The 10-Q Says What the Headline Won’t: TeraWulf Is No Longer a Bitcoin Miner

We didn’t start with the stock price. We started with the warrant line. TeraWulf reported a second-quarter net loss of $940.8 million. On that line alone, the stock should be dead. But $755.7 million of that loss came from a non-cash remeasurement of warrants — an accounting charge, not cash leaving the building. And beneath that charge, the operating story is completely different from the one most retail traders are following. HPC and AI leasing revenue hit $31.9 million in the quarter. Bitcoin mining generated $12.8 million, down 73% year over year. Total revenue was $44.8 million, roughly flat but with the mix violently shifted. AI/HPC is now 71% of revenue. This is not a miner with an AI side project. This is an AI data center lessor with a shrinking bitcoin hedge. We didn’t need a press release to know the pivot was structural. We needed the 10-Q. And the 10-Q says the transformation is already complete. Context: From Bitcoin Beta to AI Beta Let’s be precise about what TeraWulf is now. It owns and operates physical infrastructure: land, substations, power access, and buildings. Its Lake Mariner facility in New York is live with 102 MW of critical IT capacity. Another 336 MW is under construction. And Anthropic has signed for up to 401 MW in Kentucky. That is an anchor lease with a top-tier AI lab. The total contract value is reported at $190 billion, with an option to expand to $330 billion. The lease runs 20 years, plus a five-year renewal option. Add Fluidstack, which activated $600 million in Google credit support for TeraWulf’s capacity, and the counterparty quality is no longer the kind of paper-handed float that used to define bitcoin miners. Anyone who has spent time in crypto infrastructure knows how extraordinary that is. In 2017, I allocated $40,000 into an ICO with a strong technical team. The launch got overloaded, fees spiked 500%, and my position lost 30% before the crowd sale closed. That lesson is simple: technical pedigree means nothing if the infrastructure cannot handle demand. TeraWulf is now trying to sell physical infrastructure reliability to the most demanding compute buyers on the planet. The contracts are signed. The hard part is delivering. Core: The Physical Rebuild Nobody Wants to Discuss We didn’t stop at the revenue split, because the revenue split is backward-looking. The forward-looking question is whether a bitcoin mining site can be converted into an AI-grade data center without losing both time and margin. It is not a software update. Bitcoin miners are dumb boxes. They tolerate higher power usage effectiveness, they don’t need low-latency fiber, and their power density per rack is modest. AI clusters are the opposite. Modern GPU racks can draw over 100 kW each. They require liquid cooling, low-latency networking, and redundant power distribution with N+1 or 2N topology. That means the electrical distribution, cooling plant, and physical layout built for miners often has to be torn out and redesigned. TeraWulf is partially converting Lake Mariner, not building from scratch. That carries real execution risk. A conversion can take longer and cost more than a greenfield build if the original electrical buswork cannot handle high-density loads. I have spent years auditing smart contracts. In 2020, I found a reentrancy vulnerability in a yield aggregator and earned a whitehat bounty. That conditioned me to look for hidden failure modes. With code, the failure path is deterministic. With physical infrastructure, it is not. If TeraWulf misses the 2027 H2 delivery date for Anthropic, the damage is not just a penalty clause. It is the market’s confidence in the entire “miner to AI” thesis. The balance sheet tells the next part of the story. The massive warrant charge implies the company has been using warrants as a financing tool. That is not necessarily fatal — many high-growth infrastructure companies do the same — but it is a real dilution signal. And the cash-flow gap is obvious: HPC leasing only became the majority revenue line in the quarter, but the largest contract payments don’t start until late 2027. Between now and then, TeraWulf needs to fund 336 MW of construction. They can either use the $190 billion lease as collateral for project debt or print more equity. The math works only if the capital markets stay open. This is why I read the quarterly report differently from most crypto Twitter. A net loss of $1.4 billion year-to-date looks terrifying. Strip out non-cash warrant charges and the operating loss is much smaller — possibly close to break-even at the adjusted level. The market is pricing WULF like a struggling miner. The financial statements are starting to price it like a data center REIT with constrained supply and long-dated contracted cash flows. That mismatch is the opportunity, and also the trap. The competitive picture reinforces the shift. Core Scientific has signed large AI contracts, but its revenue mix is not yet led by HPC. IREN and Cipher are still earlier in the arc. CoreWeave is a purer AI data center operator, but it lacks TeraWulf’s already-owned power and physical campus portfolio. TeraWulf is not the biggest player in the AI infrastructure game. It is, however, one of the few former miners where the income statement already reflects the new identity. Contrarian: The Real Risk Is Not AI Demand. It’s Execution. The popular take is: “Bitcoin miner pivots to AI — another narrative pump.” That take has it backwards. The market has not yet repriced WULF as an infrastructure company. The CEO, Paul Prager, says control of power infrastructure is what matters at the AI buildout’s current bottleneck. That is not a marketing slogan. Grid interconnection lead times for large data centers are measured in years, not months. TeraWulf already has the grid access and the physical sites. That is a scarce resource. The contrarian angle is that the market is focused on the wrong tail risk. Everyone worries that AI demand will evaporate by 2028. I think the bigger risk is operational. A top AI lab like Anthropic will enforce service-level agreements on power availability and cooling performance. If TeraWulf’s converted mining facility underperforms, the 20-year lease won’t save the stock. The lease is only as valuable as the uptime behind it. We also didn’t buy the “single-customer” risk as the primary threat. Yes, Anthropic is the dominant future revenue driver, and customer concentration is high. But in infrastructure, a 20-year lease with a creditworthy counterparty reduces the need to guess future demand. The bigger risk is that TeraWulf has to raise capital in a high-rate environment before revenue starts. That is the hidden line item. Watch the financing announcements, not just the hash price. Takeaway: Watch the Deliverables, Not the Narrative I have learned to trust verification over stories. That is why I shorted TerraUSD in 2022 when the collateral math was already broken. That is why I sold the top of the NFT floor when volume contradicted price. And it is why I am not calling TeraWulf a sell or a buy today. The evidence is mixed by design. What I can tell you: the 10-Q proves TeraWulf has already made its strategic choice. The old business is a shrinking hedge. The new business is an AI infrastructure trade. The stock will not go back to being a simple bitcoin beta. The question is whether Paul Prager’s team can turn contracted megawatts into running data centers by 2027 H2. If they hit that date, the market will have to reclassify WULF as something closer to a digital energy REIT. If they miss it, the warrants, the construction costs, and the dilution will do the selling for them. We didn’t read the headline. We read the footnotes. The footnotes say TeraWulf is already a different animal. Now the market has to decide if it believes the delivery schedule.