Galaxy Digital announced Phase II construction of its Helios AI data center in West Texas. The headline reads like a validation of the crypto-to-AI pivot narrative. But the press release omitted the metrics that matter: power capacity in megawatts, GPU count, cooling architecture, and—most critically—any signed customer contracts. Without those numbers, this is an engineering update, not a business milestone.
I have spent the past decade auditing infrastructure transitions in crypto. In 2022, I led a team that stress-tested the data availability layer of a modular blockchain, and before that, I spent six weeks deconstructing Bancor V2's weighted constant product formula. The lesson I learned: complexity hides in the details. A bitcoin mining facility retrofitted for AI workloads is not a simple plug-and-play operation. The cooling system alone can consume 30% of the conversion budget, and the network topology for high-density GPU clusters is fundamentally different from ASIC rows.
Helios is a 100+ acre site originally built for Bitcoin mining. The land, power transformers, and grid interconnection are valuable assets. But AI data centers require precision cooling—typically liquid cooling for GPU clusters—and redundant power feeds with UPS systems that handle microsecond-level fluctuations. Bitcoin miners, by contrast, are tolerant of intermittent power outages. They can shut down and restart without data loss. AI training runs cannot afford that luxury. A single checkpoint failure can cost thousands of dollars in compute time.
Check the math, not the roadmap. Galaxy Digital has not disclosed the total power capacity of Phase II. Without that, we cannot estimate the number of GPUs, the type of chips (NVIDIA H100, B200, or AMD MI300X), or the potential revenue per megawatt. The industry standard for AI inference is around $2–$4 per GPU-hour, while training can fetch $8–$12 per GPU-hour. But those rates are only achievable if the data center has the right interconnect fabric (InfiniBand or NVLink) and low-latency networking. Bitcoin mining does not require any of that.
Audits are snapshots, not guarantees. The same applies to infrastructure. A Phase II announcement means the project has moved past the planning stage, but it does not confirm that Phase I is fully operational or profitable. Many crypto miners that pivoted to AI have faced delays: Core Scientific filed for bankruptcy in 2022 before emerging, and Hut 8's AI transition has been slower than expected. Galaxy Digital, primarily a financial services firm, lacks the operational track record of dedicated data center operators like CoreWeave or AWS. The engineering team may be competent, but scaling a data center is a different discipline from managing a crypto treasury.
Let me be specific. Based on my experience auditing modular blockchain infrastructure, the shift from ASIC-based mining to GPU-based AI computing requires a complete overhaul of the power distribution system. ASICs run at 12–18 V DC, while GPUs require 48 V or higher, with different current demands. The transformer taps, busbars, and circuit breakers must be re-specified. The cooling infrastructure must handle 10–30 kW per rack, compared to 2–3 kW for a typical Bitcoin miner. The networking layer must support 400 Gbps or higher for gradient synchronization. These are not trivial changes.
Complexity is the enemy of security. Every added layer—power conversion, liquid cooling, network routing—introduces failure points. A single pump failure in a liquid cooling loop can overheat an entire GPU cluster, causing hardware damage and training job loss. The Grid in West Texas (ERCOT) is known for volatile prices and occasional blackouts during winter storms. Galaxy Digital may have secured fixed-price power purchase agreements, but the public has not seen those contracts. Without them, the financial model is incomplete.
On the market side, the crypto-to-AI pivot narrative has driven up stock prices of mining companies like IREN, Core Scientific, and Bit Digital. Galaxy Digital is a publicly traded company (ticker GLXY on TSX), and its stock has benefited from this narrative. However, the market often prices in expectations before fundamentals materialize. The key metric to watch is the revenue per megawatt from AI services versus the cost of conversion. If Galaxy Digital signs a multi-year lease with a hyperscaler or AI lab, that would be a real signal. A Phase II construction update is not.
Contrarian angle: The pivot may actually dilute Galaxy Digital's core value proposition. The company was originally a crypto merchant bank, with strengths in trading, lending, and asset management. Moving into physical infrastructure brings capital intensity, operational risk, and regulatory scrutiny from utilities and environmental agencies. The data center business has low margins at scale, dominated by players with decades of experience. Galaxy Digital may find itself competing in a capital-intensive industry where its financial acumen is less of an advantage.
Furthermore, the AI demand cycle is not guaranteed to last. The current GPU shortage is driven by the training of large language models, but inference workloads are increasingly moving to specialized ASICs (like Google's TPU or Amazon's Trainium). If the market shifts toward custom silicon, general-purpose GPU clusters may become less valuable. Bitcoin mining, by contrast, has a fixed demand: securing the network. That demand is not going away. Galaxy Digital may be selling a cyclical narrative for a temporary premium.
Takeaway: Galaxy Digital's Helios Phase II is a technical milestone, but it is not a business validation. The missing data—power capacity, GPU type, customer contracts, cooling architecture—are the inputs that matter. Without them, the announcement is noise. Investors should demand the math before buying the roadmap. As I wrote in my 2024 analysis of sequencer centralization, “Layers add latency, not just features.” The same applies here: converting a bitcoin mine to an AI data center adds complexity, not just value.
I will be watching for the next quarterly report. If Galaxy Digital discloses a signed lease or a binding letter of intent with a credible AI customer, then the pivot has substance. Until then, the Helios Phase II is a construction project, not a thesis.
