TeraWulf's $19 billion Anthropic lease shows how AI compute is now financed

The lease that dwarfs the quarter

TeraWulf's third quarter, as the company reported it in its filing, came to $76.2 million in revenue. Lease revenue from high-performance computing was $63.0 million of that, or about 83%. Those are modest numbers for a company in the middle of the largest infrastructure buildout in recent memory. The number that matters sits next to them: a 20-year lease with Anthropic for about 401 megawatts of critical IT capacity, which the company says represents about $19 billion of contracted revenue over the initial term. That is roughly 250 times the revenue TeraWulf booked in the quarter. m:…3218

The filing fills in how the lease is being financed and built. TeraWulf says it secured a Google credit backstop for project financing in support of the Anthropic lease, and it ended the quarter with about $2.4 billion in cash, cash equivalents and restricted cash. It sold its 50.1% interest in the Abernathy joint venture for $250 million in cash during the quarter, signed an engineering, procurement and construction agreement with Fluor Enterprises for three data center buildings at its Justified campus, and increased contracted power capacity at its Muskie campus from 500 megawatts to 1,000 megawatts. It also reaffirmed its target of contracting 250 to 500 megawatts of incremental critical IT capacity every year. m:…3218

Read together, the pieces describe a landlord model rather than a mining or hosting business. The company is selling decades of capacity under contract, financing the build against that contract, and shedding a joint venture interest to fund it. The quarterly revenue is almost beside the point. What an investor in this kind of company is really underwriting is whether the lease holds, the power arrives, and the buildings get finished on the schedule the contract assumes. m:…3218

The power behind the lease

The same pattern shows up on the utility side of the meter. Evergy says it signed a letter of intent with Digital Realty for a 1.4 gigawatt project, and signed three amendments to electric service agreements that it expects will drive about 250 megawatts of incremental load. Its third quarter, as the company reported it, produced GAAP earnings of $612.9 million, or $2.59 per share, with adjusted earnings per share of $2.58 against $2.03 a year earlier. It raised its 2026 adjusted earnings guidance to a range of $4.35 to $4.50 per share, and reaffirmed a long-term adjusted earnings growth target of 6% to 8% or more through 2030, expecting growth above 8% from 2028. m:…9030

Evergy says its third-quarter results benefited from warmer weather, recovery of regulated investments, growth in weather-normalized demand and higher large-customer revenues, partly offset by higher operations and maintenance expense and higher depreciation. The detail that matters for the compute story is the demand itself: a regulated utility is now planning its growth around large customers signing for gigawatts, not around households adding air conditioners. m:…9030

BKV Corporation, an energy company, is building the power side of the same trade. The company says its BKV-BPP Power unit entered a cost reimbursement agreement with a leading frontier AI lab to increase modular power generation at the Temple Energy Complex to 400 megawatts. The agreement covers up to approximately $179 million if offtake arrangements are not agreed to by November 18, 2026, and includes an aggregate reimbursement cap of $374.1 million if the intended offtaker elects to take assignment of the equipment supply contract. BKV reported net income attributable to BKV of $29.4 million for the quarter and adjusted EBITDAX of $132.7 million, and completed a $575.0 million convertible senior notes offering at an annual rate of 1.625%. A lab paying to reserve power before it has even agreed to buy the output is a measure of how far ahead the power queue runs. m:…5774

What the credit market makes of it

Morgan Stanley's credit view of the buildout is calm on the surface and exacting underneath. Its podcast discussion notes that the AI buildout involves the deployment of trillions of dollars in capital, and that credit spreads are currently tight, which suggests AI-related financing is not yet driving up the cost of capital. Borrowing costs for issuers have still increased, the discussion notes, because of higher underlying rates and wider spreads, and issuer earnings have shown double-digit growth quarter after quarter. Willis Towers Watson's research on data center growth in Asia makes the same point from the other side of the world: power availability and grid connectivity are critical to successful data center development, and power is one of the most significant constraints facing it. m:…1205 m:…5990

The warning in the same discussion is about paperwork, not demand. Rigorous documentation is essential in AI infrastructure financing to prevent uncontrolled cash hemorrhaging, it argues, while management teams can adjust the pace of capacity building without systemic consequence. That is the lens to apply to a $19 billion lease: the value is only as solid as the documents behind it — the lease terms, the backstop, the power contracts and the construction agreements. A 20-year promise is a financing instrument before it is a revenue line, and the market is, for now, willing to fund it cheaply. m:…1205