TeraWulf Inc. (WULF) Company Overview

A digital-infrastructure company built around power access

TeraWulf Inc. is a Nasdaq-listed U.S. digital-infrastructure developer and operator. Its original economic engine was self-mined bitcoin, but its strategic center has shifted toward high-performance computing and artificial-intelligence data centers. The company now describes itself as an energy-infrastructure platform that secures large power positions, builds high-density data halls, and leases those facilities to AI and cloud customers. The official company overview emphasizes power-secured infrastructure for AI and HPC rather than cryptocurrency alone.

Why the company matters

TeraWulf matters because access to reliable electricity has become a gating factor for large AI deployments. The company’s asset base connects energy-market expertise with data-center construction: Lake Mariner in New York is the operating flagship; Justified Data and Muskie expand the Kentucky pipeline; and additional sites create optionality in other power markets. The 2025 Form 10-K frames power control, long-duration contracting, modular development, and direct infrastructure ownership as the model’s core.

Two revenue streams with very different economics

How lease revenue is structured

HPC contracts combine a right to use data-center capacity with non-lease services. Base rent creates recurring revenue; electricity is generally passed through to tenants; and customer-requested fit-out work can add revenue, often with a markup. This means reported HPC revenue is not identical to economic rent, so researchers should separate base lease income from power pass-throughs and one-time fit-out activity. The latest Q1 2026 Form 10-Q identifies the two reportable segments and their profit measures.

Operating progress was stronger than consolidated earnings

In the quarter ended March 31, 2026, revenue was nearly flat year over year, but its quality changed materially. HPC produced $21.0 million of revenue and $10.2 million of segment profit after all 60 MW of Core42 capacity had commenced. Mining revenue fell as facilities were repurposed and curtailed: TeraWulf mined 168 bitcoin versus 372 in Q1 2025, while the average bitcoin price used in revenue fell to $77,739 from $93,522. The company’s Q1 2026 earnings release therefore reads as an infrastructure-ramp update more than a conventional earnings beat-or-miss story.

Why the GAAP loss needs decomposition

The $427.7 million net loss does not map directly to current operating economics. It included $101.4 million of stock-based compensation, $28.5 million of depreciation, $25.7 million of impairment, $67.1 million of interest expense, and the non-cash warrant remeasurement. Yet adjusted EBITDA was still negative $4.1 million, so the correct conclusion is not that losses are merely accounting noise. The core segments were profitable, but corporate costs and the financing architecture remained heavy.

Strategic turning points

1. 2021
   TeraWulf was incorporated and began trading on Nasdaq, creating a public vehicle for energy-intensive digital infrastructure.
2. 2022
   Lake Mariner began bitcoin mining on a repurposed industrial power site; a long-term New York Power Authority allocation strengthened the campus power position.
3. December 2024
   The Core42 leases established a commercial bridge from mining to GPU-oriented colocation.
4. May 2025
   The acquisition of Beowulf Electricity & Data internalized development and energy expertise, increasing vertical integration.
5. August 2025
   Fluidstack leases, supported by Google credit arrangements, expanded Lake Mariner’s contracted AI capacity and enabled project financing.
6. February–May 2026
   The Justified and Muskie acquisitions added major Kentucky campuses with immediate or planned utility-scale power.
7. July 2026
   A 20-year Anthropic lease validated Justified, while the planned Abernathy stake sale signaled a preference for wholly owned, directly controlled assets.

The May 2025 Beowulf E&D acquisition and August 2025 Fluidstack agreement are especially important. The first brought critical capabilities inside the company; the second proved that power-rich mining campuses could attract large AI tenants and credit-enhanced financing.

The financial statements show the transition

Which sites matter most?

Lake Mariner is the operating proof point. At year-end 2025 it had 245 MW of legacy mining capacity and 18 MW of operating HPC capacity; by March 31, 2026, 60 MW of critical IT load was energized. Its contracted Lake Mariner program comprises 60 MW for Core42 and 378 MW for Fluidstack. Justified Data in Hawesville offers 480 MW of immediate grid-connected power, and the July 2026 Anthropic lease covers approximately 401 MW. Muskie is earlier-stage but larger, with more than 1 GW planned over time.

The May 2026 Muskie acquisition announcement targets the first 500 MW beginning in the second half of 2028 and another 500 MW in the second half of 2030. Those dates illustrate the central analytical tension: power positions can be valuable well before they produce revenue, but schedule slippage can materially defer cash flow.

Contracted customers change revenue visibility

The Anthropic agreement is expected by the company to generate about $19 billion of contracted revenue over its initial 20-year term. That headline improves backlog visibility, but valuation must still discount development cost, financing, delivery risk, ramp timing, operating obligations, and the difference between contracted revenue and free cash flow.

The moat is power development, not proprietary software

TeraWulf does not report a patent-based technology moat. Its advantage is an execution bundle: long-term control of industrial sites, interconnection rights, energy-market knowledge, reusable electrical infrastructure, liquid-cooling design, and the ability to convert those inputs into financeable leases. Brownfield locations can shorten the path to power compared with greenfield projects, while mining provides a flexible load that can be curtailed or displaced as higher-value HPC demand arrives.

Who are the main competitors?

Competition is broad because the scarce input is power rather than a single product category. A practical peer set includes former or current digital-asset miners converting sites to AI use, such as Core Scientific, IREN, Cipher Mining, Hut 8, and Applied Digital. TeraWulf also competes with independent developers, infrastructure funds, hyperscalers building their own campuses, and established data-center landlords such as Digital Realty and Equinix. The 10-K identifies site availability, power economics, capital access, execution capability, and customer relationships as the decisive factors.

Liquidity is substantial, but it is paired with substantial obligations

At March 31, 2026, TeraWulf reported $2.63 billion of unrestricted cash and $462.7 million of restricted cash. Total assets were $7.01 billion, while total liabilities were $7.09 billion and TeraWulf stockholders’ equity was a $78.8 million deficit. The debt stack included $3.06 billion of long-term debt, $1.60 billion of long-term convertible notes, $490.4 million of short-term convertible notes, and additional current borrowings. The company also closed a $250 million revolving facility. Liquidity therefore supports construction, but much of the capital structure is linked to specific projects, restricted accounts, convertibles, warrants, and contractual completion obligations.

Capital allocation is a build-and-finance cycle

FY2025 revenue rose to $168.5 million from $140.1 million, but net loss widened to $661.4 million and adjusted EBITDA was negative $23.1 million. Interest expense reached $80.2 million. Management is recycling capital as well as raising it: the planned sale of TeraWulf’s 50.1% Abernathy interest is intended to monetize roughly $450 million invested and redirect capital toward wholly owned campuses. That can simplify reporting and improve control, but it also makes project-level returns and sale proceeds critical to judging capital-allocation quality.

Founder influence is meaningful without a dual-class structure

TeraWulf has one vote per common share, so economic ownership and voting power are broadly aligned. Even so, insiders have meaningful influence. The 2026 proxy statement calculated ownership using 492.6 million shares outstanding after the April offering. Co-founder, chairman, and CEO Paul Prager beneficially owned 8.5%; co-founder and CTO Nazar Khan held 5.7%; and current directors and executive officers as a group held 15.9%.

Governance signals to interpret

The governance question is therefore not whether a controlling shareholder can override all others; it is whether founder-led execution, equity incentives, and capital-market activity create more project value than dilution. That makes share count and fully diluted securities essential operating metrics, not merely footnotes.

The upside is a multi-campus contracted infrastructure platform

The risks are concentrated in execution, customers, financing, and power

TeraWulf’s principal risks follow directly from its model. Construction delays can defer lease commencement and, in some contracts, jeopardize backstop effectiveness or tenant commitments. Customer concentration is high: Q1 2026 HPC revenue came from one customer, while future growth relies on a small number of large counterparties. Power shortages, interconnection delays, equipment constraints, tariffs, and skilled-labor availability can raise cost or move delivery dates. Cybersecurity and physical uptime also matter because HPC leases impose service obligations.

The most important risk distinction is between contracted and operating. A signed lease can improve financing and visibility, but only completed facilities generate rent. Similarly, a gigawatt-scale site can be strategically valuable, yet it requires permits, transmission work, customer commitments, equipment, and billions of dollars of capital before reaching steady-state economics.

DCF analysis should be built from megawatts, timing, and unit economics

A useful TeraWulf model should not extrapolate consolidated revenue growth mechanically. It should forecast each campus by contracted critical IT megawatts, delivery date, rent commencement, price escalator, power pass-through, operating expense, maintenance capital, and financing cost. Mining deserves a separate scenario because its revenue depends on bitcoin price, network difficulty, fleet efficiency, curtailment, and power cost. Corporate stock compensation, warrants, convertibles, and potential equity issuance must then be reflected at the per-share level.

What should researchers monitor next?