one counterparty, exposing us to heightened risk if that customer reduces, delays, or terminates its commitments." The July 27, 2026 leases with AMD for 377 MW and with a Neocloud for 152 MW are the first real diversification, and they are 15-year terms, but no revenue from them appears in this quarter.
Dilution is large and stacking. At June 30, 2026 the company excluded 182.7 million potentially dilutive shares from diluted EPS as antidilutive: 96.4 million Tranche 1 warrants, 69.6 million from the convertible notes, 16.4 million of RSUs, PSUs and MSUs, and 0.3 million of options. Against 319.6 million shares outstanding, that is 57% potential dilution before counting the new warrant issued to AMD for up to 30 million shares at $23.47, of which approximately 6.5 million had already vested at 12,222 shares per MW of leased IT load. More leases mean more vested warrant shares.
Execution and leverage. The gap between 590 MW leased and 395 MW billing is the near-term growth, and it depends on equipment lead times, labour, permitting and interconnection. The $3.30 billion of secured notes begin amortizing at 11.5% of original principal per year once rent commences and revenue credits abate, and the company provided an uncapped completion guarantee for specified data center projects. Meanwhile the mining wind-down is not free: Q1 2026 carried a $266.5 million non-cash impairment of mining assets after hashprice hit historic lows, this quarter carried a $19.5 million remeasurement of mining equipment moved to held-for-sale, and the company paid a $41.9 million loss to terminate its mining-equipment supply contract with Block, Inc. and Proto Global LLC. Getting out of bitcoin mining is costing real money.
The Bottom Line
Core Scientific's income statement is currently a poor instrument for measuring Core Scientific. The $1.16 billion loss is a share-price derivative wearing an accounting costume, and it will reverse into a headline gain the first quarter the stock falls. What the filing does show is a business that in twelve months replaced a low-margin commodity operation with a 59%-margin contracted one, went cash-flow positive from operations, and financed a $4.4 billion balance sheet against it.
The three things to watch are all capacity and counterparty, not earnings. Billable capacity against the 590 MW leased, since that gap is the contracted growth already paid for. The first revenue from the AMD and Neocloud leases, since that is the answer to the concentration risk. And whether any of the 685 MW of unleased capacity gets leased, since today's price requires it. Ignore the net loss line until the warrants are gone.
Every figure in this analysis was pulled from Core Scientific's Form 10-Q for the quarter ended June 30, 2026, filed July 28, 2026, except where a price or consensus figure is attributed to stockanalysis.com. This is not investment advice and contains no price targets.
Built on the RoboSystems SEC Shared Repository: structured XBRL filing data for every public company that files. See robosystems.ai/pricing. New customers get 50% off your first month with code ROBO50.