WULF — knowledge base

Overview

TeraWulf ($WULF) is repositioning from Bitcoin mining toward long-term high-performance computing (HPC) and AI data-center infrastructure. Its strategy is to monetize scarce, large-scale power access through 10–15-year colocation agreements, using in-house energy and generation expertise to develop sites across New York, Texas, Kentucky, and Maryland.

The August 2025 Fluidstack agreement established the commercial model: approximately 200 MW of critical IT load at Lake Mariner, backed by roughly $1.8 billion of Google support and accompanied by warrants representing approximately 8% of TeraWulf. Fluidstack remains the disclosed customer; claims that Google is the sole end tenant are unverified. Updated: approximately 272.5 MW was publicly contracted in August 2025, while management reported more than 500 MW by March 2026 and expects more than $1 billion of run-rate NOI or EBITDA exiting 2027. These later targets remain dependent on contract terms, financing, construction, interconnection, tenant performance, and on-time energization.

Colocation providers own land, power infrastructure, cooling, and building shells with 15–40 year asset lives while tenants supply GPUs depreciated over 3–5 years. Leases run 20–30 years with 2–3% annual rent escalators. TeraWulf targets mid-to-high teens yield on cost and mid-teens yield on lease. The Anthropic Kentucky lease covers up to 401 MW critical IT load.

Key facts & figures

Contracted capacity—updated: The August 2025 total was approximately 272.5 MW: roughly 72.5 MW for Core42 plus 200 MW for Fluidstack, not the 262.5 MW claimed during the discussion. By March 2026, management reported more than 500 MW of critical IT load under long-term contracts, approximately 10% online and nearly all of the balance scheduled by year-end 2026; the later company-specific figures remain unverifiable without current contracts and filings. [[s:61@00:18:42]]
Core42 correction: The claim that Core42 accounted for 62.5 MW was inaccurate; publicly announced initial and expanded commitments totaled approximately 72.5 MW of critical IT load. [[s:78@00:04:56]]
Fluidstack agreement: TeraWulf agreed to provide approximately 200 MW of critical IT load at Lake Mariner, requiring roughly 250 MW of gross capacity, with phased delivery during 2026 and the balance expected by year-end. [[s:78@00:05:28]]
Fluidstack expansion option: Fluidstack received an option for approximately 160 MW of additional critical IT load. The option is contingent and should not be included in contracted operating-capacity totals unless exercised. [[s:78@00:05:57]]
Google credit support: Google agreed to backstop approximately $1.8 billion of Fluidstack lease obligations and received warrants representing roughly 8% of TeraWulf on a pro forma basis. [[s:78@00:10:50]]
Potential Google ownership: Additional warrants could increase Google’s stake to approximately 14% if specified expansion and funding conditions are met. Claims that ownership will rise to 14%–15% were misleading because the outcome is contingent and 15% is imprecise. [[s:78@00:11:12]]
End-user identity: Recognition-agreement protections for Google’s hardware and intellectual property, together with the backstop and warrants, indicate significant Google involvement but do not prove Google is the sole or ultimate tenant; Fluidstack is the publicly disclosed contracting customer. [[s:78@00:08:56]]
Contract economics: Management describes contract terms of 10–15 years with approximately 85% cash-flow or NOI margins. Both the terms and projected margin are unverifiable and should not be treated as realized economics. [[s:61@00:19:22]]
Insider alignment: Board, management, and other insiders own approximately 25% of the equity.
Lake Mariner, New York: Management later reported approximately 500 MW of gross capacity committed under long-term agreements, with another 250 MW being pursued through the interconnection queue. A prior claim that Lake Mariner already had 750 MW of usable power was misleading because that figure included expansion potential rather than solely energized capacity. [[s:78@00:04:28]]
Bitcoin mining: Lake Mariner reportedly operates 150–170 MW of Bitcoin-mining load, which management plans to reduce as AI capacity expands; current load, reusable infrastructure, and transition timing are unverifiable. [[s:61@00:25:07]]
Cayuga, New York: Approximately 400 MW was described as prospective development capacity and a major leasing catalyst, not currently energized or contracted critical IT load. Development and interconnection work remain necessary.
Abernathy, Texas: The joint venture represents 84 MW of critical IT capacity net to TeraWulf and is located in the Southwest Power Pool, not ERCOT. The grid-market characterization is accurate, while the precise economic share remains company-reported. [[s:61@00:23:15]]
Hawesville / Muskie, Kentucky: The acquired former Century Aluminum smelter has approximately 480 MW of gross power infrastructure. Management targets roughly 380 MW of critical IT capacity after redevelopment; an Anthropic lease covers up to 401 MW critical IT load on 20–30 year terms valued at $19 billion. [[s:61@00:23:44]]
Kentucky schedule: Management targets the second half of 2027 for approximately 380 MW of critical IT load to come online; this is unverifiable forward guidance dependent on leasing, financing, permitting, and construction. [[s:61@00:28:05]]
Morgantown, Maryland: The proposed configuration comprises 1,000 MW of load, 1,000 MW of generation, and 500 MW of battery storage, with power targeted for late 2028 or 2029. These are unverifiable development plans, not operating assets. [[s:61@00:24:17]]
Regulatory review: Management’s suggestion that FERC’s Morgantown acquisition review focuses on market power without materially considering end use, repowering, or fuel mix was rated misleading. Market power is central, but FERC can also examine rates, regulation, and cross-subsidization, while other agencies may review environmental and operational matters. [[s:61@00:34:34]]
Illustrative Fluidstack valuation: A projection that 422 MW could generate approximately $700 million of revenue and $500 million of EBITDA was misleading if presented as an operating outlook. It included the unexercised 160 MW option and assumed approximately $1.65 million of annual revenue per MW, an 85% gross margin, and favorable SG&A economics. [[s:78@00:07:19]]
Financial target: Management expects more than $1 billion of run-rate NOI or EBITDA exiting 2027 if Wolf Compute and Kentucky perform as projected. This is forward-looking guidance rather than established earnings.
Development cadence: Management aims to sign an additional 250–500 MW annually and generally deliver contracted capacity 12–15 months after signing.
Capital strategy: The stated priorities are debt reduction, refinancing operating-company obligations at lower rates, maintaining moderate parent-level leverage, and ultimately supporting an investment-grade credit profile. New sites target approximately 20% equity with project-level debt at 6–7%.
Credit-rating claim: Google’s backstop should improve project financeability, but the claim that it should produce a roughly BBB issuer rating is unverifiable; ratings also depend on leverage, guarantees, corporate structure, covenants, cash flow, and agency methodology. [[s:78@00:11:48]]
Erroneous market statistic: A claim that ERCOT has roughly 88,000 GW of capacity against 280,000 GW of demand was rated inaccurate because the units are off by approximately three orders of magnitude; relevant ERCOT system figures are normally expressed in MW. [[s:61@00:54:01]]
Power costs and construction: US industrial wholesale power averages 80–100 $/MWh. TeraWulf targets 20–30 $/MWh via combined-cycle generation plus batteries. Data-center construction costs approximately 10 million $/MW; combined-cycle plants cost 2.5–3 million $/MW with heat rate of roughly 6000 BTU/kWh.
Muskie, Kentucky: Prospective interconnection from Kentucky Power subsidiary starting Q3 2028, ramping toward 1000 MW by 2030.
Anthropic lease escalators: The 20–30 year Kentucky lease includes approximately 3% annual rent escalators, producing roughly $725 million in year-one rent rising to $1.2 billion by year twenty.
Yield target: TeraWulf targets 15–19% yield on cost for colocation projects.
Grid-connected sites: All TeraWulf data-center sites are grid-connected rather than behind-the-meter.

Thesis & bull case

Institutionally validated model: The Fluidstack agreement, Google backstop, and Google warrants provide stronger external validation than a conventional speculative HPC conversion announcement.
Improved financeability: Approximately $1.8 billion of Google support may lower project-financing costs, expand lender appetite, and reduce the equity required to construct the Fluidstack capacity, although it does not guarantee an investment-grade corporate rating.
Strategic Google alignment: Google’s warrants create a direct economic interest in TeraWulf’s success. Additional warrants tied to expansion could further align the parties if the 160 MW option is exercised.
Scarce powered land: TeraWulf controls multiple sites with hundreds of megawatts of existing or prospective power infrastructure, potentially allowing it to deliver capacity faster than greenfield competitors; energized interconnection on a known schedule is the scarce input.
Large contracted base: If management’s March 2026 figures and schedules prove accurate, more than 500 MW could be operating by year-end 2026 under long-duration agreements, creating recurring, infrastructure-like cash flow.
Expansion embedded in existing relationships: The Fluidstack option could add 160 MW without sourcing an entirely new customer relationship, while Lake Mariner’s interconnection expansion may add another 250 MW.
Attractive unit economics: Ten- to 15-year terms and targeted NOI margins near 85% would support substantial operating leverage, although neither figure has yet been independently verified. Rent escalators in 20–30 year leases compound value.
Power-generation expertise: Management presents TeraWulf as an energy-infrastructure company rather than merely a data-center landlord. Experience with generation, grids, curtailment, and industrial-site rehabilitation may improve site selection and execution.
New York positioning: Lake Mariner and Cayuga may offer relatively low-latency access to major Northeast markets, supporting inference, enterprise, and other workloads for which proximity can matter more than for model training.
Regional diversification: New York, Texas, Kentucky, and Maryland expose the company to different power markets and development pathways rather than concentrating all capacity in ERCOT or one utility territory.
Embedded site expansion: Lake Mariner may add another 250 MW, Cayuga represents approximately 400 MW of prospective capacity, Hawesville could support approximately 380 MW of critical IT load, and Morgantown represents a potential gigawatt-scale generation-and-load campus; sites total multiple gigawatts of potential capacity.
Premium 2027 capacity: Management believes near-term powered capacity is scarce enough that customers will pay attractive economics for credible 2027 delivery; market discussions are increasingly shifting toward 2028 supply.
Tenant selectivity: Prioritizing high-credit customers or obtaining hyperscaler backstops can lower financing costs, increase project leverage, and reduce contract-default risk.
Cash-flow inflection: Successful delivery of Wolf Compute and Kentucky could produce more than $1 billion of run-rate NOI or EBITDA exiting 2027, creating capacity for debt repayment, refinancing, reinvestment, or shareholder returns.
Insider incentives: Insider ownership of approximately 25% creates substantial alignment with common shareholders.
Self-funding flywheel: After initial equity raises, project-level debt and operating cash flow enable future developments without dilution; by mid-2027 the company expects to fund future deals from its own cash and operating income.
Grid-connected reliability: All sites are grid-connected rather than behind-the-meter, supporting SLAs and long-term competitiveness.

Risks & bear case

Fluidstack counterparty structure: Google’s backstop materially strengthens the arrangement, but investors still need the exact scope, duration, caps, conditions, remedies, and termination provisions. A backstop is not necessarily equivalent to Google directly leasing all capacity.
End-user speculation: Claims that Google is the ultimate tenant or that Lake Mariner will support Waymo’s Northeast operations are unverifiable. Recognition agreements protect property and contractual rights but do not establish the workload or sole end user. [[s:78@00:10:13]]
Unexercised option risk: The additional 160 MW should not be capitalized as contracted capacity. Expansion depends on Fluidstack and Google decisions, funding conditions, site readiness, and demand.
Execution intensity: Bringing nearly all of more than 500 MW online by year-end 2026, followed by approximately 380 MW in Kentucky during 2027, requires simultaneous financing, procurement, construction, commissioning, and tenant coordination.
Unverified contracts and economics: Contracted megawatts, 10–15-year terms, 85% margins, tenant credit quality, and the relationship between corporate and project-level guarantees require confirmation through filings or executed agreements.
Construction complexity: Bitcoin-mining buildings and electrical systems are not automatically suitable for dense GPU deployments. HPC conversion may require new structural systems, networking, redundancy, security, liquid cooling, chillers, and higher-quality power distribution.
Kentucky leasing risk: Hawesville’s value depends on securing a creditworthy tenant on acceptable economics. Term sheets are nonbinding, and the expected signing by the end of Q2 2026 remains a forecast.
Cayuga development risk: The site’s approximately 400 MW is prospective. Full-site leasing, interconnection, permitting, financing, remediation, and construction are not assured.
Financing and leverage: Data-center campuses require substantial upfront capital. Delays, cost overruns, weak tenant credit, or higher interest rates could increase parent-level leverage and impede the investment-grade objective.
Warrant dilution: Google’s initial warrants represent approximately 8% pro forma ownership, and conditional expansion warrants could raise that stake to approximately 14%. The strategic and financing benefits must be weighed against dilution to existing shareholders.
Tenant concentration: Large campuses may depend on Fluidstack, Core42, Anthropic, or a small number of hyperscale and AI customers, making counterparty credit, guarantees, renewal behavior, and workload concentration disproportionately important.
Power and interconnection uncertainty: Gross site power does not equal deliverable critical IT load. Grid studies, transmission upgrades, redundancy requirements, utility approvals, cooling loads, and power-usage effectiveness reduce capacity or delay energization.
Maryland regulatory risk: Morgantown must navigate acquisition approval, generation and environmental permitting, interconnection, fuel and storage decisions, and potentially broader public-interest scrutiny.
Bitcoin transition risk: Reducing mining removes a flexible source of revenue before replacement AI capacity is fully operational. The claim that mining exists solely to support the New York grid was misleading, since Bitcoin production itself generates revenue. [[s:61@00:58:30]]
Customer-demand cyclicality: AI infrastructure demand is currently strong, but rapid supply additions, better chip efficiency, changing model economics, or hyperscaler capital-spending cuts could weaken future pricing.
Valuation sensitivity: Bullish scenarios depend on options being exercised, full utilization, projected revenue per MW, approximately 85% margins, timely delivery, limited dilution, and high infrastructure multiples. Small changes in these assumptions could materially reduce equity value.
Disclosure credibility: The understated historical Core42 total, misleading aggregation of energized and expandable capacity, inaccurate ERCOT statistic, and overly narrow characterization of FERC review highlight the need to verify promotional claims against contracts, regulatory records, and filings.
Regulatory moratoriums: New York restrictions may limit Lake Mariner expansion pace despite negative declarations for certain phases.

Timeline of developments

2025-08-17: TeraWulf’s announced Fluidstack transaction covered approximately 200 MW of Lake Mariner critical IT load, an option for another 160 MW, approximately $1.8 billion of Google credit support, and warrants giving Google roughly 8% pro forma ownership; discussion claims that Google was definitively the end tenant or that total contracted capacity was exactly 262.5 MW were not substantiated. [[s:78@00:10:50]]
2026-03-09: Management outlined the broader transition from Bitcoin mining to AI/HPC infrastructure, reporting more than 500 MW of contracted critical IT load, plans for Lake Mariner expansion, approximately 380 MW at Hawesville in 2H 2027, and a gigawatt-scale Morgantown development targeted for late 2028 or 2029. [[s:61@00:18:42]]
2026-07-09: Discussion highlighted colocation economics, 20–30 year leases with rent escalators, Anthropic Kentucky lease for up to 401 MW critical IT load valued at $19 billion, and TeraWulf’s mid-to-high teens yield-on-cost target.
2026-07-09: Space confirmed colocation asset lives of 15–40 years versus 3–5 years for GPUs, 20% equity / project-debt model, all sites grid-connected, and 20–30 $/MWh power-cost target via combined-cycle generation plus batteries. [[s:89]]
2026-07-23: Patrick detailed grid-connected site strategy, 20–30 $/MWh power cost target via combined-cycle generation plus batteries, Muskie Kentucky interconnection timeline, and long-term REIT conversion goal.

Open questions

What contractual developments increased publicly identified contracted capacity from approximately 272.5 MW in August 2025 to management’s reported 500+ MW by March 2026?
Which customers account for the reported 500+ MW, and what are their credit ratings, guarantees, termination rights, funding obligations, and ultimate end users?
What exactly does Google’s $1.8 billion backstop cover, and under what conditions can it be reduced, terminated, or enforced?
Is Google itself deploying workloads at Lake Mariner, or is its role limited to credit support, equipment ownership, strategic alignment, or services supplied through Fluidstack?
What conditions must be satisfied before Fluidstack can exercise the 160 MW option, and what construction funding, pricing, and additional warrants would follow?
How much of the contracted capacity is take-or-pay, and how much remains conditional on construction milestones, financing, customer acceptance, or upstream end-user commitments?
Does the targeted 85% margin refer to site-level NOI, unlevered cash flow, EBITDA, or another non-GAAP measure, and which expenses are excluded?
What total capital expenditure, TeraWulf equity contribution, and debt funding are required to deliver the contracted pipeline?
How much do Google’s backstop and warrants lower the cost of debt and required TeraWulf equity compared with an unbacked Fluidstack lease?
What restrictions apply to Google’s warrants, and can they be sold, transferred, pledged, or exercised before related project debt is repaid?
What portion of project debt will be nonrecourse, and what guarantees or cross-collateralization will remain at the parent company?
Will Cayuga’s prospective 400 MW be leased to one customer or multiple tenants, and what interconnection, remediation, permitting, and construction milestones remain?
Will Hawesville secure one anchor tenant or multiple customers, and what economics are required to support the projected 380 MW deployment?
What grid, transmission, and redundancy work is necessary to convert Hawesville’s 480 MW gross power into approximately 380 MW of critical IT capacity?
Has the additional 250 MW at Lake Mariner received an interconnection position, study results, or firm delivery timetable?
How quickly can the 150–170 MW Bitcoin-mining load be converted to AI use, and how much equipment, building, cooling, and electrical infrastructure can be reused?
What are Morgantown’s acquisition-closing conditions, permitting pathway, proposed generation technology and fuel source, and expected capital cost?
How much of the projected $1 billion-plus run-rate NOI or EBITDA exiting 2027 comes from executed contracts versus unsigned Kentucky or other prospective capacity?
What leverage ratio and ratings-agency metrics define management’s “moderate” parent-level leverage and investment-grade goal?
Once free cash flow materializes, how will management prioritize debt repayment, new campuses, acquisitions, buybacks, or dividends?
What is the precise interconnection status and timeline for Muskie reaching 1000 MW?

Notable predictions to track

Fluidstack and Google will exercise the 160 MW expansion option; Perry Lin assigned this a 95% probability, but the option remained contingent. [[s:78@00:05:57]]
Google’s ownership will rise from approximately 8% to about 14% if the specified expansion and funding conditions are satisfied. [[s:78@00:11:12]]
Cayuga’s entire prospective 400 MW will be leased to one customer within six months of the August 17, 2025 discussion, implying approximately February 17, 2026.
Nearly all of the reported 500+ MW of signed critical IT load will be online by December 31, 2026. [[s:61@00:18:42]]
TeraWulf will sign an additional 250–500 MW annually, with typical delivery 12–15 months after contracting.
A high-credit customer will be secured for Hawesville by the end of Q2 2026, with term sheets already under negotiation; both the negotiations and deadline are unverifiable. [[s:61@00:31:45]]
Approximately 380 MW of Hawesville critical IT capacity will become operational in 2H 2027. [[s:61@00:28:05]]
Lake Mariner will obtain another 250 MW of capacity and progressively replace its 150–170 MW Bitcoin-mining load with AI compute.
TeraWulf will largely exit Bitcoin mining within approximately 18 months, implying around September 2027 from the March 2026 discussion; this remains an unverifiable forecast. [[s:61@00:58:30]]
Wolf Compute and Kentucky will support more than $1 billion of run-rate NOI or EBITDA exiting 2027.
Morgantown will ultimately support 1,000 MW of load, 1,000 MW of generation, and 500 MW of battery storage, with power available in late 2028 or 2029. [[s:61@00:24:17]]
Google’s support will materially lower financing costs and help TeraWulf progress toward an investment-grade credit profile, though a specific BBB rating cannot be inferred from the backstop alone. [[s:78@00:11:48]]
Operating-company debt will be refinanced at lower rates while parent-level leverage remains moderate and progresses toward an investment-grade rating.
Muskie site reaches 1000 MW by 2030.