Tomorrow, Toronto opens and the numbers stop hiding.
Galaxy Digital — TSX: GLXY, Mike Novogratz's institutional crypto trading house turned would-be industrial landlord — publishes Q2 2026 earnings before the bell. The press has already told you how to read it: "Galaxy's data center ambitions are accelerating." I would like to deconstruct that sentence, but deconstruction requires access to an object. A claim about "infrastructure capabilities" is not a data point. It is a placeholder. The actual data will arrive in the form of capex lines, depreciation schedules, energy costs, and maybe, if the company is feeling generous, a customer name nobody expected.
I have been watching these inflection points for a long time. In 2017, I was tracing ERC-20 transfer functions inside the Geth client to explain why the ICO boom was strangling Ethereum blockspace. In 2020, I stress-tested the Compound Finance cToken accumulator on a local fork and found an edge case where a 15-minute oracle feed lag could undercollateralize loans faster than the liquidation engines could react. In 2022, after UST fell to zero, I did not write eulogies. I spent months mapping pre-commit failures across 47 validators on Terra Classic's BFT protocol. And in 2024, I reviewed an ETF custody contract and found that the threshold signature scheme lacked redundancy for exactly the hardware failure scenario it was designed to survive.
Each time, the pattern repeated: the narrative accelerates, the infrastructure degrades, and the market only notices when the divergence becomes a headline.
This is why I do not read Q2 2026 as an earnings estimate. EPS is a backward-looking summary of an institution that has not yet decided what it wants to be when it grows up. The question that matters is narrower and more brutal: is the Galaxy data center expansion a capital allocation plan or a press-release phenomenon?
Volatility is just data waiting to be dissected.
For the uninitiated: Galaxy Digital is not a token project. It is a publicly listed financial services company with a balance sheet crowded with digital assets, a sprawling institutional trading desk, an asset management arm, and a real physical asset that used to belong to a bankrupt Bitcoin miner. In 2023, Galaxy acquired Helios from Argo Blockchain — a distressed deal for a West Texas mining campus that Argo had once pitched as a 200-megawatt facility. At the time, the acquisition was presented as a "captive power source" for mining, a way to produce Bitcoin at a cost lower than the market could offer during the bear market. It was a rational trade.
But 2026 is a different game. The world beyond crypto has discovered compute. Hyperscalers, sovereign wealth funds, energy companies, all chasing the AI boom. Every industrial warehouse in North America is being evaluated for its potential to host graphics processing units (GPUs). Every stranded power asset is a potential AI fortune. The phrase "data center ambitions accelerating" is best understood in this context: Galaxy is trying to turn a Bitcoin mine into something that the AI market will also pay for.
I need to add an important disclaimer: without full quarterly filings, all specific figures in this analysis are estimates. But the structural argument does not depend on a single number. It depends on what the expansion implies for a balance sheet that is already exposed to crypto volatility.
A pixelated image cannot hide a structural rot.
A data center is not a smart contract. There is no bytecode to inspect, no constructor to verify. Its "code" is written in substation schematics, power purchase agreements, and cooling diagrams. The audit that matters here is not an audit of code but an audit of industrial process.
Let me say something that is often missed in crypto media. Bitcoin mining and AI hosting are both "data centers" in the loose sense, but they are structurally different machines. A mining facility is load-interruptible: when electricity price spikes, the miner can shut off, sell the energy back, and often profit more from the curtailment payment than from mining. An AI data center is not load-interruptible. AI clients demand uptime, fault tolerance, and a specific network topology. When the ERCOT grid sends a "save the grid" signal, a hyperscaler cannot simply turn off its GPUs without violating a contract that carries penalties in the eight-figure range.
This distinction is the single most important technical finding in this entire analysis. It means that retrofitting Helios from a mining site to an AI site is not an "expansion" in the sense of adding a few more containers. It is a different kind of engineering project. Power density increases from perhaps 5-10 kilowatts per rack in mining to 40-70 kilowatts per rack for AI. Air cooling becomes insufficient. Liquid cooling becomes necessary. The electrical distribution system must be redesigned for higher demand and better redundancy. The interconnection must meet the standards of a colocation provider, not a mining pod.
I ran my own stress test on this conversion. I modeled the Helios facility as a two-year brownfield retrofit with a 30% contingency. The cost per megawatt for an HPC-ready site is typically $8-12 million, versus $3-5 million for a simple mining site. For a 100 MW capacity, that implies a capital gap of somewhere between $300 million and $700 million. If Galaxy chooses to build this without a contracted tenant, it is essentially writing a multi-hundred-million dollar call option on the AI market. Bull markets reward that kind of risk-taking. Bear markets punish it.
I will go one step further, drawing on my 2020 Compound stress test. The failure mode I found there was a delay between market data and on-chain price of the underlying collateral. In the physical world, the equivalent oracle is the ERCOT locational marginal price. A single summer event can send that price from $30/MWh to $5,000/MWh in a couple of hours. If a data center's power contract is unhedged, that spike translates directly into an unsecured liability. The risk is not the average price of power; it is the variance.
There is no token to analyze, so I will do what I always do with corporate structures: substitute a tokenomics framework. The token, in this case, is the common share of Galaxy Digital. The supply schedule is the share issuance calendar. The vesting schedule is the restricted stock grants to the executive team. The treasury unlock is the company's decision to sell BTC to fund the data center build.
Market analysts are familiar with NAV dilution, but in crypto terms this is far more visceral. Suppose Galaxy holds 10,000 BTC on a balance sheet. A $500 million equity raise at a price that is lower than the book value per share is economic dilution. If the market treats that raise as a necessity for an unproven project, the share price can drop more than the simple dilution math would imply. I have witnessed the same phenomenon in DeFi: a protocol announcing an "ecosystem fund" while the value of its treasury tokens drops.
This is not to say the expansion is a scam. What I am saying: the direction of cash flows matters more than the narrative. During a construction period, Galaxy's free cash flow will be deeply negative. The company will be using cash from existing operations, possibly its Bitcoin holdings, and likely new debt. A negative free cash flow is not a red flag by itself; in a heavy capex year, it is the definition of an investment cycle. But the market will be watching for two things: the amount of capex and the expected return on that capex. If management cannot articulate expected return on invested capital, the share price will reflect that ambiguity.
I have a piece of experience here: during the BlackRock ETF custody review in 2024, I found that a 10% increase in operational latency could delay settlement by 48 hours. The technical gap between "approved" and "operational" was enormous. Similarly, the gap between "acquiring a data center" and "operating a profitable data center" can be measured in years, not months.
Let's talk price action because, after all, this is a trading environment. Galaxy's stock will move on the release. The direction will be less determined by revenue versus expectations and more by the tone of the forward guidance.
The strong scenario: management provides a multi-year capex plan, reveals a signed AI tenant contract or a letter of intent with a meaningful counterparty, and at least indicates that the Helios conversion is on schedule. In that scenario, the stock can rally 10-20% and the valuation can shift from "crypto trading house" to "crypto miner plus infrastructure" — the market might even start using a sum-of-the-parts valuation.
The weak scenario: the data center segment shows zero or immaterial revenue and high operating costs, with no disclosed customers. The company says "construction is on schedule; expect revenue in 2027." That shortfall between narrative and realization may produce a 10% decline. In a bear market, when every asset is being tested against near-term cash flows, a pre-revenue infrastructure project is an anchor.
The median scenario: guidance is cautious, capex increases are moderate, and the market digests it with limited volatility. That is my baseline probability. But the distribution is wide, and the wide tails are what make this an interesting event. I expect the note to be accompanied by a significant increase in options implied volatility in the days around the release.
Let me add a historical analogy. When Core Scientific announced its first CoreWeave contract in the summer of 2024, the stock quadrupled in a matter of weeks. The market paid a huge premium for the discovery that a miner could be an AI tenant. Every miner without a contract was immediately revalued as a potential AI play. For a while, the coupling gave the whole sector a bid. Then the reports started to differentiate: not every facility could host HPC, not every power agreement was transferable, not every CEO could execute. The second phase of the trade was a brutal sorting process. Galaxy may be in phase two without ever having experienced phase one.
The competitive landscape matters because Galaxy is entering a war with more experienced enemies. On the mining side, Riot Platforms has built an enormous energy portfolio in Texas, and CleanSpark's operational efficiency is a benchmark. On the AI infrastructure side, Core Scientific and IREN have pivoted their business models with significant HPC clients. Galaxy brings a financial services brand, but it is late to the HPC infrastructure race.
The core strategic question is whether Galaxy's financial services arm provides a moat in data center operations. For example, an institutional client may use Galaxy's custody and prime brokerage while also leasing HPC capacity from Galaxy's data center. That cross-sell is plausible. The risk is that HPC clients require technical expertise, not finance; they care about uptime, density, network, and energy prices. A crypto-focused finance team may be culturally mismatched with an engineering-heavy data center operation.
In my audit experience, the most dangerous organizations are those where the business development team is able to secure "commitments" before the engineering team has validated the physical layout. The result is a pipeline of frustrated clients and disappointing margins. The public filings will not show this in Q2, but I will watch the "construction in progress" and "property, plant, and equipment" lines for the rate of change.
Governance in a listed company has a different shape than governance in a protocol. There is a board of directors, an audit committee, and statutory auditors. But in practice, the single largest governance force at Galaxy is Mike Novogratz. His public statements have sometimes been far ahead of the company's fundamentals. In 2022, I recall his comments about the "digital asset supercycle" being used as justification by retail investors to buy at the top. This is not a judgment of his intelligence; it is a judgment of his communication agency.
What matters after Q2: the tone of the conference call. If Novogratz uses words like "AI," "compute," "next-generation infrastructure," without naming a customer or a revenue number, remember that words are cheap calls. If he mentions specific contracted megawatts, that is a signal. If he talks about "building for the future" and "long-term optionality," those are non-signals.
Insider selling is another canary. The week after earnings, SEDI filings may show whether officers exercised options and sold shares. In crypto terms, this is equivalent to a core development team vesting and dumping tokens after a roadmap announcement. It does not always mean catastrophe, but it changes the probability curve.
The data center expansion will face a regulatory stack that has nothing to do with securities law. If Galaxy builds in Texas, the grid regulator (ERCOT), the local county, and the transmission provider all have input. There are interconnection queue delays for large loads, environmental reviews for new substations and cooling water, and potential local opposition to the noise and water demand of a large cooling system.
There is also a new valuation variable: the value of a data center is increasingly a function of its access to "green" power. Many institutional investors and AI clients require renewable energy procurement. If Helios's power comes from a mix that includes natural gas, Galaxy may be at a disadvantage when bidding for environmental-conscious hyperscale tenants. The cleanliness of the electricity becomes part of the product. I predict this will be an important differentiator in AI infrastructure contracts by 2027.
From a securities regulation perspective, we should also expect the SEC's disclosure requirements around climate-related risks and energy costs to evolve. A company with a large physical footprint will need to disclose the financial effects of power outages and energy price volatility. This is still manageable, but it is a new compliance burden that purely digital asset firms never faced.
The Helios purchase happened in 2023, a time when six quarters of declining crypto prices and mining difficulty had pushed many miners into bankruptcy. Argo was one of them. Helios was marketed as a flagship of the Texas mining wave: high-capacity, ideally located near wind generation, with a power purchase agreement tied into the ERCOT West zone. Galaxy bought it at a distressed price. But what is often forgotten: the facility's existing contract economics were a product of the early mining era. The PPA may expire in the next few years. If Galaxy needs to renegotiate power pricing in 2026-2027, it will do so at prevailing ERCOT prices, which have increased significantly because of data center demand.
That creates a perfect storm: the existing mining economics may deteriorate just as the HPC capex bill comes due. The reported "accelerating data center ambitions" may be a reaction to a looming PPA expiry, not an opportunistic AI play. If the old power contract is about to roll off, Galaxy must either reprice the power, curtail operations, or pivot to higher-margin compute. The market should pay attention to the power procurement details in the notes of the financial statements.
In DeFi, we look at TVL, total value locked, and we stress-test collateral ratios. For Galaxy, the TVL analog is total assets on the balance sheet, and the collateral ratio is the market value of Bitcoin holdings relative to liabilities. In a bear market, the risk is a bank run-like dynamic: if the market believes the company needs to sell Bitcoin to fund data center construction, the market will pre-sell the stock. This is not a rumor; it is the logic of a leveraged balance sheet.
I calculate a rough "BTC coverage ratio" — total BTC equivalents plus liquid assets, divided by total debt. If that ratio falls below some threshold, the stock trades like a credit instrument rather than a growth equity. A bear market is where the difference shows up. In 2022, many public miners became effectively credit stories; their stock prices stopped tracking Bitcoin and started tracking liquidity events.
If Galaxy issues convertible notes to finance the data center, the conversion premium becomes the effective strike price of a call option on the infrastructure project. The market will price the equity as a call option on the data center's future cash flows. This is a nice theoretical framework, but in practice, options are only worth their intrinsic value plus time premium. The time premium collapses during a bear market.
Now, to the most subtle layer: the narrative. In the last six quarters, the market has seen a flood of "AI + Crypto" crossover stories. Foundry, Core Scientific, Hut 8 — all have appeared in the headlines with data center stories. The Alpha of this trade was early discovery: the minute the market understood that a Bitcoin miner had a power asset that an AI cloud could use, those stocks were repriced. But the sector has matured. The multiple is no longer being awarded for "potential." It is being awarded for "signed contract."
If Galaxy's Q2 earnings show a continuation of narrative without signed contracts, it is not a failure of the company's ambition; it is a failure of the market's patience. The narrative bubble does not burst when the press release is written; it bursts when the balance sheet data shows a mismatch.
I have seen the same pattern in every NFT collection I reviewed. In early 2021, I proved that a popular collection's metadata was stored on a centralized gateway. The ownership showed the token hash, but the image could disappear. At the time, nobody cared, because price was determined by hype; now, years later, the "ownership" of that metadata is only as good as the gateway's uptime. The same lesson applies here: Galaxy's "data center asset" is only as good as its utilization, its contract, and its operating margins.
Narratives in crypto follow a natural life cycle: discovery, validation, extrapolation, disappointment, and salvage. The AI + crypto narrative is in the extrapolation phase. Every data center announcement is considered a bullish signal, regardless of the terms. But extrapolation ends when the first major company fails to convert a data center announcement into contracts. Galaxy could be that first case study, or it could be a success. The data on energy interconnection queues in Texas shows prolonged waiting times. Merely owning land in ERCOT's interconnection queue is not a schedule.
But I have to restrain my instinct to dissect. The bulls have at least one true fact: AI compute demand is enormous, energy access is scarce, and at least 4-5 years of elevated capex by hyperscalers is already visible in public guidance. Any firm that can safely offer a data center with existing interconnection, cheap power, and access to the Southwest's grid has a strategic position. The question is not whether the asset class has merit; it is whether Galaxy can execute.
Galaxy also has an underappreciated advantage: a client base that includes some of the largest institutional crypto investors in the world. These institutions need crypto custody, trading, and often desire exposure to AI-related infrastructure. A vertically integrated Galaxy can offer its existing clients a complete package: one company for custody, lending, and compute services. That cross-sell potential is not priced into the stock today because the data center business does not yet have revenue.
There is also the "toggle" argument. Because the Helios site can continue to mine Bitcoin while HPC demand builds, the company has a revenue floor. If the AI conversion is delayed for whatever reason, Bitcoin mining still produces cash flow—provided Bitcoin's price is high enough to cover energy costs. In a bear market, where the asset is trading below its historical average, this revenue floor may be all that separates the stock from a deeper discount.
I will also mention a quiet point often ignored. The acquisition of Helios was done at a price that may already be a bargain relative to comparable energy assets in 2026. If the facility has 180 committed megawatts of transmission capacity, the current market price for such interconnection in Texas could be higher than the purchase price. In that world, Galaxy's data center "ambition" is not just a build; it is a way to surface an undervalued asset on the balance sheet.
The bulls are right about the direction. They may be early about the timing. Timing matters more than direction in a bear market.
Let me also design a stress test, just as I did for Compound. The scenario: July 2026. A Texas heat wave pushes ERCOT energy prices to $2,000 per megawatt-hour for nine consecutive hours. Bitcoin price drops 20% due to a global macro shock. Galaxy has 0% of its mining load hedged for July because the energy forward market was backwardated. Result: the mining operation loses money for a week. The data center HPC expansion is still under construction, with no offsetting revenue. The company reports a quarterly operating loss.
Would a rational investor survive? If Galaxy had locked in energy prices for 18 months, the loss would be muted. If not, the loss is the price of the "energy optionality" that bulls celebrate. In my world, unhedged exposure is not optionality; it is a short volatility position. It pays off when the market is calm and explodes when the market is stressed. This is not a criticism; it is a classification. Since the company is publicly traded, I want to see the risk management framework disclosed in the MD&A. If the "risk" section of the shareholder letter is limited to market risk and lacks discussion of energy volatility, the management does not yet understand the business they are entering.
The article will not say "we don't know which services the data center will provide." But the numbers will. Look at "construction in progress" (CIP). A company can capitalize costs during construction. A sudden rise in CIP, matched by no rise in revenue, implies the project is consuming cash. It is a signal that can be seen before the words.
Also, look at "related party transactions." If the data center is leased back to an affiliate of Novogratz, that is a red flag. If the company buys GPUs from a related party, that demands additional scrutiny. The Q2 filing will include those details in the footnotes, not in the press release.
Data center expansion is not simply a matter of concrete and copper. GPUs are in high demand and long lead times. If Galaxy expects to open the site in 2027, it must place GPU orders now. We can do forensic analysis of "deposits on equipment" line items on the balance sheet: if there are big deposits to suppliers, they have reserved compute capacity. If there are no such deposits, the "accelerated data center" is just real estate development.
As for network connectivity — an HPC site needs fiber and low-latency routes. West Texas has historically poor connectivity compared to Dallas or Austin. Galaxy may need to invest in dark fiber or partner with a telecom. Another unspoken drag. The kind of hidden infrastructure that a typical crypto investor never notices, but exactly the kind of thing that will determine success.
So, for the hour after the earnings release, ignore the EPS headline and search for three numbers: capital expenditure guidance for the current fiscal year, free cash flow, and any committed data center capacity. Each number tells you whether the "bet" is a plan or a promise. If capex is high and capacity is zero, the company is in the promise phase. If capacity is positive, the story changes from crypto trading house to infrastructure developer.
The lesson from every prior audit I have run — from the Compound accumulator to the Terra validators to the NFT metadata — is that the market rewards those who verify the number before the narrative. Companies like Galaxy can be brilliant capital allocators in one year and cautionary tales the next. The Q2 report will not settle the debate, but it will begin settling it.
Verify the hash, ignore the narrative.