Two markets are pricing the same asset. A buyer evaluating a 50MW mining site in West Texas as a bitcoin facility sees it through one lens — power cost, hashprice, ASIC efficiency, payback period. A hyperscaler or neocloud evaluating the same site as a future AI training campus sees it through a completely different lens — fiber proximity, water access, interconnection queue position, and the ten-year cost of avoiding the wait. The result is one of the widest valuation spreads we've ever seen on infrastructure assets that look superficially identical.
The two pricing regimes
Mining-to-mining transactions in 2026 are clearing in a tight band. The data we're seeing across roughly two dozen disclosed and undisclosed transactions points to $300K-$500K per MW for an operational facility with average power costs and a credible operator at the helm. The high end of that band requires sub-4-cent power, redundant substations, and an established operations team transferring with the asset.
Mining-to-AI conversions are pricing in a different universe. Confirmed and rumored transactions are clearing at $3M-$8M per MW, with the upper bound reserved for sites that already have fiber on-property, sufficient water for closed-loop liquid cooling, and an interconnection capacity headroom that doesn't require a new queue submission. The Riot/AMD lease that closed earlier this month implied roughly $1.24M per MW per year on a long-dated lease basis — capitalize that at a reasonable rate and you're back in the $6M-$8M per MW range as an outright sale comparable.
What actually drives the multiple
The temptation is to attribute the spread entirely to demand-side enthusiasm. That's part of it, but the structural drivers are more durable than the cycle. We see five factors that consistently separate a $400K/MW mining sale from a $5M/MW AI sale on otherwise comparable sites.
- 01Interconnection queue position. A site with executed LGIA (Large Generator Interconnection Agreement) capacity in place for the announced load is worth multiples of a site that needs to queue. In ERCOT and MISO the queue is 36–60 months deep; PJM is functionally closed for new submissions through 2028.
- 02Fiber proximity. AI training workloads need diverse routes to at least two long-haul carriers. Sites within 5 miles of dark fiber routes price meaningfully above sites that need a 20-mile lateral build.
- 03Water rights and cooling readiness. Liquid cooling is now table stakes for any new training facility. Sites with sufficient water rights or proximity to municipal supply price up; dry sites in arid regions either need closed-loop systems with expensive chillers or get discounted to mining-comp value.
- 04Power contract structure. A take-or-pay PPA at 3 cents looks great to a miner and limiting to a hyperscaler. Sites with flexible structures — block-and-index, behind-the-meter generation, or curtailable load with credit — command premium prices because they translate to either valuation model.
- 05Operator credit and lease history. A site that's transferred operating control once before with clean documentation and audited financials prices above a site that requires the buyer to underwrite a first-time conversion.
The bid stack today
We track who's actively bidding across these transactions. The mining-side bid stack in 2026 is thinner than at any point since 2022. Public miners are deleveraging, not acquiring. Private operators with conviction on hashprice are still active but writing smaller checks. Sovereign and institutional capital that funded the 2023 buildout has largely rotated into AI-adjacent assets or exited the sector entirely.
The AI-side stack is the opposite story. Neoclouds raising Series B and C rounds are deploying that capital aggressively into capacity. Established hyperscalers are buying optionality. A new category — well-funded sovereign data initiatives — is starting to bid on the upper end of the spread.
“The mining-comp value is a floor, not a target. Any operator running a site that could plausibly be converted owes it to themselves to underwrite both numbers.”
What we expect through Q4 2026
The spread will narrow, but not collapse. As more conversions complete, more buyers will get comfortable underwriting the conversion premium, and arbitrage will tighten. We expect the mining-comp band to drift up modestly (toward $400K-$600K per MW) as supply gets absorbed, and the AI-ready band to compress somewhat as more sites come online with completed conversions priced in.
Where the spread stays wide: sites with already-executed interconnection agreements for upsized load, sites with dual fiber paths, and sites in jurisdictions with predictable permitting timelines. Where the spread compresses fastest: sites that require queue-jumping, water builds, or fiber laterals, where the conversion math gets dragged down by execution risk and 24+ month timelines.
