The Hook: A Save That Isn't
The Q2 numbers hit like a fist in the chest for anyone still clinging to the merchant-bank fairy tale. Galaxy Digital — the swaggering crypto financial services firm that Mike Novogratz built into Wall Street's favorite bridge to digital assets — just reported an $85 million net loss, dragged down by a crypto operation that refuses to stop bleeding in this bear market. Right beside that wound, management flashed its salvation narrative: roughly $80 million per quarter in projected AI data center revenue. Clean symmetry. $85 million out the door on crypto; $80 million coming in from artificial intelligence. The crowd reads that as a save.
The chart reads it as a trap.
Because that $80 million is top-line revenue, not profit. And the AI machine that's supposed to produce it is financed with $3.507 billion in senior secured notes carrying a 9.875% coupon — an annual interest bill of roughly $346 million. Against an annualized top line of $320 million. Do the math once and you'll feel it: even at project-level EBITDA margins north of 90%, there's a hole in the floor. About $60 million wide. Every year. Smile while the liquidity drains.
I've spent more than two decades watching market structure bend under stress, and I can tell you exactly what that gap means. It means no room for error. No construction delay that goes unpunished. No tenant that pays late. The market's excitement about the AI pivot is real, but the balance sheet is the referee — and right now, the referee is calling a foul on the home team.
This is a story about how a crypto company became a landlord for the AI gold rush, how it financed that transformation with the most expensive secured debt on the block, and why the redemption arc the headline writers are stitching together doesn't survive contact with the actual numbers. The chart lies. The crowd feels. And what the crowd is feeling right now is hope — hope that hasn't been reconciled with a 9.875% coupon, a 2027 delivery clock, and a single tenant that holds the entire revenue stream in one hand.
Context: The Company That Forgot Which Face It's Wearing
Let's establish what Galaxy Digital actually is, because the category error at the heart of most coverage of this company starts there.
Galaxy Digital has no token, no protocol, no DAO, no on-chain governance. Adjust your mental frame accordingly. This is a public company — listed in Toronto, reporting through the SEC's disclosure machinery — with audited financials, a board of directors, and the entire weight of traditional corporate governance pressing down on its quarterly reporting. For the better part of a decade, its pitch was elegant in its simplicity: we are the institutional-grade crypto financial services firm. Trading. Custody. Asset management. Investment banking for blockchain-native companies. The bridge, the gateway, the grown-up in the room.
Then came 2022. Terra/Luna collapsed in a algorithmic stablecoin death spiral. Three Arrows Capital vaporized. FTX turned out to be a house of cards with a Bahamian beach house. The regulatory machinery in Washington went from benign neglect to active enforcement. And the bear market that started all of it refused to die, dragging on through 2023, 2024, and into a 2025 that felt like a slow-motion car crash for institutional crypto trading volumes.
The lending books that once generated fat, easy revenue became landmines. The trading desks that profited from volatility found themselves staring at dead spreads. And Galaxy — like so many of its peers — discovered that being the grown-up in a shrinking room doesn't pay the bills.
So it pivoted. Or, more accurately, it started building a second company underneath the first one.
The second company is an AI data center landlord. Through a project subsidiary called Galaxy Helios II LLC, the firm has been acquiring land, securing power rights, and constructing high-density data centers sized for the machine-learning boom. Not for itself. For a tenant: CoreWeave, the aggressively expanding AI cloud provider that has become one of the most important customers of NVIDIA's GPUs and one of the most voracious capital raisers in the entire technology ecosystem.
The structure is Project Finance 101 with a crypto flourish. The data center subsidiary issues secured notes. The notes fund construction. CoreWeave signs a 15-year lease. The lease backs the debt. The revenue stream is contracted, visible, and — in theory — predictable. This model has attracted tens of billions of dollars of institutional capital into the AI infrastructure space, and Galaxy has positioned itself squarely in the middle of that flow.
Phase I is done. 133 megawatts of critical IT load fully operational, switched on, delivering power to CoreWeave's machines. That is a real milestone — not a CGI rendering, not a press release about groundbreaking. The megawatts are live. The lease is running. Management guided to roughly $80 million per quarter of revenue off that Phase I book, and project-level adjusted EBITDA margins above 90%. If those numbers hold, the AI infrastructure business is, on a narrow project basis, an absolute cash machine.
Phase II is the monster. 260 additional megawatts, financed by that $3.507 billion debt raise, targeted for delivery and handover in 2027. More than double the first phase, with all the complexity that entails — power procurement, cooling systems, construction sequencing, equipment availability, and the simple fact that large-scale infrastructure projects have a nasty habit of running late and over budget. Based on my audit experience with mining and data center operators, I can tell you that the difference between 133 MW and 393 MW total is not linear. It's exponential in terms of operational complexity.
The Q2 financials show just how early this transformation sits. The AI data center segment generated only $20 million in adjusted gross profit and $11 million in adjusted EBITDA. That's the entire AI empire, as of last quarter: $11 million of project-level earnings against a $3.5 billion commitment. The crypto trading and Treasury segment, meanwhile, posted a $42 million adjusted gross loss. The company as a whole lost $85 million.
Let me be direct about what that means, because the crowd is already inventing a redemption arc and the chart hasn't caught up: Galaxy Digital is a story stock. A story stock with real infrastructure attached — I'll grant it that. But the storyline — "AI saves Galaxy" — is running about three years ahead of the cash flows, and the balance sheet is the referee.
Core: The Debt Math Nobody Wants to Do
This is the part my market surveillance training makes me obsessive about. The capital stack. The interest coverage. The gap between narrative and arithmetic. The headlines all grabbed the symmetry — $85 million lost, $80 million coming — but none of them did the division. Let's do it together.
Start with the debt. $3.507 billion in senior secured notes. Coupon: 9.875%. Maturity: 2031. Secured by project assets and pledged equity of Galaxy Helios II LLC. Notice what's missing: a parent guarantee. The debt sits at the project subsidiary, which means creditors have claims on the data center assets, not on Galaxy Digital's broader corporate assets. That's a deliberate risk-isolation structure. It protects the parent's crypto operations from a data center implosion, but it also means the data center has to stand alone financially. No parent lifeline. No balance-sheet bailout from the trading desk.
Annual interest on that note: $346 million. Roughly. Year in, year out, until 2031. The coupon doesn't care about crypto winter or AI hype cycles. It capitalizes during construction? Maybe — some of the interest may be capitalized rather than paid in cash, which shifts timing but increases the principal burden. The filings don't give us the full picture of how much interest is being capitalized versus paid in cash. But the obligation exists, and it compounds.
Now the revenue side. The $80 million per quarter run-rate from Phase I. Annualized: $320 million. And here we need to slow down and use the correct lens. The $80 million per quarter is top-line lease revenue. Not profit. The company, to its credit, has been transparent that the project-level adjusted EBITDA margin is above 90% — but that's a management guidance figure, not a reported EBITDA number.
Even accepting the 90%-plus margin at face value, we arrive at roughly $288 million in annual project-level EBITDA. Against $346 million in annual interest expense. The difference: about $58 million to $60 million, annually. Roughly 20% of EBITDA, negative.
In plain language: the AI project, as currently configured, doesn't generate enough cash to pay its own interest bill. The shortfall must be covered by something else — new financing, drawing down whatever liquidity reserves exist, favorable interest capitalization terms that push payments further into the future, or Phase II revenue arriving faster than its costs. But Phase II revenue cannot arrive before 2027, by the company's own timeline. The construction clock runs. The interest clock runs faster.
Now, before the pedants sharpen their knives, let me acknowledge the caveats. This is a conservative stress test, not a precise forecast. There are levers I am not modeling. Interest capitalization during the build-out phase could push cash interest payments further out and smooth the near-term crunch. Phase II revenue, if and when it hits, changes the denominator dramatically. There may be other liquidity sources sitting in the parent company's treasury, and the crypto business might — might — stage a recovery that contributes cash rather than consuming it.
But the point stands, and it's worth putting in bold: a business that generates $320 million in top-line revenue against $346 million in annual interest expense is a business living on the kindness of lenders and the hope of future deliveries. That's not a rescue narrative. That's a high-wire act with a $60 million deficit in the funding gap.
And here's the other elision hiding in the 90% margin figure. The project-level adjusted EBITDA margin does not translate to company-level profitability. That margin excludes management overhead, corporate expenses, and the losses from other segments. The crypto Treasury segment, remember, lost $42 million at the adjusted gross level in Q2. The company's total net loss was $85 million. The 90% margin refers to a narrow project-level slice of the pie, and presenting it as anything more is the kind of accounting obfuscation that builds bull traps.
I want to add a pressure-test scenario, because this is what market surveillance actually does — you don't just look at the base case; you look at how a structure behaves when the market moves against you.
Scenario one: Phase II slips nine months. Construction delays in data centers are not exceptions; they are the rule. Grid interconnection queues, transformer shortages, cooling equipment lead times — the physical world is unkind to ambitious timelines. If Phase II moves toward late 2027 or 2028, the project absorbs fixed cash costs with no matching revenue for longer. The carry cost widens. And the phrase from the filing — that the project company will assume fixed cash costs starting in 2027 — turns into a guillotine. Fixed costs don't wait for the grid to energize. They don't wait for CoreWeave to take delivery. They show up on schedule, every quarter, like a debt collector with a key.
Scenario two: CoreWeave stumbles. Its own balance sheet is levered to the hilt, its customer base is concentrated among a handful of AI labs with their own funding dependencies, and its market narrative requires continuous capital infusion to sustain growth. The credit quality of the tenant is the credit quality of the rent. A 15-year lease is only as strong as the tenant's ability to honor it.
Scenario three: the AI bubble's air starts leaking. This is the VanEck scenario, and it deserves a whole section of its own.
The 9.875% Signal: The Loudest Number in the Room
Let me put that coupon in context, because it's the most underappreciated data point in this entire story.
Investment-grade corporate issuers today borrow at 4% to 6%. High-yield credits of reasonable quality might pay 7% or 8%. A 9.875% coupon on a secured note is the kind of rate you'd expect from a distressed company, or from a project that the market believes has a nontrivial probability of default. The bond market is not in the business of sentiment — it prices risk with cold precision. The fact that Galaxy could sell $3.507 billion of these notes at all is a testament to the demand for AI infrastructure exposure. The fact that it had to pay nearly double digits to get there is a testament to what the lenders actually think of the risk.
The 9.875% coupon is simultaneously the strongest evidence of institutional demand for AI data centers and the clearest warning that the market views Galaxy's specific execution risk as severe. Those two statements are both true. The market wants the asset class, but it's demanding a steep risk premium to hold this particular issuer's paper.
Think about what that means for the equity. If the debt costs nearly 10%, the equity returns required to make this project value-accretive are astronomical. The project has to clear that interest hurdle before a single dollar of shareholder value is created. And given the current funding gap, the equity isn't just not earning returns — it's watching the debt stack consume the project's entire cash generation.
The bond market is also signaling something about the timeline. A 2031 maturity with a 2027 fixed-cost trigger means the company faces a refinancing test before the notes come due. If Phase II is late, if the AI demand environment cools, if interest rates stay elevated, the refinancing that Galaxy will likely need — to cover the gap, to fund further build-out, to smooth the cash flow cliff — will come at terms that make 9.875% look generous.
CoreWeave, the Tenant, and the $20 Billion Elephant
Now let's talk about the tenant. This is where the narrative gets interesting, because CoreWeave is not a sleepy utility. It's an aggressive, debt-fueled AI cloud provider that has become one of the most important customers of NVIDIA's GPUs and one of the most aggressive capital raisers in the entire tech ecosystem. It just raised $20 billion, and that headline matters on two levels.
First, it's a validation. CoreWeave has the market's attention and the capital to keep pursuing its own growth. That's good news for Galaxy's rent check. A well-funded tenant is a tenant that can honor a 15-year lease. The $20 billion raise de-risks, at least temporarily, the single most important assumption in Galaxy's AI business model.
Second, and this is the part most analysts are sleeping on: CoreWeave's $20 billion raise is happening at the exact moment capital is flowing out of bitcoin mining and crypto infrastructure. The AI narrative isn't just lifting Galaxy's data center business — it's actively cannibalizing crypto's capital base. Every dollar that flows into CoreWeave's AI war chest is a dollar that might otherwise have gone to a bitcoin miner, a crypto exchange, a DeFi protocol, a layer-2 team. That's not a side effect. That's the story.
We've seen this migration in real time across the mining sector. Riot Platforms, IREN, Cipher Mining — the pure-play bitcoin miners are all rebranding themselves as AI/HPC data center operators, chasing the same CoreWeave-style lease deals, the same institutional capital, the same NVIDIA allocation. The market is rewarding them with premium valuations before most of their AI capacity is even delivered. Wall Street is literally paying for infrastructure that hasn't been built yet. And Galaxy — a crypto financial services firm — has shoved itself into the middle of that land grab, financed by 9.875% debt, betting on enormous execution.
The single-tenant concentration deserves emphasis, because it's the kind of risk that looks small in a bull market and fatal in a correction. Galaxy's quarterly filing explicitly acknowledged that the data center segment depends heavily on this one AI infrastructure customer in its early phases. That's not a hedge; that's an admission. Fifteen years is a long lease, and CoreWeave is a formidable company, but "formidable" is not "bulletproof." Credit risk is the slow-moving threat in every landlord business. It doesn't announce itself. It accumulates silently in quarterly reports, in working capital lines, in refinancing terms.
Here's the darker version of the same story: if the AI build-out frenzy cools — if the funding environment tightens, if the model-training economics disappoint, if the power costs escalate — CoreWeave's own margin compresses, and the pressure ratchets down through the entire chain. The tenant's struggle becomes the landlord's struggle. In a 15-year lease, there's no escape hatch. Just a long, slow grind.
VanEck's Warning and the Premium Problem
VanEck — one of the more thoughtful voices in the digital asset management world — issued a warning that should be taped to every trader's monitor: AI-linked miners are receiving premium valuations before most of their leased capacity has even been delivered. The market is paying for deliveries that haven't happened. Galactic-scale premiums for phantom megawatts.
That warning applies to Galaxy with almost uncomfortable precision. Galaxy's Phase I is real — I'll say it again, 133 megawatts of delivered, operational capacity — but the stock's AI premium is being driven largely by Phase II, by the 260 megawatts that won't produce revenue until 2027, by the narrative that a crypto company can become a core AI infrastructure provider. VanEck's point is that the market has historically been generous to AI conversion stories before evidence arrives, and brutal when the evidence disappoints.
We saw the same pattern in the crypto mining sector in 2023 and 2024. Every miner that announced an AI partnership juiced its stock. Then the deliveries, the revenues, the actual operating results started trickling in — and the differentiation began. The companies that executed on their AI leases compounded their premiums. The ones that didn't — the ones that announced, raised, and stalled — got repriced downward with the same speed that had carried them up. The market's memory for unmet delivery promises is short, but its reaction to them is violent.
The question for Galaxy is not whether AI data centers are a good business. They are — the long-term demand for compute is real, and the supply constraints are genuine. The question is whether this specific company can execute at the scale required, with the debt load carried, on the timeline promised. The bond market's 9.875% coupon is the market's honest answer to that question. The equity market's enthusiasm is the crowd's emotional answer. They are currently in disagreement. That disagreement is where the risk lives.
Contrarian: The Story Everyone's Missing
Here's what the consensus narrative gets backwards. The mainstream telling is simple and soothing: "Galaxy lost money on crypto, but its AI pivot will save it." Flip the frame and the picture darkens. The AI pivot isn't saving Galaxy from crypto — crypto is subsidizing Galaxy's AI ambitions, and the real risk isn't that the AI project fails, it's that the AI project succeeds just enough to keep the debt machine turning while the crypto business slowly starves.
Think about the capital allocation logic. Galaxy is a crypto financial services company raising $3.5 billion in the debt markets to build data centers for an AI cloud provider. The interest burden alone exceeds the current revenue of the AI project. The crypto operations are losing money. The company has effectively placed a massive bet that AI infrastructure will generate superior returns to its core business — a bet funded by borrowing at nearly 10% during a period of high interest rates. That's not a hedge. That's a leveraged reallocation of the entire firm's future.
The symmetry the headlines are built on — $85 million lost versus $80 million earned — is a mirage, and chasing it is how you end up long a broken balance sheet. The $85 million is a net loss: the bottom line of the entire company after all expenses, all impairments, all segments. The $80 million is a top-line revenue projection for one segment that hasn't even been reported yet — management's guide for Q3, not Q3's actual print. The actual reported Q2 number from the AI segment was $20 million in adjusted gross profit and $11 million in adjusted EBITDA. There is no mathematical operation that makes those figures commensurable. Yet the market is being sold "offset" as if it were arithmetic.
This is where my old ICO-sprinter instincts kick in. Back in 2017, I was one of the first people to write about EtherDelta before the DEX narrative exploded. I learned then that markets reward the first mover on narrative, not the most rigorous analyst. But I also learned that narratives have a half-life. The crowd that felt "DEX volume will explode" didn't wait for orderbook DEXs to get front-run into oblivion by centralized matching engines. The same principle applies here: the crowd feels "AI saves Galaxy" today, but the market will eventually weigh the latency between the story and the delivery schedule. Latency is everything, in trading and in infrastructure. You can't front-run a construction schedule.
There's another uncomfortable parallel that fits the moment. You've heard me say this before, but it bears repeating in this context: there are dozens of layer-2s now, all claiming to scale Ethereum, all actually slicing an already-scarce liquidity pool into smaller fragments. That isn't scaling; it's fragmentation. The AI infrastructure build-out is doing something similar to crypto's capital. Every dollar Galaxy raises for AI is a dollar pulled out of crypto market-making, crypto lending, crypto venture. The escape narrative hides this. The data doesn't.
Galaxy, whether it intends to or not, is a participant in a capital rotation that's leaving crypto thinner. When the market's most sophisticated crypto financial services firm is borrowing billions to build for an AI tenant, that's a signal about where the smart money thinks the future lies. It's also a self-fulfilling prophecy: the more capital leaves crypto for AI, the thinner crypto's liquidity becomes, the more reason the smart money has to leave. The chart lies. The crowd feels. And what the crowd feels right now is FOMO about AI while the underlying asset class it claims to serve is quietly losing its institutional depth.
Let me also flag a governance angle that the mainstream coverage misses. By issuing debt at the project subsidiary level rather than the parent, Galaxy has structurally isolated the AI project from its crypto business. That's good risk management in one sense — a data center failure won't drag the trading desk down. But it also means the parent company has limited ability to redirect the AI project's assets toward its crypto operations if the crypto business needs support. The structural isolation cuts both ways. The parent can't bleed the project to save the trading floor, but the project can also fail without the parent being able to quickly extract value. There's no organic synergy between a crypto financial services firm and a physical data center landlord. The only connection is the balance sheet — and the balance sheet is stretched.
The Regulatory Quiet: Boring but Telling
On the compliance front, this story is refreshingly dull. There's no token launch, no Howey test anxiety, no DAO governance weirdness, no anonymous founder with a Dubai mansion. Galaxy Digital is an SEC-reporting company with regulated subsidiaries — CFTC registrations, FINRA oversight, the whole institutional apparatus. The $3.507 billion note issuance is textbook corporate debt. The security structure is standard project finance. The disclosure requirements are being met, quarter after quarter, in the ordinary course of business.
But even the boredom is data. The fact that a fully compliant, SEC-reporting, CFTC-regulated institution has to pay 9.875% on secured debt tells you the market doesn't care about compliance boxes when it's pricing delivery risk and execution risk. The compliance box is checked. The execution box is not. This is the quiet irony of the AI gold rush: the most boring regulatory profile in crypto is attached to one of the most aggressive balance sheets.
The financing choice is also revealing. Galaxy chose debt over equity — debt at nearly 10% — rather than diluting existing shareholders. That's a signal. It suggests management believes the equity market wouldn't value the AI project at a price they consider fair, or that they're unwilling to accept the dilution that a $3.5 billion equity raise would impose on current holders. Debt preserves ownership. But it also imposes a rigid, unforgiving obligation. In a project with this much execution risk, the choice of debt over equity is the board saying: "We believe in this, and we'll pay a premium to keep the upside." The bond market's answer: "Fine, but we'll charge you for the privilege."
There's also the question of what happens if the project needs more capital before Phase II delivers. Based on comparable build costs in the industry, a 260 MW expansion could plausibly exceed the $3.507 billion already raised. That's a medium-confidence inference, not a documented fact, but it keeps me up at night as an analyst. If Phase II requires additional fundraising while the 9.875% notes are still outstanding and unharmed, the new capital will come at terms that reflect the same risk — or worse. The debt stack gets taller. The fixed costs get heavier. The equity gets thinner.
The Three Gates: What to Watch
So where does that leave us? Watch three gates, and watch them carefully.
Gate one: Q3. The first real test of the "$80 million per quarter" run-rate guidance. If the actual Q3 print comes in light — and remember, Q2's actual AI segment numbers were a fraction of the guide — the symmetry narrative dies in real time. I will be listening to the earnings call for any softening of the run-rate language, any hedge about "timing," any reference to "ramp-up" that sounds like a delay dressed in a suit. The difference between $80 million and $50 million in reported run-rate revenue is the difference between a credible pivot and a paper castle.
Gate two: CoreWeave's health. Watch its funding events, its customer announcements, its own quarterly disclosures. CoreWeave's $20 billion raise was a strong signal, but it was also a demand for more oxygen. The company needs to keep converting its lease pipeline into contracted, revenue-generating capacity to justify its own valuation. Any wobble in CoreWeave's balance sheet, any customer concentration revealed in its filings, any delay in its own delivery schedules — all of it is a direct threat to Galaxy's rent.
Gate three: Phase II execution. Every construction update, every financing announcement, every grid-connection notice, every equipment procurement milestone. The 2027 fixed-cost clock is the countdown timer on this entire thesis. If Phase II delivers on time and on budget, the additional 260 MW could transform the revenue profile and close the interest coverage gap. If it slips, the gap widens, the refinancing pressure builds, and the 9.875% coupon starts looking like the cheap part of the deal.
Takeaway: Smile, Then Watch the Delivery Dates
The chart lies. The crowd feels. And right now, the crowd is feeling a rescue narrative around Galaxy Digital while the chart shows a leveraged project that cannot yet pay its own interest bill. The real story is not "AI saves crypto company." The real story is "capital is fleeing crypto for AI, and Galaxy is both the escapee and the escape hatch."
The Q2 numbers deserve a cold, unblinking read. $85 million lost on crypto. $11 million of AI adjusted EBITDA. $3.5 billion of 9.875% debt. A $60 million annual interest coverage gap. A single tenant. A 2027 delivery clock. Those facts don't add up to a redemption story. They add up to a stress test — the most important stress test in Galaxy's corporate life.
Smile while the liquidity drains. Then watch the delivery dates. Because in a market this unforgiving, the only thing that matters is whether the machines turn on, whether the tenant pays, and whether the interest bill gets covered by something other than hope. The crowd will keep feeling the AI story. I'll keep reading the filings. And when the Q3 numbers land, we'll all find out which one of us was reading the right ledger.
