The market is pricing every miner with a press release about AI as a CoreWeave-lite. But the code—or in this case, the electrical load—does not lie. Galaxy Digital announced Phase II of its Helios AI data center in West Texas, converting a former bitcoin mining site into an HPC facility. The narrative is seductive: stranded energy assets repurposed for the AI boom. But the engineering reality is a different beast entirely.
Context: The Crypto-to-AI Pivot Factory Galaxy Digital is a publicly traded crypto financial services firm, not a data center operator. Helios is its flagship mining site in West Texas, a region known for cheap wind and solar power, but also for grid instability. The Phase II announcement signals a shift from ASIC-packed warehouses to GPU clusters. This is part of a broader wave: Core Scientific, Hut 8, IREN, and others are all pitching the same story. The market loves it—stocks have rallied on thin air. But the devil is in the MW.
Core: The Engineering Gap Converting a bitcoin mine to an AI data center is not a simple plug-and-play. Bitcoin mines are designed for ASICs: low power per unit, air cooling, high tolerance for heat, and minimal latency requirements. AI data centers demand high-density GPU racks, liquid cooling, redundant power feeds, and low-latency networking. The structural load on floors, the cooling capacity, the electrical distribution—all need overhaul.
Galaxy has not disclosed the power capacity of Phase II. Typical AI data centers run at 50-100 MW per facility. A single Nvidia H100 GPU draws 700W; a cluster of 10,000 requires 7 MW just for the GPUs, plus cooling overhead. Without knowing the MW, the GPU count, or the cooling method, we are flying blind. Precision is the only hedge against chaos, and here there is none.
Based on my experience auditing mining operations and evaluating data center capex, the transition costs are substantial. Retrofitting an existing shell can save 20-30% versus greenfield construction, but only if the electrical infrastructure is already high-voltage and the cooling system is scalable. Helios, being a bitcoin mine, likely has a 138 kV substation and transformer capacity. That is a head start. But the cooling system? ASIC miners use air; AI requires liquid or chilled water. That means new piping, pumps, and possibly a chiller plant. The cost per MW for AI-ready space is $8-12 million. For a 100 MW site, that is $800 million to $1.2 billion. Galaxy's balance sheet? Market cap around $5 billion, but they have debt and other operations. They will need financing—either equity dilution or debt. Check the gas, then check the truth—in this case, check the capital raise.
Another hidden issue: the West Texas electricity market (ERCOT) is prone to price spikes. In winter storms, grid prices can hit $9,000/MWh. Bitcoin miners simply shut down; AI data centers cannot. They require 24/7 uptime. That means Galaxy must either secure firm power contracts, build battery storage, or accept the risk of curtailment. The latter is unacceptable for AI clients. So the site needs additional on-site generation or costly PPA hedges. Alpha hides in the friction of liquidity—here, the friction is power reliability.
Contrarian: The Narrative vs. The Reality The retail crowd is piling into mining stocks as AI plays. They see the headline and assume instant value creation. But the smart money is asking: who are the clients? Core Scientific signed a 12-year deal with CoreWeave for 200 MW. Hut 8 has a deal with AI startup. Galaxy has not announced any customer. Without a lease, this is speculative construction. The market is pricing the option, not the cash flow.
Moreover, Galaxy is a financial firm, not an operator. They will likely outsource the management to a data center specialist or a colocation partner. That introduces margin compression. The real value capture is in the land and power, not the operation. This is a capital-intensive real estate play, not a technology breakthrough. Volatility is the tax on uncertainty—and the uncertainty here is high.
Another contrarian angle: the crypto-to-AI pivot is a double-edged sword. If Galaxy dedicates Helios to AI, it reduces its bitcoin mining capacity. That could hurt its mining revenue if bitcoin surges. The company is effectively betting that AI margins will exceed mining margins. That may be true today, but the AI GPU market is volatile. GPU supply is tight, but demand could shift. The hyperscalers (AWS, Azure, Google) are building their own capacity. The mid-tier AI firms may not survive the shakeout. If AI demand softens, Galaxy is left with a white elephant.
Takeaway: The Real Metric to Watch The Phase II announcement is a signal, not a proof. The key numbers are not construction milestones but signed contracts: power purchase agreements, colocation leases, and client commitments. Without them, this is just a press release. Backtest the assumption, not just the data—assume the narrative is priced in, then look for the gap. The gap is currently empty.
Yield is never free; it is rented. Galaxy is renting hope from the market. The real test will come when they need to sign a 10-year lease with a tenant at $0.10/kWh. Until then, I am watching the MW, the cooling system, and the client list. The code does not lie, but it does hide. In this case, it hides behind a press release.
