Crypto Mining to AI Data Center Conversion: The Workflow Developers Need

Former mining sites offer power and land. They also carry real risk. Here is how to evaluate and convert one correctly.

The $500 million acquisition of Genesis Digital Assets by AiOnX in June 2026 is the latest data point in a pattern that has been building for two years. Bitcoin mining facilities -- purpose-built for large power draws, often located in power-rich secondary markets, and now economically stressed by halving cycles -- are being repositioned as AI and high-performance compute data centers.

For data center developers, this is a genuine opportunity with a distinct risk profile. It is not a shortcut to a completed data center. The conversion workflow is substantive, and the failure modes are specific.

Why Mining Sites Have Appeal

The obvious thesis is power. A functioning mining facility with a signed utility interconnection has already cleared the hardest gate in data center development in 2026. Depending on the vintage and scale of the facility, you may be acquiring 50 MW to 300 MW of contracted load, substation infrastructure, and in some cases owned real estate at land and hard cost basis that is significantly below new-build replacement value.

The secondary benefit is geography. Mining operators historically located in markets with cheap, abundant power: Wyoming, Texas (ERCOT markets), Montana, Kentucky, Washington state (Columbia River basin). Several of these markets are now exactly where data center developers want to be -- either because of low power cost or because greenfield power access is constrained in primary markets.

AiOnX's Polaris Forge facility in Wyoming is the current flagship example. Applied Digital's repurposing of mining infrastructure for HPC. Bit Digital and Hive Blockchain have both announced AI pivot strategies backed by facility conversions.

Where Mining Sites Fall Short

The power advantage is real but partial. The rest of the conversion workflow is where deals either work or fail.

Cooling infrastructure is built for air cooling, not liquid. Mining hardware is air-cooled at relatively standard density. AI training workloads -- GPU clusters from Nvidia, AMD, or custom silicon -- require rack densities of 50 to 100+ kilowatts per rack, which air cooling cannot support at acceptable PUE levels. Immersion cooling or direct-to-chip liquid cooling systems must be installed. That means floor-up mechanical upgrades, a water loop design, heat rejection engineering, and potentially new water sourcing or cooling tower permitting depending on local regulations.

Structural loads may not support the required density. Mining facilities are typically single-story, non-critical construction -- concrete slab, metal panel walls, industrial roof systems. The raised floor or slab-on-grade system may not carry the concentrated point loads of GPU rack deployments with liquid cooling loops. Structural assessment by a licensed engineer is not optional.

Power delivery topology often needs reconfiguration. Mining operations draw power at relatively uniform loads without the redundancy requirements of Tier II or Tier III data center operation. The existing UPS, switchgear, generator, and distribution architecture may not support N+1 or 2N redundancy at the density levels an AI tenant requires. Depending on the vintage and quality of the original electrical installation, the conversion may require significant electrical scope in addition to the cooling upgrades.

Fiber and connectivity may need significant investment. Mining sites were not selected for latency. A facility in rural Montana with excellent power may have limited fiber options within a practical route distance. AI inference workloads in particular have latency requirements that may disqualify otherwise attractive sites. Training workloads are more tolerant of network latency but still require high-bandwidth, redundant connectivity.

Permitting for change of use varies significantly by jurisdiction. A site permitted as an industrial manufacturing or energy facility may need new use permits, noise variance amendments, or stormwater reconsideration for changed cooling infrastructure. Some jurisdictions have data center moratoriums or conditional approval requirements that apply to new AI infrastructure regardless of whether the underlying facility already exists.

The Conversion Workflow

A structured conversion diligence and delivery workflow runs across six stages.

Stage 1: Power and utility validation. Before any other analysis, confirm the contractual status of the utility interconnection. What tariff class applies? Is the load contract transferable to a new operating entity? Does a change of use or ownership trigger a new service application? In states with activated large load tariff frameworks, a change in ownership or load profile may trigger re-enrollment in the new tariff structure.

Separately, model whether the existing contracted load matches the proposed AI workload. Mining loads are typically continuous and flat-profile. GPU clusters can have highly variable draw profiles depending on workload type. Confirm that the utility service agreement and backup generation sizing support variable load operation.

Stage 2: Structural and civil assessment. Commission a structural engineer to review the existing facility for floor loading capacity at target rack density, roof loading if cooling towers or dry coolers are required on the structure, and seismic performance if the site is in a zone with meaningful seismic risk. Civil scope includes stormwater management for expanded cooling systems and site access for heavy equipment delivery during construction.

Stage 3: Electrical system audit. A licensed electrical engineer should survey the existing distribution topology against the proposed redundancy standard. Identify what can be retained, what must be replaced, and what new scope is triggered by the cooling upgrade. Produce a basis of estimate that separates existing system reuse from new-work costs. The gap between "existing switchgear" and "interconnection-compliant switchgear for AI data center" can be larger than the acquisition team expects.

Stage 4: Cooling design and feasibility. Engage a mechanical engineer to develop a cooling concept based on target rack density and PUE targets. Single-phase immersion, two-phase immersion, and direct-to-chip liquid cooling each have different site requirements. Water availability and local permitting for water use are often the binding constraints in secondary markets that lack municipal water infrastructure at the required scale.

Stage 5: Fiber and connectivity plan. Before the deal closes, map available fiber routes, identify providers, and establish whether redundant paths are achievable within a cost-effective route distance. If the site requires new fiber construction, model the capital cost, timeline, and its impact on lease commencement. Some AI tenants will provide or fund connectivity as part of a pre-leased development arrangement.

Stage 6: Tenant requirement alignment. The conversion scope should be driven by a specific tenant requirement or a clearly defined target tenant profile. The investment decision for a Tier II colocation product differs from the investment for a 50 MW hyperscale AI training facility. Commencing construction before tenant alignment is confirmed increases speculative development risk substantially.

AI's Role in the Conversion Workflow

AI assists at several stages of this workflow without replacing the engineering judgment that drives the core decisions.

In the due diligence phase, AI can aggregate utility tariff data across the site's jurisdiction, parse the existing service agreement for transfer provisions, monitor permit databases for change-of-use requirements, and compile a comparative analysis of recent mining-to-data-center conversions to establish a cost benchmark range.

During construction, AI can manage the procurement register for the long-lead electrical and cooling components, track RFIs and submittals for the conversion scope, and monitor schedule dependencies between the electrical, mechanical, and structural phases.

What AI cannot do: determine whether the site's power delivery infrastructure is worth the conversion cost versus a greenfield alternative. That call requires an engineer's field assessment and a developer's judgment about market timing and capital allocation. The site economics have to work without an AI providing the answer.

The Acquisition Frame

Mining-to-AI conversions can create genuinely attractive development economics, but the acquisition price has to reflect the real conversion scope. Buying a mining facility at replacement cost for a data center because it has power misses the conversion capex. Buying it at a meaningful discount to greenfield replacement cost while modeling a realistic conversion budget is the correct frame.

The facilities that convert successfully share a few characteristics: large utility interconnection at a voltage level that supports data center loads, reasonably modern electrical infrastructure, proximity to fiber (or a credible fiber plan), and a structural envelope that can accommodate the required cooling systems.

The facilities that become expensive mistakes are the ones where the power story was compelling enough to move forward before the cooling, structural, and connectivity gaps were priced in.

That gap is the workflow. It is not administrative. It is where the deal economics are determined.