When Galaxy Digital bought a distressed Bitcoin mining campus in Dickens County, Texas, at the end of 2022, the move looked opportunistic rather than transformative. The $65 million rescue of Helios—acquired from an overextended miner—was, at the time, just another line on a balance sheet known for eclectic digital-asset bets. Less than three years later, that “accident” has evolved into Galaxy’s central strategy: a pivot from crypto finance to owning and operating one of the largest AI-ready data center platforms in North America. In a world where power, not processors, has become the hard constraint, the company’s wager is audacious and, if it works, generational.
This is the story of how a crypto-era holding company is recasting itself as an AI infrastructure owner; why power contracts and grid interconnections are the new currency of cloud; and what the Helios campus tells us about the next decade of the data economy.
Galaxy Digital was built to be a diversified, publicly listed vehicle for the digital-asset economy. Its operating businesses—trading and prime services, market making, derivatives, lending, investment banking, asset management, custody, staking—gave it the profile of a crypto-native Goldman Sachs with a sprinkling of Berkshire Hathaway’s holding-company DNA. It also carried a meaningful book of liquid crypto assets and venture stakes across the ecosystem.
That architecture still matters. The trading franchise and asset-management arm supply fee income, counterparty reach, and deal flow. Its custody technology (via GK8) and staking services equip the firm to help institutions hold and use tokens at scale. A significant balance sheet—net cash and investments measured in the billions—gives management the option to acquire, invest, and build through cycles. Those capabilities explain why Galaxy could move quickly when Helios came up for sale and, later, why it could finance an industrial-scale retrofit for AI.
But they do not explain the strategic audacity. That came later, when management leaned into an emerging reality: the AI boom would not be limited by chip supply forever. It would be limited by electricity, permits, and the ability to turn megawatts into usable, reliable GPU hours quickly. Helios had exactly what the market would soon prize—an 800MW approved interconnection with ERCOT and a pathway to 2.5GW at a single, scalable campus.
The Scarcity That Matters Now
For a decade, conventional wisdom held that hyperscale computing was constrained by semiconductors: first CPUs, then the GPU accelerators that power machine learning. That was true—until it wasn’t. Over the last two years, the chokepoint shifted decisively from silicon to electrons. U.S. electricity demand, flat for much of the 21st century, is now bending upward. Utilities and independent system operators face multi-year backlogs for new grid connections. High-voltage transformers can take two years to arrive. The planning and permit process for a fresh interconnection can easily run 36 months.
For AI developers and hyperscale clouds, those timelines are incompatible with competitive reality. Training frontier models requires dense clusters of the latest GPUs, purpose-built electrical and cooling infrastructure, and—above all—assured, affordable power. That is why existing, large-scale campuses with live or near-term grid capacity command a premium. It is why operators able to deliver hundreds of megawatts in one place, with fiber to major metros and designs that can accommodate liquid cooling for successive GPU generations, are suddenly in the driver’s seat.
Helios checks those boxes. The campus sits adjacent to a major switching station in West Texas, tying into one of the country’s most dynamic power markets. Its 800MW of approved capacity represents more than two percent of current U.S. data center power at a single site—unusual in an industry whose incumbents historically distributed smaller facilities across dozens of metros. An additional 1.7GW is in the approval pipeline. If even a portion clears, Helios scales into the same order of magnitude as the portfolios of publicly traded data-center REITs, but concentrated in one campus optimized for AI.
A Triple-Net Bridge to AI Scale
Galaxy’s retrofit strategy is straightforward: own the shell, upgrade the power and cooling backbone, and lease capacity on long-term, triple-net terms to tenants that bring their own chips and servers. The economics, when executed well, resemble a data-center REIT more than a miner or a merchant compute operator. Rent escalators provide inflation protection. Tenant-paid operating costs protect margins. And long-dated leases create annuity-like cash flows that can be levered with project finance rather than diluting equity.
The company’s landmark agreement with CoreWeave—an “AI hyperscaler” focused solely on GPU-based compute—validated the approach. The initial ~600MW lease ramps in phases beginning in 2026 as Helios completes its conversion to AI-grade infrastructure. The structure is emblematic: roughly $720 million per year in base rent, three percent annual escalators, and a triple-net framework in which the tenant bears property-level costs. The resulting margin profile is unusually high for infrastructure: management points to ~90 percent EBITDA margins on this stream, implying roughly $650 million of EBITDA when fully online.
The key, for Galaxy and for investors, is that this lease covers less than a quarter of the site’s total potential capacity. If the company signs out the next gigawatt—management expects an additional 800MW approval in the near term and is planning for a further 900MW—contracted EBITDA climbs past $1.7 billion on Helios alone. Applied to market REIT multiples, that math pushes enterprise value for the campus into the tens of billions. Importantly, the equity requirement for each subsequent tranche falls as stabilized rent rolls support refinancing and recycling of capital into new phases.
Why Galaxy’s Deal Looks Better Than the Miners’ Pivots
Galaxy is not the only firm that spotted the arbitrage between mining and AI. Several publicly traded miners have announced conversions, partnerships, or ambitions to reallocate power to GPU workloads. A few have signed credible leases. Yet the market continues to treat most miners as structurally challenged. The reasons are practical.
First, balance sheets. Many miners spent the last cycle issuing stock to survive drawdowns and halving-driven revenue compression. That leaves them with limited cash and constrained debt capacity—hardly ideal when a single 200MW tranche can demand hundreds of millions in construction spend before the first rent dollar arrives. Galaxy’s capital position, by contrast, gives would-be tenants confidence the site will be finished on schedule.
Second, credibility and focus. Galaxy has fully exited Bitcoin mining to remove perceived conflicts and execution drift. The company has recruited senior operators across construction, procurement, finance, and development. A successful Phase I delivery for CoreWeave will give risk-averse hyperscalers comfort that the team can hit industrial-grade milestones. That feedback loop—prove one tranche, unlock the next—compounds quickly in a market where every major cloud is hunting for power.
Third, embedded expansion. The Helios approval pipeline is unusually large for a single campus. That matters to tenants who want to scale with demand rather than overcommit too early. It also matters to financiers, who prefer to underwrite platforms with line of sight to replication rather than one-off conversions. And it matters to regulators and local stakeholders who want to see coherent, staged development rather than piecemeal builds.
Tenant Concentration and the CoreWeave Debate
No discussion of Helios is complete without the obvious risk: concentration. In the early years, CoreWeave is a dominant counterparty, and the market has wrestled with questions about its leverage, business model novelty, and sensitivity to AI cycles. Some of the skepticism is warranted. Non-investment-grade tenants pay more, and concentration reduces optionality if conditions change. But much of the fear confuses operating model with credit structure.
CoreWeave’s growth has been contract-led and capitalized accordingly. The company signs multi-year, often prepaid agreements with blue-chip AI customers, then draws on delayed-draw facilities to buy GPUs and stand up capacity. That sequencing means leverage is tied to committed revenue, not speculative build-it-and-they-will-come bets. Crucially, as more tenants sign and the revenue base diversifies, the risk profile improves. And for landlords like Galaxy, tenant selection is not only about credit ratings; it’s about the ability to ramp on schedule, pay escalators, and renew at higher densities as GPU generations advance.
For investors, the practical mitigation is time and execution. Deliver Phase I. Prove out the economics. Then layer in additional tenants—whether neoclouds or traditional hyperscalers—against incremental approvals. The faster that flywheel spins, the less any single counterparty defines the story.
Design for the Next GPU, Not the Last One
Retrofitting a Bitcoin mining campus into an AI-grade facility is not an exercise in swapping machines. It requires re-engineering the electrical backbone, cooling modality, and rack design to anticipate densities that climb with every GPU generation. Air-cooled halls built for ASICs cannot support the thermal loads of liquid-cooled Blackwell systems. Power distribution must be designed for clusters that knit tens of thousands of accelerators with high-speed interconnects. Fiber must be provisioned for both training backhaul and latency-sensitive inference paths to major metros.
Helios’s staged rollout is a feature, not a bug. Delivering 200MW in 2026 and 400MW in 2027 allows each tranche to incorporate the latest standards for immersion and direct-to-chip liquid cooling, switchgear, and redundancy. It also allows Galaxy to evolve contract structures with tenants as the market standardizes on service-level expectations for AI clusters, including thermal envelopes, rack footprints, and maintenance windows. The on-site freshwater pond, while hardly the sole determinant of thermal design, adds flexibility for liquid-cooling strategies in a region where temperatures and water politics must be managed prudently.
Just as important is long-haul fiber. At ~250 miles from Dallas–Fort Worth, Helios is close enough to serve inference workloads once fiber paths are optimized for 10–15 millisecond round-trip times. That dual-use profile—training at scale and latency-tolerant inference, with the option to pinch-hit for metro-adjacent inference—gives the campus a longer demand tail. It is also a hedge against the narrative that all value will shift from remote training clusters to edge inference. In reality, the pie is growing; training’s share may fall, but absolute dollars and megawatts dedicated to it will not.
Financing the Build Without Diluting the Future
The capital model for Helios is textbook project finance adapted to a new asset class. Construction loans and delayed-draw facilities bridge the capex while phases are built. Triple-net rent rolls, once stabilized, support take-out refinancing at lower rates, with excess proceeds recycled into subsequent tranches. With early phases contracted at high margins, the equity need per incremental 200MW falls as the platform matures. That is how an owner can scale from hundreds of megawatts to gigawatts without serial equity issuance.
Here, Galaxy’s holding-company structure and market listing confer advantages. The NASDAQ uplisting broadens the investor base, reduces the cost of capital, and opens a path to index inclusion, which improves liquidity and valuation resilience. The crypto operating businesses, far from being a distraction, provide fee income and relationships that can be parlayed into data-center deal flow and tenant introductions. Over time, management may pursue a structural separation—spinning the data-center platform into a REIT-like vehicle—once revenue is seasoned and the investor base is ready to value each segment on its merits.
The Strategic Logic for Hyperscalers and Neoclouds
Why should a hyperscaler lease rather than build? The answer is velocity. Even giants with in-house development teams face grid, transformer, and permit bottlenecks that no amount of capital can compress. Leasing from a third-party campus with ready interconnection is a way to pull forward compute delivery while captive sites wend their way through approvals. The choice is not binary; clouds routinely blend owned and leased capacity. For AI specifically, a portfolio approach—some campuses dedicated to frontier training, others for enterprise inference—will dominate.
Neoclouds like CoreWeave exist because customers value specialization. Running massive, liquid-cooled GPU clusters with high reliability and utilization is a different sport than traditional VM provisioning. Customers want operators that can tune networks, schedule jobs efficiently across generations of GPUs, and deliver committed capacity with predictable economics. Those operators need landlords who can finish on time and scale in place. That complementarity is why a competent campus owner can sign meaningful leases before the first hall goes live—and why early delivery can unlock a queue of additional tenants.
Why the Market Still Misprices Galaxy
If the thesis is so compelling, why doesn’t the market fully reflect it? Partly because the story is complex and, to many investors, novel. Galaxy is neither a pure-play REIT nor a pure-play crypto bank. It is a holding company with a powerful core business and an emergent infrastructure platform, both operating in sectors prone to volatility and narrative swings. Equity research coverage has been slow to catch up. Many institutions with mandates against foreign small/mid-caps could not own shares prior to the NASDAQ listing. And sell-side models rarely credit multi-gigawatt optionality until permits are granted and PPAs or leases are inked.
The other reason is execution risk. Building one of the largest AI campuses in the country on a compressed timeline, during a period of acute supply-chain tightness for thermal and electrical components, is non-trivial. The market will rightly demand proof: milestones met, spend controlled, leases executed, phases energized. That is why each quarterly update matters more than commentary. And it is why management’s decision to shed the mining business entirely—to clarify focus and remove a source of equity overhang—was not cosmetic but strategic.
Governance, Incentives, and the Value of Saying “We Got Lucky”
Leadership and alignment often read like boilerplate. They aren’t here. CEO Mike Novogratz owns a majority stake, making him the rare public-company chief with both control and a visible, long-term economic incentive to create durable value. President and CIO Chris Ferraro brings private-credit rigor to the financing architecture underpinning Helios and future sites. That matters when coordinating lenders, tenants, equipment suppliers, and grid operators across a capital stack that must be resilient through cycles.
Equally notable is the candor with which management describes Helios. They did not foresee the timing or intensity of the AI power crunch when they bought the site. They have been explicit about luck, and about the decision to lean into it rather than spin a revisionist narrative. That credibility—admitting the origin story and focusing on execution—strengthens the case they make to hyperscalers: we have the power, we will finish on time, and we are building a platform around it.
Competitive Landscape: Old Rules, New Game
Traditional data-center leaders—Digital Realty, Equinix, American Tower’s data assets, and private-equity-backed platforms—remain formidable. They possess procurement clout, operator muscle memory, and deep customer relationships. But the center of gravity is shifting. What counted as hyperscale in the cloud-SaaS era—10MW sites near major metros—looks small against 500MW to multi-GW AI campuses with liquid cooling and high-density racks. Many legacy sites are hard to retrofit economically. New sites face the same interconnection gauntlets as everyone else.
That is why the next decade of the sector will likely be a barbell: incumbent networks continue to dominate metro-adjacent enterprise and interconnect-heavy workloads, while a new class of “AI megacampus” owners races to bring power-dense training and inference capacity online. Galaxy’s ambition is to be in the second group—first with Helios, then with a portfolio of converted or greenfield sites. The company has surfaced a pipeline of roughly 40 potential acquisitions averaging ~500MW each, much of it drawn from mining infrastructure now out of favor with financial markets. If even a fraction close on Helios-like terms, the platform extends from a lucky break to a defensible franchise.
Valuation: Thinking Like a REIT, Acting Like a Platform
How should executives and investors think about valuation when a company straddles categories? Start with first principles. Every 200MW tranche at Helios (roughly 133MW of critical IT load at a PUE of 1.5) translates into about $240 million in annual rent and roughly $216 million of EBITDA under a triple-net structure. Apply a conservative mid-20s multiple—a discount to where mature data-center peers trade—and the enterprise value uplift per tranche is on the order of $4.9 billion, partly funded with non-recourse project debt. Because the tenant pays opex, incremental EBITDA is highly visible once energized.
Then add the holding-company layers: a sizeable balance sheet of liquid crypto, a profitable institutional services platform, custody and staking technologies, and an early position in a regulated euro stablecoin via AllUnity. Individually, none match the near-term cash generation of a leased campus. Together, they support the corporate center, seed optionality in tokenization and stable-value infrastructure, and—critically—reduce the need to raise dilutive equity for Helios and beyond.
It is not difficult to build paths from today’s valuation to scenarios where contracted EBITDA, at REIT-like multiples, implies enterprise values that eclipse the company’s current market cap several times over. The hard part is not the spreadsheet; it is the schedule. Tranches must be delivered, approvals granted, fiber lit, transformers installed, and tenants ramped. If those things happen on or ahead of plan, the equity calculus changes quickly.
Galaxy appears materially undervalued on a sum-of-the-parts basis:
- Balance sheet ~$6/share
- Crypto business ~$14/share
- Helios data center ~$25/share
Fair value ~$45/share today, with upside north of $100 if Helios is executed flawlessly.
The Uplisting as a Structural Catalyst
Galaxy’s uplisting to the NASDAQ is more than a marketing milestone. It addresses mechanical barriers that kept large pools of capital on the sidelines: mandates against foreign listings, minimum liquidity thresholds, index eligibility. It lowers the company’s cost of capital, which matters in a capital-intensive buildout. It also invites a different research and investor-relations dialogue—one that forces a clearer articulation of segment economics and, eventually, may lead to a structural separation that allows public markets to value the data-center platform as infrastructure rather than as an appendage of a crypto holding company.
In the meantime, the uplisting serves a simpler purpose: it broadens the base of potential long-term owners just as the company begins to report tangible milestones at Helios. If those milestones are met, coverage follows flows, and flows follow proof.
What Could Go Wrong—and How to Watch for It
Three families of risk dominate.
First, delivery risk. Supply-chain slippage in transformers, switchgear, or liquid-cooling components can push energization dates. Galaxy must demonstrate procurement foresight, diversified vendors, and contingency plans. Watch for milestone granularity in updates: pad dates, energization windows, and acceptance testing, not just ribbon-cuttings.
Second, concentration and counterparty risk. Until tenant diversification arrives, CoreWeave’s health matters. Investors should track not just headlines but the cadence of customer prepayments, tenor and pricing of its debt stack, and evidence of successful delivery to its end customers. Positive signals include additional long-dated contracts, steady debt costs, and public-market access on reasonable terms.
Third, regulatory and power-market risk. ERCOT’s policy environment is business-forward, but transmission politics and curtailment debates are dynamic. The safeguard is smart interconnection design, on-site resilience where feasible, and a portfolio approach over time that includes multiple ISOs and states.
None of these risks is idiosyncratic to Galaxy; all are endemic to the AI buildout. The difference lies in balance-sheet strength, governance alignment, and the discipline to sequence growth behind contracted cash flows.
The Broader Context: Data Centers as Industrial Policy
There is a final reason Helios matters beyond its contribution to a single company’s fortunes. It illustrates a shift in how nations will think about digital infrastructure. The last great buildout—mobile broadband and cloud—ran on spectrum auctions, fiber backbones, and massive metro campuses. The next one runs on electrons, water rights, and interconnection queues. It will be shaped by utility commissions and ISO planning as much as by developer roadmaps. In that world, the most valuable companies will be those that can translate power into AI capacity faster and more reliably than peers—without losing sight of environmental and community obligations.
That is not a technocratic footnote. It is a competitive weapon. The firms that master it will not only lease racks; they will help define the contours of the 2030s economy: what can be trained, where, and how quickly ideas can move from model weights to products in customers’ hands.
The Road Ahead for Galaxy Digital
Galaxy’s North Star is clear. Execute Phase I and II at Helios on time. Convert approvals into power. Sign the next tenants. Refinance construction debt as rent rolls stabilize. Extend the platform with a measured cadence of acquisitions, favoring sites with near-term interconnections and credible paths to scale. Professionalize reporting so investors can see the annuity taking shape beneath the crypto holding company chassis.
If management does those things, the market will do the rest—re-rating the enterprise not as an interesting crypto conglomerate with a lucky windfall, but as an early leader in a new class of critical infrastructure. In that future, Galaxy’s origin story is not a curiosity but a competitive advantage: a firm built in the volatility of crypto that learned to move fast, finance complex assets, and cultivate institutional relationships—skills that, in the age of AI power scarcity, translate directly into value.
Helios began as a rescue. It may end as a template.
These are some of the key numbers that define Galaxy’s business today:
- 🌐 Market Cap: ~$9.3B
- 🌍 Employees: 600+ worldwide
- 💰 Balance Sheet: ~$2.1B (in digital assets, excluding stablecoins and cash)
- 📈 Q2 2025 Profits: $211M
- 🏦 Assets Under Management (AUM): ~$9B
- 🤝 Institutional Clients: 1,400+ served
- 🏦 Institutional Loans: $1.1B issued
- ⚡ Data Center Capacity: 800 MW approved for their data center