Galaxy Digital's $85 Million Loss Hides a 5.7-Gigawatt Option

Galaxy Digital reported a net loss of $85 million for Q2 2025. Shares fell roughly 14%, from $22.14 to $19.07, in the session that followed. The causal chain looks simple: crypto prices pressured earnings, earnings missed, equity paid. That is the standard read. It is also incomplete. The macro view reveals what the micro ledger hides. The same press release that disclosed the loss disclosed something larger: 133 megawatts of AI data center capacity delivered at Helios, a 15-year lease signed with CoreWeave, and a power pipeline exceeding 5.7 gigawatts across four Texas sites. The market priced a loss. The filing described a transition. Galaxy Digital is not a token project. It is a Nasdaq and TSX-listed holding company, founded by Mike Novogratz, with regulated entities, institutional clients, and a balance sheet that spans crypto trading, asset management, principal investments, and physical infrastructure. Revenue in Q2 was $87 million, down from $102 million in Q1. Net loss narrowed from $216 million to $85 million, but EBITDA remained negative at $77 million. The digital asset segment produced adjusted gross profit of $66 million, up 34% quarter over quarter, while industry trading volumes fell roughly 7%. More profit per unit of flow in a quieter market is not the profile of a broken trading desk. It is a market-making operation extracting spread from dispersion. The losses are not coming from the desk; they are coming from the transition. I did not reach this view from the press release. In 2022, I spent four weeks reverse-engineering TerraUSD's reserve mechanics after the collapse. The lesson was not about stablecoins. It was about framing: a headline loss can obscure a structural change, and a structural change can obscure a headline loss. The same discipline applies here. A 14% stock drop following a loss is a reflex. The filing is the actual evidence. Let me walk through the asset side, because that is where the real accounting lives. Helios was originally a Bitcoin mining site. Galaxy has repurposed it into an AI data center campus. Phase I is delivering 133MW. The company holds three additional Texas sites, and the combined power pipeline is more than 5.7GW. Those are not mining machines. They are rack slots, cooling systems, and long-dated power agreements. The anchor tenant is CoreWeave, an AI cloud provider. The lease runs 15 years. The data center segment already recorded $20 million in gross profit in Q2, and management guided to roughly $80 million of incremental quarterly revenue beginning in Q3, with margins above 90%. Bitcoin mining is the wrong template for understanding this business. A mining machine consumes power and produces a commodity with fluctuating value. An AI lease consumes power and produces a contracted cash flow with a fixed price. The architecture looks similar. The cash flow profiles are opposites. That is why Helios is not a hedge. It is a substitution of one revenue class for another. This is project finance, not crypto speculation. A 15-year lease at 90% margins is a bond-like instrument wrapped in a landlord agreement. At $80 million per quarter, the data center line reaches a $320 million annualized run-rate. That does not transform the income statement on day one, but it changes the composition of future earnings. A balance sheet is a map of choices. Leverage is the compass. Galaxy's map points to Texas power corridors and a lease schedule, not to the next Bitcoin cycle. The funding side matters more. Galaxy raised $3.5 billion in senior secured notes due 2031 to fund Helios Phase II and the broader expansion. That is a large amount of leverage for a company with roughly $2.7 billion in equity. It is also a deliberate structure. The debt is secured by physical assets and contracted cash flows, not by a token. The collateral is a set of grid interconnection rights, land, and a 15-year tenant. Code does not lie, but it often obscures intent. Corporate statements perform the same trick. The line items are legible. The intent is visible only when the balance sheet is read as a sequence of bets. There is a timing mismatch. The notes mature in 2031, but the development pipeline will take years before each phase produces revenue. Senior secured debt does not wait for narratives. It accrues interest. With consolidated EBITDA still negative, Galaxy is borrowing against future lease payments. The structure only works if the construction calendar and tenant demand remain aligned. In project finance, that alignment is called execution risk. In crypto, it used to be called hope. The difference is that this time the counterparty signed a 15-year contract. The 5.7GW figure is the most misunderstood number in the release. It is not capacity. It is an option on future capacity, an option secured by permits, substations, and ERCOT relationships. To put it in context, 5.7GW is roughly the size of a mid-sized nuclear station. The market usually discounts illiquid rights like these. But in an AI arms race, power access is the bottleneck. CoreWeave signed a long lease because power, not chips, is the scarce input. Galaxy is not building a foundation model. It is renting the electrical floor to someone who needs it. That distinction matters. Real estate trades on contracted cash flow. Trading desks trade on beta. Galaxy is currently priced as the latter. Here is the contrarian position: Galaxy has already begun to decouple from Bitcoin, and the market has not updated its classification. The stock still moves with crypto because the revenue mix is still dominated by digital assets. But the marginal revenue that will define 2026 is locked into a 15-year infrastructure contract. That contract does not care whether Bitcoin trades at $40,000 or $80,000. The decoupling thesis does not claim that Galaxy is immune to crypto bear markets. It claims that the second engine is no longer a pitch deck. It is a signed lease with an AI cloud provider. The trap is leverage and concentration. CoreWeave is a credible tenant, but it is one tenant. If AI compute demand pauses, if CoreWeave renegotiates, or if a construction milestone slips, equity absorbs the first loss. The $3.5 billion note is secured. Bondholders sit ahead of shareholders in the capital stack. The 5.7GW pipeline will take years to develop, and carrying costs do not pause for permits. In 2020, I ran liquidity stress tests across Aave and Compound during a simulated stablecoin depeg. The conclusion was that correlated positions inside the same wrapper fail together. Galaxy is now a wrapper around two cash-flow engines with different time horizons. When the wrapper carries $3.5 billion of debt, timing mismatch is the risk. The market is also stuck on the wrong peer group. Riot and Marathon have announced similar pivots, but most of those announcements remain schedules, not assets. Galaxy has delivered 133MW and booked a named tenant. That is the difference between a narrative and a construction schedule. It does not make Galaxy safe. It makes it the first real test of whether the mining-to-AI route can create shareholder value or simply transfer it to bondholders. The next two earnings reports are the first honest data points. Watch three things: data center revenue, CoreWeave's payment cadence, and the ratio of EBITDA to interest expense. If AI revenue arrives near $80 million in Q3, the market will be forced to reclassify Galaxy as a hybrid infrastructure asset. If it misses, the debt stack becomes the story, and no Bitcoin rally will change that. The macro view reveals what the micro ledger hides. The micro ledger shows an $85 million loss. The macro view shows a 15-year lease, a 5.7GW power option, and a company borrowing against time. Whether that is a bridge or a trap depends entirely on execution. For now, the ledger is honest enough to watch.