The 9.875% Confession: Galaxy Digital's $3.5B AI Pivot and the Math That Doesn't Close

Galaxy Digital lost $85 million on its crypto business in Q2. Management points to $80 million in projected quarterly AI revenue as the offset. The symmetry is seductive. It is also an accounting fiction. The $85 million is a net loss. The $80 million is top-line lease revenue. They live on different planes of financial reality. Push past the headline arithmetic, and you find a number nobody in the coverage is quoting: $346 million. That is the annual interest expense on the $3.507 billion in senior secured notes Galaxy issued through its subsidiary, Galaxy Helios II LLC, at a 9.875% coupon. Against Phase I's projected $320 million annualized lease revenue. Against roughly $288 million in project-level adjusted EBITDA, if you trust a >90% margin guidance that excludes overhead, excludes interest, excludes everything that actually makes a company solvent. The math doesn't close. Not yet. That's the story. I've audited enough ERC-20 contracts to recognize a reentrancy vector in narrative form. When a crypto financial services firm borrows at nearly ten percent to build infrastructure it doesn't operate, under a single-tenant lease, and tells the market "the revenue will come," you aren't looking at a growth story. You are looking at a leveraged options position on construction execution and counterparty solvency. The architecture of trust, stripped to its bones, is a debt covenant wearing a data center costume. Let me be clear about what Galaxy has actually done, because the transformation is significant. This is not a bitcoin miner bolting GPU racks onto existing power infrastructure. Galaxy has constructed, from the ground up, a separate project entity that owns physical data center capacity. Phase I is 133 MW of critical IT load, fully operational, leased to CoreWeave for fifteen years. Phase II adds 260 MW, financed by those $3.5 billion in notes, with handover expected in 2027. The strategic rationale is familiar. The AI compute demand curve is steep. Physical infrastructure with contracted power and liquid cooling is the scarce asset in that equation. Galaxy wants to be the landlord in a market where every square megawatt is being fought over. The playbook mirrors what CoreWeave itself did to NVIDIA in the GPU procurement wars — secure the asset, sign the lease, monetize the scarcity. But the financing terms betray the true risk assessment. A 9.875% secured coupon is not an investment-grade rate. Investment-grade infrastructure paper moves at four to six percent. This is a credit spread of roughly 400 to 500 basis points over comparable corporate debt, on a secured note with physical collateral. The bond market is saying: we believe in the collateral, we are less certain about the timing and the cash flow coverage. This is not a vote of confidence in the AI thesis. It is a premium charged for execution risk. THE INTEREST COVERAGE GAP The central calculation deserves repetition because it anchors everything else. $3.507 billion multiplied by 9.875% equals $346 million in annual cash interest. Management guidance: $80 million per quarter of run-rate lease revenue. Annualized, that is $320 million. Project-level adjusted EBITDA margin: above 90%. That implies roughly $288 million in annual EBITDA from Phase I at full run rate. $288 million against a $346 million interest bill. A shortfall of $58 million per year. Twenty-two percent of EBITDA. Before principal amortization. Before maintenance capital expenditures. Before the crypto business bleeds another quarter. Three escape hatches exist. Interest capitalization during construction. Phase II revenue arriving earlier than scheduled. Additional financing at better terms. All three are assumptions. None is discretionary. The market will look at the $85 million loss and the $80 million revenue number and declare them roughly offset. Two wrong numbers canceling into a comfortable illusion. But in reality, the income statement and the cash flow statement are telling different stories in different currencies. The gap between them is where companies die. This kind of analysis is not speculation. It is the same structural logic I applied during the 2020 DeFi stress tests, when my team simulated high-frequency trading against Uniswap V2's AMM mechanics under extreme volatility. The liquidity pools looked mathematically sound right up until they didn't. A 90% margin on paper meant nothing when the marginal constraint was a single liquidity pool's depth under adversarial conditions. Here, the marginal constraint is a single tenant's willingness and ability to keep paying rent for fifteen years while its own leveraged position compounds. WHY THE MARGIN METRIC MISLEADS The >90% project-level EBITDA margin looks spectacular. It is also a managed figure. It excludes general and administrative expenses. It excludes interest. It excludes every other segment of the business. It is project-specific: a real estate net operating income metric dressed up as a profitability signal for the whole enterprise. The company-level picture is less flattering. Q2 net loss: $85 million. Data center segment adjusted gross profit: $20 million. Segment adjusted EBITDA: $11 million. Treasury segment adjusted gross loss: $42 million. The company's core crypto market-making and asset management operations are still bleeding. This is the inverse of what a healthy transformation looks like. A transition story should show the new business growing while the old business stabilizes. Instead, the new business is growing from a small base while the old business is still in active drawdown. The total is not yet less risky. It is simply two risks stacked in sequence. The treasury loss is particularly instructive. $42 million in adjusted gross loss on a treasury book is not a small miss. It reflects the ongoing volatility and structural headwinds in crypto market making. And if the company's attention is shifting to data center construction, the treasury operation's risk management may become less proactive. This is the diversion-of-attention risk that no spreadsheet will capture. THE SINGLE-TENANT CONSTRAINT Galaxy's quarterly filing deserves credit for one thing: honesty. It explicitly acknowledges that the data center segment "initially relies heavily" on the AI infrastructure customer. That customer is CoreWeave. One tenant. One hundred percent of Phase I revenue. The deal structure is clean. A 15-year lease. CoreWeave pays rent for the 133 MW of critical IT load. Galaxy holds the physical asset. It doesn't manage GPUs, doesn't chase inference workloads, doesn't need to speak the language of model training clusters. It is a straightforward real asset lease. The risk, however, is a concentration risk so severe that it warrants its own stress-test framework. CoreWeave is not a utility. It is a highly leveraged AI cloud provider whose own business model depends on the continuous inflow of institutional capital to fund GPU purchases and data center expansion. CoreWeave's cost of capital is elevated. Its client concentration is similarly narrow — a handful of AI labs and, increasingly, a few hyperscale tenants. CoreWeave just raised $20 billion in financing. That is a validating signal. It shows that the market still wants exposure to AI compute. But it also demonstrates the fragility of the underlying business model. Twenty billion dollars is not raised cheaply. It comes with covenants, delivery obligations, and the expectation of rapid revenue growth. If AI inference demand softens, or if a major customer cancels capacity, CoreWeave's own financial stress will transmit directly to Galaxy's rent payments. The chain of leverage: bondholders → Galaxy → CoreWeave → AI labs → VC-funded model companies. Every layer is levered. Every layer depends on the next layer's growth continuing. When a chain of this kind breaks, it doesn't break at the top. It breaks at the weakest link. And the weakest link is whichever layer experiences the first demand shock. If CoreWeave asks for rent relief, Galaxy has few options. Physical data centers have no liquidity. You cannot sell 133 MW of liquid-cooled capacity in a fire sale and recover book value. The asset has a very narrow buyer pool. The 15-year lease is the only thing giving it value, and the lease is only as good as CoreWeave's balance sheet. This is why the single-tenant risk cannot be diversified away. It is structural. It is the entire business model. And it is precisely what the bond market is charging 9.875% for. THE 2027 FIXED COST CLIFF The part of this story that is getting the least coverage: the fixed cost obligation that activates in 2027. The Phase II notes mature in 2031. Handover of the 260 MW expansion is expected in 2027. But the key clause is this: from 2027, the project company will bear fixed cash costs. This means that on a specific date, regardless of whether Phase II is productive, regardless of whether the tenant has taken delivery, regardless of whether the revenue has started flowing, the fixed costs begin. The implication is a liquidity rigidity that construction finance professionals recognize immediately. The costs are a hard commitment. The revenue is contingent. Between the 2027 cost activation and the point where Phase II reaches full utilization, there is a gap. That gap must be bridged by cash reserves, additional financing, or interest capitalization provisions. If Phase II is delivered on schedule and immediately revenue-producing, the gap is small and temporary. If it is delayed — by supply chain friction, electrical inspection regimes, cooling system commissioning issues — the interest bill keeps accruing, the fixed costs keep running, and the company's liquidity buffer erodes. This is where equity dilution becomes likely. Not as a choice but as a necessity. The market should price that dilution today. When I modeled the interoperability challenges between Bitcoin spot ETFs and CBDC settlement systems in 2024, one pattern kept surfacing: two systems that each perform well in isolation can produce unpredictable frictions when forced to interoperate. Same logic here. Phase I performs. Phase II performs in a spreadsheet. The seam between them — construction schedule, commissioning, tenant acceptance — is where the value destruction happens. THE CREDIT MARKET'S VERDICT The 9.875% coupon is the most objective piece of information in this entire story. It is the market's collective price discovery on the risk of this structure. Let's decode it. Investment-grade corporate debt in the current market: 4% to 6%. High-yield with significant risk: 7% to 9%. A secured note backed by physical infrastructure at 9.875% is signaling measured distrust. The collateral exists. The sponsorship is credible. But the market is not confident in the timing or the cash flow sufficiency. Probability-weighted, a 9.875% coupon over a 15-year horizon implies the market is pricing in a meaningful chance of distress. Not necessarily bankruptcy. But a restructuring, a covenant breach, an asset sale at depressed values. The financing is at the subsidiary level. Galaxy Helios II LLC. Not the parent. That structure is deliberate risk isolation — the parent does not put its entire balance sheet behind the project. Bondholders accept the project risk without a parental guarantee. This is the kind of structure that protects the parent while leaving the project company exposed to higher funding costs. For investors in Galaxy stock, the subsidiary-level debt is a double-edged sword. It limits downside contagion to the parent if the project fails. But it also means the parent cannot easily extract value from a successful project beyond the equity stake in the subsidiary. The value separation is real. THE CAPITAL EFFICIENCY PROBLEM Let's run the return on capital math in open air. Galaxy has deployed $3.5 billion into AI infrastructure. The revenue run rate is approximately $320 million per year. That is a 9% gross yield on capital deployed, before interest costs, before maintenance, before operating overhead. Nine percent. On physical infrastructure. With construction risk. With single-tenant credit risk. With obsolescence risk in a sector where the underlying technology changes every generation. In a low-rate environment, a 9% yield with long-duration contracted cash flows might be acceptable. In a higher-rate environment, where the risk-free rate is meaningfully above 3%, a 9% yield on a leveraged structure with binary tail risk is thin compensation. The market is compensating for this through terminal value assumptions. The narrative assumes the AI infrastructure will be worth significantly more in five years than it costs today. That is plausible. It is also unverifiable. The construction delivery schedule and the rental trajectory over the first three years will define whether the terminal value thesis holds. When I was stress-testing Uniswap V2 in 2020, the equivalent error was assuming that impermanent loss arithmetic was symmetrical. It isn't. Losses and gains are not mirror images. In the same way, the upside case for Galaxy's AI project is not the mirror image of the downside case. Upside requires multiple variables to align: construction on schedule, tenant growth, AI demand growth, sustained pricing power in a competitive capacity market. Downside requires only one: a failure of any single link in the chain. Q3 IS THE VERIFICATION POINT This is why Q3 is the most important reporting period Galaxy has faced since its introduction. The guidance is $80 million per quarter at run rate. Q2 showed the ramp phase — $20 million in gross profit, $11 million in segment adjusted EBITDA. Q3 is the first full quarter where Phase I should be operating at full capacity. If Q3 delivers near the $80 million revenue run rate, the project economics start to validate. The interest coverage gap narrows. The story becomes credible. If Q3 delivers $50 million, the market will begin discounting the 2027 Phase II narrative and the stock will re-rate. The reaction window is bidirectional. In a market that has already assigned premium multiples to AI-related infrastructure concepts — as VanEck has repeatedly cautioned, noting that AI-linked miners trade at premium valuations before most of their leased capacity is physically delivered — the downside asymmetry is more acute than the upside potential. Expectations have run ahead of financial reporting reality. VanEck's warning is not about Galaxy specifically. It is about the entire category. When a sector's valuations are predicated on future capacity that has not yet produced revenue, the first missed guidance is a sector-wide repricing event. Galaxy is the test case for that dynamic in the crypto-crossover space. THE COMPETITIVE LANDSCAPE Galaxy occupies an unusual position. It is neither a pure crypto mining company pivoting to AI nor a pure data center REIT. It is a hybrid: crypto financial services with an AI data center arm. Compared to Riot Platforms, IREN, or Cipher Mining, Galaxy is ahead in one dimension — Phase I is already operational and generating contract revenue. Those miners are still in earlier stages of their AI/HPC transformations. But Galaxy is behind in another dimension: none of those miners has a $3.5 billion debt instrument with a 9.875% coupon attached to its AI project. Compared to CoreWeave itself, Galaxy is an asset holder rather than an operator. That distinction matters. CoreWeave has deep expertise in GPU cluster deployment, workload optimization, and AI customer acquisition. Galaxy has a lease. When the AI market experiences a soft patch, the operator has levers to pull — infrastructure optimization, customer negotiation, workload diversification. The landlord does not. The landlord waits for the rent check. The pure-play data center REITs — the Equinixes of the world — have a different credit profile. They have diversified tenants, decades of operating history, and investment-grade balance sheets. Galaxy has one tenant, one project, and a construction pipeline. The risk premium embedded in its coupon reflects exactly that distance from the established infrastructure players. THE CRYPTO DIVISION'S PARALLEL REALITY While the AI division is the headline story, the crypto treasury operation continues to produce losses. Q2's $42 million adjusted gross loss in the treasury segment is not a rounding error. It is evidence that the cyclical volatility of crypto market making remains alive and well. The narrative that the AI business will "stabilize" the crypto volatility misses a more important point. The AI business is not a hedge. It is a separate, equally volatile risk profile expressed in physical assets instead of digital assets. Crypto trading losses are mark-to-market. Data center losses are construction overruns, tenant defaults, and commissioning delays. Both are volatile. Neither stabilizes the other. The company's capital allocation decision to fund data center construction through debt while its treasury segment is losing money is a signal. It says management believes the AI opportunity is more attractive than the crypto trading opportunity. If that is the thesis, investors in Galaxy stock are effectively buying a leveraged AI infrastructure bet wrapped around a declining crypto business. One of the hidden ironies: the very capital flowing into AI infrastructure is draining from crypto markets. CoreWeave raised $20 billion. Galaxy raised $3.5 billion. That money is not buying bitcoin. It is not providing DeFi liquidity. It is not funding derivatives trading. The institutional capital rotation toward AI is structurally bearish for crypto market depth, even as it creates new revenue streams for crypto-adjacent companies. Galaxy is both a beneficiary and a victim of this migration — its data center business gains, its treasury business loses. REGULATORY DIMENSIONS The financing structure itself is unremarkable from a securities law perspective. Secured notes issued by a subsidiary, registered with the SEC, governed by traditional corporate bond documentation. No novel token issuance, no decentralized governance, no regulatory gray zone. This is as conventional as corporate finance gets. For the crypto side of the business, the regulatory picture remains murky. Galaxy operates licensed subsidiaries in various jurisdictions. Its SEC reporting obligations are routine. But the changing U.S. regulatory landscape around digital asset classifications continues to pose a tail risk to the treasury business. The data center business faces a different regulatory class: energy permitting, grid interconnection approvals, environmental compliance, and construction oversight. These are real risks but manageable ones. Large data center projects routinely face delays in permitting and power delivery. The 2027 handover date leaves limited margin for such delays. What would be interesting from a regulatory design perspective: the interaction between the data center subsidiary's physical infrastructure and the parent's crypto financial operations in the event of stress. If Galaxy Helios II faces liquidity pressure, can the parent provide support without triggering cross-guarantee provisions? The structural separation is clean on paper, but the market will test it in the first stress scenario. THE 2026 AI+Crypto CONVERGENCE ANGLE The macro context for this deal: we are in the early phase of what I've been calling the AI-crypto convergence. Autonomous agents need settlement layers. AI-driven trading systems need cryptographically verifiable execution infrastructure. The efficiency gains from AI in market participation are contingent on blockchain's trustless execution layer. In 2026, I developed a prototype where AI-driven trading bots settled micro-transactions on a modular blockchain, reducing gas fees by 40% through batch processing. The lesson from that experiment: AI's gains in a financial system are only as strong as the settlement layer underneath. Galaxy is betting on a different corner of that convergence — not the settlement layer, not the execution layer, but the physical compute substrate. It is a real and growing demand signal. AI needs data centers. Data centers need financing. Financing needs predictable cash flows. The question is whether the physical layer can generate the same kind of economic efficiency gains as the digital layer. Physical construction doesn't benefit from Moore's Law. It benefits from slow, linear progress. And every large construction project runs the risk of compounding cost overruns. The convergence thesis is correct. The specific capital allocation may not be. AUDITING THE INVISIBLE HANDS OF MONETARY POLICY Zoom out and look at the macro liquidity map. We are in an environment where monetary policy expectations are driving capital flows into tangible infrastructure assets. Data centers are the new toll roads. The same institutional appetite that once bid up renewable energy assets and fiber optic networks is now chasing AI compute capacity. The scale of capital being raised is historic. CoreWeave's $20 billion. Galaxy's $3.5 billion. Multiple billions more across the mining sectors. This is a broad-based infrastructure buildout funded by debt markets that are still optimistic about AI's demand curve. From a macro observer's perspective, the risk is a synchronized repricing event. If AI capital deployment at the institutional level hits a confidence pause, the leveraged structures at the bottom of the chain — the Galaxys, the debt-financed data center owners — will suffer outsized losses. The bond market's willingness to fund these structures at 9.875% is a sign that the risk premium is being recognized, but it is not yet being fully discounted. CONTRA Let me now argue against my own analysis, because the contrarian angle here is more interesting than the obvious bear case. The conventional bull narrative: Galaxy has successfully transitioned from a pure crypto trading shop to a diversified financial services firm with a real asset arm. Phase I is operational. The tenant is a market leader. The 15-year lease provides unprecedented revenue visibility. The crypto business is cyclical, but the AI business is secular. Eventually, the project-level cash flows will cover the interest bill, the stock will re-rate, and the transformation will be vindicated. There is genuine substance to this. Real data centers are operating. Real rent is being paid. This is not a whitepaper story — it's a functioning business. But the contrarian thesis I want to press is different. It's not about whether Phase I works. It's about what the transformation has done to the risk profile of Galaxy as an investment. Galaxy has replaced a market risk (crypto price volatility) with a binary risk (construction, tenant credit, and AI demand). Market risk is continuous. You can hedge it, size it, and manage it. Binary risk is harder. A single missed milestone, a single tenant default, a single supply chain failure can produce an outsized, discontinuous loss. If bitcoin drops 20%, Galaxy loses money on its treasury book. It can also reduce exposure, adjust positions, and rebuild. Market risk is recoverable. If CoreWeave defaults, or if Phase II slips by a year, Galaxy has $3.5 billion in construction debt on an asset that's generating less than its interest bill. You cannot sell a data center fast. You cannot hedge a construction delay. You can only absorb the loss and hope the bond market is still open. The second contrarian point: the market's enthusiasm for the AI infrastructure asset class may itself be a top signal. When leverage becomes freely available for speculative infrastructure buildouts, it usually marks the late stage of the cycle. The 9.875% coupon is not just a cost of capital. It is a velocity gauge. It tells you how fast the money is flowing into AI capacity. The third: the idea that AI leases are "contractual revenue" oversimplifies. Contracts are only as good as the counterparty's ability to perform. CoreWeave's revenue is contractually backed by its own clients. But those clients are AI labs whose funding is not guaranteed. When the AI lab funding cycle cools — and it will, because every technology capital cycle cools — the rent transmission chain gets stressed. Navigating the storm with empirical precision means watching the data, not the narrative. The data says: the interest bill exceeds Phase I's EBITDA. The data says: one tenant, one project, one construction deadline. The data says: 9.875%, which is the market's own estimate of the risk. The contrarian thesis is not that Galaxy loses. It is that this transformation has made Galaxy a less predictable, more binary investment than the crypto trading company it replaced. And for investors, predictability is a form of safety that no growth narrative can substitute. TAKEAWAY The architecture of trust, stripped to its bones: Galaxy has staked its balance sheet on a data center it doesn't operate, serving a tenant that serves AI labs it cannot control, financed at 9.875%, with a fixed-cost cliff in 2027 and an interest bill that exceeds Phase I's projected EBITDA by roughly $58 million per year. Q3 is the first public test. Watch the lease revenue. Watch the delivery milestones. Watch the interest coverage ratio. Clarity emerges from the chaos of verification. The numbers will tell the story long before the narrative does.