There's a particular silence that settles over markets when a story's arithmetic stops adding up. Not the loud silence of a crash, or the anxious hum of a sell-off, but the quieter moment between cycles when a headline like "Galaxy Digital lost $85 million on crypto as its projected $80 million in AI revenue must offset $3.5 billion AI investment" gets shared, retweeted, and filed away without anyone running the numbers.
I've spent the better part of 13 years listening for these moments โ the intervals where narratives begin to crack under their own weight. This one carries particular resonance, because it involves a company that built its name on crypto's volatility now mortgaging its balance sheet to the most capital-intensive promise of the AI era. The symmetry in that headline โ $85 million lost, $80 million earned โ is dangerously seductive. It's also an accounting illusion that evaporates the moment you look at what those two numbers actually measure.
Galaxy Digital has always occupied an unusual seat in the digital asset ecosystem. It's not a protocol, not a miner, not an exchange. It's a financial services firm that trades, manages, and warehouses crypto for institutions. That identity gave it a front-row seat to every major cycle: the 2017 ICO mania, the DeFi Summer of 2020, the 2022 contagion that took down FTX and a dozen lesser names, and the 2024 spot ETF approval that finally pulled Wall Street's liquidity into Bitcoin. Mike Novogratz, its founder, has been perpetually bullish regardless of weather. But the second quarter of 2025 introduced a different kind of narrative โ one where the crypto business lost money and the company's future increasingly depends on a sector that hasn't historically cared about blockchain at all.
The core facts are straightforward. Galaxy's Phase I data center project โ a 133-megawatt facility leased to CoreWeave, the AI cloud provider that has become the poster child for GPU-backed debt โ is fully operational and generating contracted revenue under a 15-year lease. Management guides to roughly $80 million in quarterly revenue from this agreement, or about $320 million annualized. Project-level adjusted EBITDA margin, again per management guidance, exceeds 90%. To fund Phase II, a 260-megawatt expansion currently under construction, Galaxy has raised $3.507 billion in senior secured notes carrying a 9.875% coupon, maturing in 2031. The project company will begin assuming fixed cash costs when Phase II starts its phased delivery in 2027, regardless of whether the associated revenue has come online.
In the broader landscape of crypto-adjacent companies pivoting toward compute, Galaxy stands partially ahead. Riot Platforms, IREN, and Cipher Mining are all pursuing AI/HPC transitions at various stages, but most remain in construction or pilot phases. Galaxy has delivered its first tranche and is collecting rent. That is not nothing. But the difference between delivering 133 megawatts and delivering 393 megawatts is not linear. Large-scale data center construction is a discipline where complexity compounds: power procurement, liquid cooling systems, chip availability, and construction timelines all interact in ways that defy neat scheduling. Phase I succeeded. Phase II is a different animal entirely.
Now let me do the arithmetic the headline never gets around to. Annualized lease revenue of approximately $320 million, at a 90%-plus EBITDA margin, yields roughly $288 million in project-level EBITDA. Interest expense on $3.507 billion at 9.875% comes to approximately $346 million per year. That's a gap of roughly $60 million annually โ about 22% of the project's EBITDA โ that must be covered by additional liquidity, interest capitalization, or Phase II revenues that haven't been confirmed yet. This is not a forecast of insolvency. It's a pressure test, the kind I learned to run during the summer of 2017, when I manually audited 15 early-stage ICO smart contracts for a Seattle crypto meetup. The most dangerous vulnerabilities I found then weren't in the code. They were in the assumptions embedded in the token models โ the way projects counted future users as current revenue, the way they treated vesting schedules as solved problems. Galaxy's situation isn't a code problem, but it has the same flavor: the timely delivery of Phase II, the ongoing solvency of CoreWeave, and crypto's recovery are all being treated as certainties rather than probabilities.
Let me be precise about what the $85 million loss and the $80 million revenue projection are and aren't. The loss is a net loss for the quarter, dragged down by a $42 million adjusted gross loss in the treasury segment and substantial expenses elsewhere. The revenue projection is top-line, unadjusted, lease-based income. Comparing them is like comparing the depth of a river to the height of a waterfall โ related but fundamentally different measurements. The Q2 filing shows the AI segment contributing $20 million in adjusted gross profit and $11 million in adjusted EBITDA. That's positive, real cash generation. But it's nowhere near sufficient to cover a $346 million annual interest bill, let alone offset the ongoing drag from crypto market volatility. The accounting truth is that "80 million" is not "80 million." One figure is a ceiling, the other is a floor, and between them sits a company stretched across two of the most capital-hungry industries on Earth.
The 9.875% coupon is where I want investors to focus, because it is the market's most honest assessment of this entire venture. Investment-grade corporate debt trades at 4% to 6% right now. A 9.875% yield on a secured note with a 2031 maturity signals that bond buyers are demanding substantial compensation for real risk. That risk isn't the crypto business โ the notes are secured by project assets, isolating them somewhat from Galaxy's crypto operations through the Galaxy Helios II LLC structure. The risk is execution: building 260 megawatts of high-density liquid-cooled data center capacity, delivering it on a timeline, and doing so while CoreWeave remains the sole tenant. When I studied the 2024 ETF inflows and their correlation with crypto volatility, I watched institutional capital pour into Bitcoin while retail absorbed the turbulence. The same dynamic is now playing out in AI infrastructure โ except this time the institutions aren't buying a liquid asset. They're buying a promise of future compute delivery, secured by land, a lease, and a debt structure that requires flawless execution to remain solvent.
From a regulatory perspective, this is the most traditional financing imaginable. These are SEC-compliant securities, issued by a public company, backed by project assets. There's no Howey analysis needed, no token classification debate, no novel legal ground. It's a bond. But that's precisely what makes the 9.875% coupon so striking: this is mainstream finance looking at a crypto-founded company's AI venture and pricing in substantial default risk. The market is not treating Galaxy's transformation as a sure thing.
What makes this a governance story as much as a financial one is the choice to finance through debt rather than equity. Issuing $3.507 billion in secured notes at a project-company level allows Galaxy to avoid diluting existing shareholders. But it also signals that the equity market's appetite for underwriting AI infrastructure at this scale may be limited. The bond market, in contrast, was willing to lend โ at a coupon that reflects genuine uncertainty about the project's outcome. This is the kind of capital structure I spent my doctoral research analyzing: the way risk gets repackaged and redistributed across different claim-holders. Equity holders bear the downside of execution failure; bond holders bear the downside of bankruptcy; and the tenant, CoreWeave, bears the operational risk of its own aggressive expansion. Everyone is exposed, but nobody is exposed in the same way. That's the design. It's also the vulnerability.
There's a deeper irony that the cheerleaders of the "AI + crypto convergence" narrative have missed. CoreWeave just raised $20 billion. That money is being absorbed by AI infrastructure at a scale that makes most crypto treasury operations look like pocket change. The capital isn't flowing into crypto alongside AI โ it's flowing into AI instead of crypto. VanEck has flagged this explicitly, noting that AI-linked miners are receiving premium valuations before the majority of their leased capacity is even delivered. Wall Street is paying for undelivered megawatts. In my 2020 work mapping liquidity flows across Uniswap and Aave during DeFi Summer, I watched projects celebrate TVL figures that evaporated the moment incentives stopped. The parallel is uncomfortable but direct: capital that chases a narrative without verifying the underlying economics produces the same outcome, whether the underlying asset is a yield farm or a data center. The difference is the scale of the burn. DeFi Summer's excesses were measured in hundreds of millions. AI infrastructure's potential excesses are measured in tens of billions.
I keep returning to the human dimension, because I've learned that markets are ultimately made of people making decisions under uncertainty. During the 2022 bear market, I hosted twelve webinars on custody, verification, and the psychological toll of watching 80% of your portfolio evaporate. The best-prepared investors weren't the ones with the most sophisticated models. They were the ones who understood the difference between a narrative and a balance sheet. Galaxy's story is compelling โ a crypto firm that saw the future of compute and positioned itself as the landlord of the AI boom. But the narrative and the balance sheet are currently misaligned. The income statement says crypto is still bleeding. The debt schedule says AI isn't generating enough to cover its own cost of capital. The tenant concentration says one bad quarter at CoreWeave becomes an existential problem. None of these are death sentences. They are, however, structural risks that compound quietly until the market decides to price them.
The contrarian angle cuts against the entire "AI saves crypto" thesis that has animated so much of the current bull market. The dominant story is that crypto companies pivoting to AI will be rewarded with a second growth curve. The data suggests something more uncomfortable: AI is competing with crypto for the same institutional capital, the same electricity, the same builder talent, and the same attention. CoreWeave's $20 billion raise is not evidence of convergence. It is evidence of divergence โ a reallocation of resources away from digital assets and toward the physical infrastructure of machine intelligence. Galaxy sits at the intersection, yes. But an intersection is also where you get hit by traffic coming from both directions. The market has already priced 50% to 70% of the AI narrative into Galaxy's stock. The remaining 30% must be earned through actual cash flow โ starting with Q3, the first full quarter at the guided run rate.
The third quarter of 2025 will be the first true test of the $80 million revenue guidance. If numbers come in range, the story holds, and focus shifts to Phase II's 2027 delivery timeline. If they miss, the premium valuation evaporates quickly, because the debt market's 9.875% coupon already contains a significant margin of doubt. I'd advise investors to stop asking whether Galaxy's AI pivot is real. It is real. The megawatts are operational. The lease is signed. The right question is whether the mathematics of leverage and delivery leave any room for error. In my experience, listening to the silence between market cycles, the structures that survive are not the ones with the grandest narratives. They're the ones with the fewest hidden obligations. Galaxy's obligations are no longer hidden. They're printed on a bond coupon at 9.875%, maturing in 2031. The market will read those numbers every quarter. So should you. So should every investor who believes that the intersection of crypto and compute is a safe place to stand โ because at this intersection, the lights are flashing in both directions, and the only thing certain is the silence before the next report lands.


