Hook Galaxy Digital just bought dirt. 230 acres of scrubland near Waco, Texas, for a data center campus that won’t see full power until 2028. The market cheered: another mining company pivoting to AI, another narrative to pump the stock. I see a different number: 1.63 GW approved by ERCOT for the Helios site, fully leased to CoreWeave for 15 years. That’s the headline. The fine print? That power comes from a grid that froze solid in 2021 and is now asked to feed an AI beast that never sleeps. The ledger bleeds faster than the logic holds.
Context Galaxy Digital (GLXY) started as a crypto financial services firm—trading, lending, mining. Then came the 2022 bear, and CEO Mike Novogratz shifted gears. The company bought the Helios mining facility in Texas, secured massive power allocations from ERCOT, and inked a 15-year lease with AI cloud provider CoreWeave for the entire 1.63 GW. Now they’ve added a second site: 230 acres, initial 74 MW from ERCOT, targeting 300 MW, with a mix of AI/HPC and Bitcoin mining. Average power cost: $0.43 per kWh. Smart money sees a pivot from cyclical mining to stable infrastructure cash flows. I see a single point of failure wearing a cowboy hat.
Core: The Grid is the Bottleneck Let’s dissect the mechanics. ERCOT approved 1.63 GW for Helios. That’s enough to power 1.3 million homes. One customer: CoreWeave. One grid: the Texas grid that failed during Winter Storm Uri, killing hundreds and leaving millions without power. Since then, ERCOT has added capacity, but peak demand from AI data centers is a new stress test. Galaxy’s new campus adds another 300 MW. The total draw could exceed 2 GW by 2028.
Based on my audit experience, this is the kind of concentration risk I flagged in 2017 ICOs—centralized dependency hidden behind a glossy pitch. In 2020, I wrote Python scripts to monitor Uniswap liquidity pool imbalances; today I monitor ERCOT’s reserve margin. The numbers don’t lie: the Texas grid has less than 10% reserve margin during summer peaks. A single heat wave or cold snap could force rolling blackouts. Data centers aren’t hospitals; they don’t get priority. CoreWeave can survive a few hours of downtime. Galaxy’s P&L cannot.
The lease is a double-edged sword. The 15-year term with CoreWeave provides stable cash flow—smart hedging against Bitcoin volatility. But it also locks in a single tenant. If CoreWeave stumbles (overbuilding, AI demand slowdown, or a grid failure that violates their SLA), Galaxy has no revenue buffer. The $8M interest income from the seller note on the land acquisition is trivial.
And the power cost? $0.43/kWh is cheap for AI (typical hyperscale rates are $0.05-0.10), but it’s expensive for Bitcoin mining. That explains the pivot: mining with that power would bleed cash at current hash rates. The AI lease gives them a 10-15% margin above mining returns. But that margin is thin insulation against a grid failure. I count the cracks before the dam breaks.
Contrarian Angle: The AI Narrative is a Distraction The market is repricing GLXY as an AI infrastructure play—higher multiples, bigger buzz. The contrarian truth is that Galaxy’s real asset isn’t the power capacity; it’s the regulatory approval. ERCOT’s blessing to connect 1.63 GW to the grid is the true bottleneck. Competitors can buy land, but getting ERCOT interconnection is a multi-year, multi-million-dollar process. That’s the moat.
But retail is missing the fragility. In 2022, I shorted LUNA/UST because I saw the death spiral mechanism—a system that looked stable until it wasn’t. Same here: the Texas grid looks stable until a polar vortex hits. The market assumes ERCOT will invest in hardening. History says they’ll kick the can. Galaxy’s business model relies on a regulator that is politically starved for funds. Risk is not a number; it is a feeling you ignore.
Also, the AI demand narrative has a catch. If AI compute demand plateaus or shifts to edge inference, CoreWeave may not need the full 1.63 GW. The lease might have step-up clauses, but we don’t know. In 2025, I built an AI agent to trade options on decentralized derivatives; I learned that data is only as good as the assumptions under it. The assumption that AI demand grows at 100% CAGR forever is a dangerous extrapolation.
Takeaway: The Next Crack Galaxy Digital is executing a clever strategy: convert mining real estate into AI real estate. The 1.63 GW approval and CoreWeave lease are tangible wins. But the thesis hinges on a single grid that has already failed. When the next winter storm hits, watch the panic in GLXY options. The premium on those puts will tell you more than any analyst report. Survival is the only alpha that compounds.