TeraWulf held two bitcoin at the end of June. Fair value $133,000, against total assets of $8.05bn. A company that mined 1,496 coins last year keeps 0.0017 per cent of its balance sheet in the thing it was built to produce.
The quarter
In the three months to 30 June 2026, revenue was $44.8m: $31.9m of rent from leasing data centre capacity, $12.8m of bitcoin. A year earlier the split was nil and $47.6m. Leasing is 71 per cent of the top line, from zero in twelve months, and it is on the income statement, not in guidance.
The reported net loss was $940.8m, which sounds catastrophic and mostly is not. $755.7m of it is a mark-to-market charge on warrants issued to Google, a liability that grows when TeraWulf's own share price rises; $83.9m is stock compensation. The operating loss was $140.5m and adjusted EBITDA negative $18.3m — the latter defined unusually strictly, stripping out $29.0m of interest income, without which it would be roughly positive $10.6m.
Mining is finished here
The mining segment lost $78,000 in the quarter, before depreciation, overhead or interest. A year earlier it made $22.0m.
Its own disclosure is blunter. Cash cost per bitcoin was $44,547; add miner depreciation, as TeraWulf does in a second line, and it becomes $97,486 against a realised $71,704 a coin. That is roughly $25,800 destroyed on each of the 179 coins produced, against a comfortable margin a year earlier, and only 28,300 of some 53,700 miners owned were still running. This is deliberate liquidation rather than failure, since every megawatt moved to a ten-year lease is margin — but the bitcoin business is no longer the investment case.
| Revenue (rent $31.9m) | $44.8m |
| Net loss (warrant mark $755.7m) | $940.8m |
| Cash / debt principal | $2.62bn / $5.73bn |
| Bitcoin held | 2 |
| Cost to mine vs value | $97,486 vs $71,704 |
| Shares outstanding, year on year | +27.3% |
| Capacity commenced | 102 MW |
| Lease revenue in the accounts | $1.46bn |
What is contracted and what is announced
The 10-Q reports 438 megawatts contracted at Lake Mariner — 60 to Core42, 378 to Fluidstack — of which 81 was earning revenue at the end of June, rising to 102 by the date the accounts were issued. All four Core42 leases have commenced; of the three to Fluidstack, one was partially delivered. Google backstops Fluidstack's obligations, which is why that paper is financeable, but only lease by lease at commencement.
Then the largest item, another 401 megawatts. In July the company disclosed a twenty-year lease to Anthropic expected to generate “approximately $19 billion of contracted revenue over the initial lease term”. That sits in a filed exhibit, so it is not puffery. But the lease has not commenced, the document itself was not filed, it came under Item 8.01 rather than as a material definitive agreement, and the $19bn appears nowhere in the financial statements. The larger $33bn, assuming two extensions, comes from a furnished deck three weeks later.
Announced, $19bn. In the accounts, $1.46bn.
What the accounts do contain is a maturity schedule for commenced leases: $1.46bn. Everything beyond sits outside it, and the options behind the $33bn are ones the company's own policy calls not reasonably certain to be exercised. One datapoint the release omits: Core42's right to take more capacity expired unused.
The capital structure
Debt principal is $5.73bn against quarterly revenue of $44.8m, and interest paid in the first half was $131.1m against half-year revenue of $78.8m. Cash is large at $2.62bn, and $3.2bn of the debt is ring-fenced in a subsidiary secured on the data centre assets at 7.75 per cent, which is a sensible shape for infrastructure. The converts are where the dilution lives: three tranches totalling $2.53bn at cash coupons of 2.75, 1.00 and zero per cent. The first two, $1.5bn between them, are already classified current because holders' triggers have been met, so they are either a near-term cash call or about 139m shares. Capped calls hedge those two; the $1.03bn 2032 tranche, some 51m shares, has none. Shares outstanding rose 27.3 per cent over the year on the two quarterly cover pages, and the overhang is roughly 305m shares against 499m outstanding.
The verdict
The bull case is partly proven: leasing made $9.0m of segment profit on 81 of 438 contracted megawatts, the buildout is funded, and the counterparties are investment-grade or backstopped by one. If the rest arrives on the terms signed, today's loss is a construction artefact.
The bear case is equally plain. The business earning money today is 102 megawatts old, the asset carrying the valuation starts delivering in late 2027, the $19bn behind it has no counterpart in the accounts, and $1.5bn of convertible debt sits at the holders' option. Anyone buying this underwrites delivery, not mining.
Retrieved 28 September 2026 from primary filings on SEC EDGAR: TeraWulf Inc, CIK 0001083301, fiscal year ending 31 December. Quarterly figures are the three months to 30 June 2026 from the 10-Q filed 5 August 2026, comparatives the same quarter of 2025; prior-year coins from the 10-K of 27 February 2026. The Anthropic lease and the $19bn come from the 8-K and Exhibit 99.1 of 6 July 2026, both filed. Contested: the $33bn extension figure, the $600m of Google credit support and the 839 MW platform total appear only in furnished exhibits, never in the 10-Q, which reports 438 MW contracted at Lake Mariner; the share increase is measured on the two cover pages, other bases giving 25.5 to 27.9 per cent; and the 10-Q misstates its own six-month loss as the three-month figure. Figures in US dollars, as reported. Not a recommendation to buy, sell or hold anything.