Core Scientific’s second-quarter press release says the company has “more than $24 billion of potential contracted revenue”. The lessor table in the financial statements filed the same day supports $9.19bn — $6,087,667,000 of future operating lease payments expected to be received, plus about $3.1bn on leases not yet commenced.
Both went out on 28 July 2026. One was filed. The other was furnished — the distinction that does most of the work in this sector and almost none of the reporting.
Filed, furnished, and in the accounts
An exhibit to an Item 2.02 earnings release is furnished under Regulation FD: not part of the reviewed financial statements, and not carrying the same liability. We checked every headline figure across both. Revenue of $164.2m, gross profit of $70.0m, the net loss, the balance sheet and the adjusted EBITDA reconciliation are identical line for line.
Four numbers are not in the 10-Q at all: the $24bn, a separate “more than $14 billion of potential base contracted revenue”, “approximately 1.1 GW” of leased capacity where the filed metric is 590 MW, and “437 MW of capacity as of mid-July, representing approximately $635 million in average annualized colocation GAAP revenue”. The 1.1 GW reconciles once you add leases signed the day before the release, so it is a post-period figure set against in-quarter accounts. The $14bn appears in no filed document at all.
The leases themselves are not on file. Nor is the thirty-million-share warrant.
That matters more than the arithmetic. The AMD and Neocloud leases, the credit support agreements between them and the AMD warrant agreement are not filed as exhibits. The 10-Q’s exhibit index runs to an indenture, a note form, a merger agreement and the certifications. The operative terms do appear in a filed footnote — 377 MW to Advanced Micro Devices and 152 MW to an unnamed “Neocloud”, fifteen-year terms with three five-year options — but the contracts do not.
The pivot is real
Colocation revenue went from $10.6m to $136.7m year on year and is now 83 per cent of the total, up from 13. Colocation gross margin went from 11 per cent to 59. Billable capacity has compounded 120, 225, 395 MW across three quarters against 590 contracted, and the gap is already built or building. Operating cash flow turned from minus $11.0m to plus $230.9m across the half. None of that is a press release; it is in the accounts.
Self-mining has been retired rather than defended. The company mined 301 bitcoin in the quarter — a figure it no longer discloses, derived here from holdings and sales and confirmed against its own stated 53 per cent decline. It stopped publishing hashrate after 15.7 EH/s at the end of 2025. Marginal cash cost was $59,339 a coin against $71,545 realised, so power is covered; fully loaded it was $111,960, a loss of $40,415 on every coin. The segment lost $12.2m at a minus 56 per cent margin.
What the balance sheet says
Debt went from $1.06bn to $4.30bn in six months. Adjusted EBITDA of $41.1m in the quarter does not cover the $52.4m of interest incurred in the same quarter, and the first coupon on the $3.3bn of 7.75 per cent notes falls due on 15 November. Shareholders’ equity is negative $2.42bn.
Two things deserve more weight than the growth. First, the company restated 2024 and all three quarters of 2025 for improperly capitalising assets committed to demolition — $122.9m of overstated property — and the material weakness behind it was still unremediated at 30 June, in precisely the area the whole thesis runs through. The auditor’s controls opinion is adverse. Second, adjusted EBITDA has been quietly redefined: the prior-year quarter was $21.5m as reported and appears as $28.5m here, a change the 10-Q does not disclose, and $27.0m of the $41.1m is an add-back of recurring start-up costs.
| Total revenue | $164.2m, of which colocation 83% |
| Contracted revenue, filed | $9.19bn on 590 MW |
| Contracted revenue, furnished | “more than $24 billion” |
| Adjusted EBITDA | $41.1m |
| Interest incurred, same quarter | $52.4m |
| Total debt | $4.30bn carrying, $4.39bn principal |
| Stockholders’ equity | negative $2.42bn |
| Potential dilution | 182.7m shares on 321.3m outstanding |
The bull case is that an infrastructure business is visibly being built, that it diversified away from a single customer at the first opportunity, and that AMD holds a free option on a further 1,925 MW through December 2028 with a strike 77 per cent above the last close. The bear case is that the latest reviewed accounts are four months old, no third-quarter filing has arrived, the capital expenditure for 529 MW of self-funded greenfield is not quantified anywhere, and the numbers the market is trading on are furnished.
All figures from Core Scientific’s own SEC filings, fetched from EDGAR on 2 November 2026 with a descriptive user agent: the 10-Q for the quarter ended 30 June 2026, the FY2025 10-K, the restated FY2024 10-K/A, the Q2 2025 10-Q, and 8-Ks covering the restatement, the terminated CoreWeave merger, the AMD and Neocloud leases and the Polaris closing. Fiscal year ends 31 December. Bitcoin mined and the cost-per-coin figures were computed here from segment data because the company publishes neither — the 301 coins were validated against its own stated 53 per cent decline. Dollars throughout, as a US issuer. Unverifiable and not relied on: hashrate at 30 June, the capital expenditure for the AMD and Neocloud sites, the identity of the Neocloud, and the $14bn and $24bn figures, which appear in no filed document. Not a recommendation to buy, sell or hold any security.