CORZ — Q2 2026 Financial Report Analysis

Revenue doubled as AI data-center rent replaced bitcoin mining — but a $1.0 billion warrant charge swamped the income statement

Core Scientific used to make most of its money mining bitcoin. In the second quarter of 2026 (three months to 30 June), 83% of its revenue came instead from renting out power-dense data-center space to CoreWeave, a company that runs AI computing clusters. Total revenue rose 109% to $164.2 million, and gross profit — what is left of revenue after the direct cost of running the sites — jumped from $5.0 million to $70.0 million. The reported net loss of $1.16 billion is almost entirely a non-cash accounting charge: the company's own share price rose, which increased the value of warrants it owes to former creditors, and that increase is booked as a loss.

At a glance

• Colocation (data-center rental) revenue: $136.7 million, up from $10.6 million. Capacity CoreWeave is actually paying for rose to 395 megawatts (MW) from 225 MW at the end of March and 120 MW at the end of 2025, so this line is still ramping.
• Net loss: $1,155.3 million, of which $1,045.5 million is the change in warrant and related liability values. Strip that out and the loss is about $110 million — still a loss, mainly from a $41.9 million contract-termination charge, $27.0 million of site start-up costs and the first quarter of interest on new debt.
• Debt went from about $1.1 billion to $4.4 billion (face value) in six months. A $3.3 billion bond at 7.75% interest, issued in May, funds the build-out; its interest alone comes to roughly $256 million a year on our arithmetic, a large claim against a business whose colocation gross profit was $80.0 million this quarter.

The numbers

"Operating margin" is the share of revenue left after all operating costs, before interest and tax; here it is negative. "n/m" means not meaningful: the per-share loss calculation treats two kinds of warrants differently (explained below), so the two EPS figures are not comparable. Adjusted EBITDA is earnings before interest, tax, depreciation and a long list of items the company excludes, including the warrant charge, stock-based pay, impairments and start-up costs.

How the business changed: from mining bitcoin to renting power to AI

Core Scientific owns 11 data centers across seven US states with about 2.1 gigawatts (GW) of utility power under contract. In 2024 it signed CoreWeave as a colocation customer — meaning Core Scientific supplies the building, power, cooling and security, and CoreWeave puts its own AI servers inside. That contract has grown to about 590 MW of leased capacity across five sites. The company bills only once a block of capacity is built and handed over, so the key operating figure is billable MW:

Colocation revenue went from $77.5 million in Q1 (implied from the six-month total) to $136.7 million in Q2, up 76% in one quarter, in step with billable capacity. The 10-Q attributes the year-on-year increase "primarily" to "incremental billable customer power capacity." The remaining 195 MW of the CoreWeave contract is "in various stages of construction and commissioning," which is the most visible source of further growth.

One detail flatters the revenue line: power is passed through to CoreWeave at cost and booked as both revenue and cost. That means higher electricity prices raise colocation revenue without raising profit, and part of the revenue growth is power being resold rather than margin being earned. Colocation gross profit — $80.0 million, a 59% margin — is the cleaner figure to track.

Bitcoin mining is being wound down. Self-mining revenue fell 65.5% to $21.5 million because the company mined 53% fewer bitcoin (power was moved to colocation) and the average bitcoin price was 27% lower than a year earlier. The segment now loses money at the gross level: a gross margin of −56%, against +5% a year ago. Hosted mining for outside customers is expected to end by 31 December 2026.

What the headline numbers hide

• The $1.16 billion loss is mostly the warrants, not the business. Warrants issued to creditors when the company emerged from bankruptcy in January 2024 are carried as a liability at fair value. When the share price rises, that liability grows and the increase is booked as a loss. The change in fair value of warrants and contingent value rights was $1,045.5 million this quarter and $910.0 million in Q2 2025. It is non-cash, and management attributes it to "changes in our stock price." Excluding it, the Q2 loss is about $110 million.
• Why EPS jumped from −$0.04 to −$3.32. The company has two warrant classes. For the $0.01 "penny" Tranche 2 warrants, the shares are already counted as outstanding and their fair-value change is removed from the EPS numerator. A year ago most of the warrant loss ($923.5 million) came from Tranche 2, so it barely touched EPS. This quarter $972.3 million (on our arithmetic: $1,046.6 million total warrant change less $74.2 million for Tranche 2) came from the $6.81-strike Tranche 1 warrants, which do count against EPS. The EPS swing is an accounting-mechanics effect, not a twelve-fold decline in the business.
• A dilution decision is coming within months. 96.4 million Tranche 1 warrants remain, exercisable at $6.81 and expiring on 23 January 2027; their fair value of $1.81 billion is now a current liability. If all were exercised for cash at the $6.81 strike, the company would receive about $656 million and issue about 96 million shares — roughly 30% more than the 319.6 million outstanding at 30 June. Separately, the two convertible note issues could convert into about 69.6 million shares. In July the company also issued AMD a warrant for up to 30 million shares at $23.47, with about 6.5 million vested so far.
• Operating cash flow mostly came from selling bitcoin. Six-month operating cash flow was $230.9 million, against a net loss of $1.5 billion. Of that, $208.2 million was the proceeds of selling 2,385 bitcoin, which the company now classifies as operating cash flow. Without those sales, operating cash flow was about $23 million, and that figure is helped by $98.4 million of customer prepayments booked as deferred revenue. Holdings fell from 2,537 bitcoin ($222.0 million) at year-end to 848 bitcoin ($49.7 million).
• One-offs in this quarter. A $41.9 million loss from terminating a mining-equipment purchase contract with Block, Inc. and Proto Global; a $19.5 million write-down on mining and electrical equipment moved to "held for sale"; and $5.4 million lost on repaying a $1.0 billion term loan that was drawn in March and repaid in May. The first quarter also carried a $266.5 million impairment on mining equipment and infrastructure, reflecting what the filing calls "hashprice reaching historic lows" — a low point in the revenue a miner earns per unit of computing power.
• The interest bill has started. Net interest swung from $1.2 million of income a year ago to $23.8 million of expense, from the $3.3 billion senior secured notes (issued in May) and the short-lived term loan. Because the bonds were issued partway through the quarter, Q2 does not yet show a full quarter of interest. The notes also require principal repayments of 11.5% of the original amount a year (about $380 million) once rent at the financed sites is fully running; the filing's maturity table assumes that starts on 15 November 2029.
• Adjusted EBITDA excludes a lot. The $41.1 million figure adds back the warrant charge, $13.9 million of stock-based pay, $15.5 million of depreciation, $27.0 million of colocation start-up costs and all of the one-offs above. Start-up costs have risen from $11.7 million a year ago and, by management's own account, will recur as new sites and customers are added.
• Controls are still flagged as not effective. The company reports an unremediated material weakness — a significant gap in its financial-reporting controls — around accounting for planned demolition of buildings converted from mining to AI hosting. It led to restatements of previously issued 2025 statements, and management does not expect to finish fixing it before its year-end 2026 assessment.
• One customer pays for nearly everything. CoreWeave accounted for about 77% of six-month revenue and all colocation revenue. Some CoreWeave contracts were moved this year to a CoreWeave-owned special-purpose vehicle, with CoreWeave remaining a primary obligor. The filing notes that recourse in a default "would depend on its financial condition at the time."

Takeaway: The operating turnaround is real: colocation produced $80.0 million of gross profit at a 59% margin, up from $1.1 million a year ago, and billable capacity more than tripled since December. But the company now carries $4.4 billion of debt at face value, relies on one customer for most of its revenue, and faces roughly 96 million potential new shares from warrants expiring in January. How fast leased megawatts become billing megawatts matters more than the headline loss.

Outlook

The company gives no revenue or earnings guidance in the 10-Q. What it does say:

• Near-term growth is already contracted. 195 MW of the 590 MW CoreWeave contract is not yet billing. Management expects colocation revenue "to increase as additional contracted capacity is commissioned and delivered." Billable capacity rose from 225 MW to 395 MW during Q2, so not all of it billed for the whole quarter. Q3 should therefore grow from Q2 even if no new capacity is added.
• A second anchor customer arrived after quarter-end. On 27 July the company signed 15-year leases with AMD for 377 MW at Pecos (Texas), Muskogee (Oklahoma) and Hunt County (Texas), plus 152 MW with an unnamed "Neocloud" at Auburn (Alabama) and Dalton (Georgia), which AMD supports through credit-support agreements. That is about 530 MW in total. AMD also holds a right to lease up to a further 1,925 MW through December 2028. On our arithmetic, adding the July leases to the 590 MW already leased would cover most of the 1,275 MW of leasable capacity. This is the first real step away from depending on CoreWeave, but those sites still have to be built.
• The build-out needs a lot of cash. At 30 June the company had $1.77 billion of unrestricted cash and $781.7 million of restricted cash (a debt-service reserve, a $120 million escrow for the pending $421 million Polaris DS acquisition next to its Muskogee site, and construction escrows). It is committed to about $1.0 billion of spending over the next 12 months, $264 million of which is passed through to CoreWeave, and it says it "may from time to time seek additional financing." Hunt County and Muskogee are not funded by the customer the way the CoreWeave conversions were, so tariff and construction-cost overruns land on Core Scientific.

Our read: the quarter supports the strategy. Colocation margins are high and capacity is coming online quickly. The risk has moved from bitcoin prices to execution and the balance sheet. Three things to watch in Q3: billable MW against the 590 MW contracted (the clearest driver of revenue), the first full quarter of bond interest against colocation gross profit, and how the Tranche 1 warrants are settled before January, which will decide how much of the coming growth goes to existing shareholders.