TeraWulf's 71% Signal: The Bitcoin Miner That Quietly Became an AI Landlord

The most important number in TeraWulf's second-quarter report did not appear in any headline. It was $31.9 million. That is the revenue generated from HPC and AI data center leasing. It represents 71% of total quarterly revenue. The Bitcoin mining company, listed on NASDAQ under the ticker WULF, now earns less from mining Bitcoin than from renting out physical infrastructure to AI customers. Mining revenue collapsed 73% year-over-year to $12.8 million. Total quarterly revenue fell roughly 6% to $44.8 million. Net loss for the quarter: $940.8 million. Year-to-date: approximately $1.4 billion. On the surface, this reads like a disaster. It is not. The headline loss is largely an accounting artifact, and the revenue split tells a very different story about where this company is headed. TeraWulf has stopped being a Bitcoin mining stock. The market simply has not fully adjusted its valuation framework to reflect what it has become: a digital energy landlord. TeraWulf operates two critical sites. Lake Mariner in upstate New York, where the company ran its Bitcoin mining fleet. And Justified, the Kentucky campus that will host the Anthropic buildout. The capacity roadmap reads like a staged infrastructure deployment, not a mining operation. 81 megawatts live today. 102 megawatts of critical IT capacity after the current delivery phase. 336 megawatts under construction. And 401 megawatts earmarked for the Anthropic lease. There is an additional layer worth noting. New York state has imposed a moratorium on certain crypto mining operations. Kentucky is far more receptive. That regulatory geography may have accelerated the strategic decision to repurpose physical assets rather than expand the mining footprint. CEO Paul Prager has been explicit about the thesis. Control of power infrastructure is where value accrues. The constraint on AI is not chip supply. It is the electrical substation, the building shell, the cooling loops, and the ability to bring capacity online within a credible schedule. This is not an eccentric view. CoreWeave built a massive business on this exact premise. Core Scientific is executing the same playbook. But TeraWulf has moved faster than most of its listed mining peers on the revenue side. The company is now targeting 250 to 500 megawatts of new signed contracts every year. That is not a pace for a company testing the waters. That is a pace for a company that believes it has structural advantages in securing power and locked-in sites with real tenants attached. Let's begin with the accounting, because this is where most market participants will get the story wrong. The year-to-date net loss of $1.4 billion looks catastrophic at first glance. The Q2 net loss of $940.8 million looks even worse. But $755.7 million of the Q2 figure is a non-cash warrant liability remeasurement. TeraWulf issued warrants as part of its capital raising process. Under US GAAP, those warrants are marked to market every quarter. The fair value change flows directly through the income statement. The stock appreciated, the warrant liability appreciated, and the company recorded a loss that involved no cash leaving the business. The operational picture is considerably less dramatic. Adjusted EBITDA, which the report does not disclose, could be close to break-even or better. During my years analyzing public equity disclosures from crypto-adjacent companies, the single most misread accounting line has been the non-cash warrant remeasurement. It tells you the company has used structurally complex financing instruments. It does not tell you the business is failing. But it does mean investors should ask how many warrants remain outstanding, what their exercise prices are, and what future dilution sits embedded in the capital structure. We don't have those details in the filing. That gap is material. Now the revenue signal. Bitcoin mining contributed $12.8 million in Q2. HPC and AI leasing contributed $31.9 million. The composition is now 71% infrastructure and 29% mining. Twelve months ago, this split was nearly reversed. It took one year to invert the revenue structure entirely. Core Scientific took years to reach a comparable structural transition. The next decisive event is the second half of 2027. That is when Anthropic's rent payments begin. The lease runs 20 years with a five-year renewal option. Base contract value: $19 billion. Maximum: $33 billion. Put those numbers next to the current quarterly revenue of $44.8 million. The scale gap is enormous. This is a company that is supposed to be earning billions annually by the late 2020s, assuming it delivers the infrastructure on schedule. And there is the Fluidstack arrangement. That lease activated $600 million in credit support from Google. This is the most underappreciated data point in the entire report. Google does not attach its balance sheet to infrastructure projects it expects to fail. The fact that Alphabet is willing to backstop rental payments for a company that was primarily a Bitcoin miner twelve months ago is third-party validation that no press release can manufacture. Now the technical reality check. Bitcoin mining ASICs are forgiving infrastructure. They tolerate high power usage effectiveness. They are largely indifferent to network latency. They can operate in simple warehouse configurations with basic cooling. AI GPU clusters are a completely different species. A single high-density rack can draw more than 100 kilowatts. Liquid cooling is mandatory. Low-latency fiber networking is non-negotiable. The medium-voltage electrical distribution in a typical mining facility was never designed for loads at this density. TeraWulf has stated that part of Lake Mariner is being converted rather than built fresh. In my experience auditing infrastructure transitions, conversions are not automatically cheaper or faster than new builds when the end-state requirement is fundamentally different. The existing power distribution backbone likely needs to be redesigned. The cooling architecture must be replaced. Network infrastructure requires a complete overhaul. The cost and timeline of that re-engineering are the firm's biggest unproven variables. The real asset here is not the mining machines. It is the power access rights and the physical campus. That is where the scarcity lives. But the conversion risk is where value could evaporate. Comparisons to CoreWeave are useful but not exact. CoreWeave is the purest AI infrastructure operator, with elite GPU capabilities and hyperscaler anchor clients. TeraWulf cannot claim that level of AI-native operational expertise. What TeraWulf has secured is electricity access and physical sites at a scale that is becoming increasingly rare. The company's edge, if it holds, is land and power. The conversion risk sits on top of that edge. And the innovation here is not technological. This is not a novel protocol or a breakthrough engineering design. It is asset repricing: existing electrical assets reclassified into a higher-value use case. When the innovation is financial rather than technical, the margin for execution error narrows. There is no novelty premium to cover mistakes. The most dangerous way to read this story is as a straightforward rerating. Bitcoin miner pivots to AI. The beta changes. The market pays up. Volume spikes don't tell you whether the infrastructure will arrive on time. Three uncomfortable truths do not fit the clean narrative. First: single-customer concentration. The Anthropic lease dominates TeraWulf's future contracted revenue. Anthropic is a top-tier AI laboratory. It is also a single point of failure. If Anthropic's compute needs shift, if its financial trajectory bends, if its own competitive position weakens, TeraWulf is left with several hundred megawatts of custom-built capacity and few alternative tenants. The 20-year lease is a long lock-in. But the lock is bilateral. TeraWulf is committed to a specific buildout for a specific tenant. Flexibility evaporates. Second: the 2027 delivery window. Two to three years is an eternity in AI. If inference efficiency improves faster than expected, if specialized ASIC chips reduce dependence on general-purpose GPU clusters, if the macro environment pushes rental pricing downward, the world that generated a $19 billion contract may look very different by the time the first rent check arrives. Third: the valuation grammar is still confused. The market keeps framing WULF as a cryptocurrency proxy. Mining is now under a third of revenue. The correct frame is closer to a data center REIT with a contracted backlog and a 20-year anchor tenant. That is a different pricing regime. Price-to-sales multiples and EV/EBITDA replace hash price forecasts. The market has not fully internalized this switch. The code doesn't lie. But the construction schedule is a different kind of truth. Between the hash and the human, there is a silence: the gap between what the contract promises and what the engineering team delivers. Three signals will define the next year. Delivery milestones: does the 102-megawatt critical IT capacity target get hit on time? New signings: does the 250-to-500 megawatt annual pace hold? And the financing: TeraWulf will need capital before 2027H2. The shape of that raise, whether debt secured by contracted cash flows or dilution-heavy equity, will tell investors whether this transformation compounds value or transfers it. We don't know the answers yet. That is the trade. The transition is real. The execution is unproven. Between now and the first Anthropic rent check, delivery is the only story that matters.