Riot Platforms signed a $9.1 billion AI deal this week. It's retrofitting a bitcoin mine, not building a new data center.

Riot Platforms disclosed a 20-year, $9.1 billion data-center lease on August 10 — its Rockdale, Texas campus, previously a bitcoin mining operation, will supply 191 megawatts of capacity to what multiple outlets reported as Anthropic. It's not a new build. It's a retrofit, and it's the fourth deal of its kind announced in three weeks.

What's actually happening here?

Bitcoin miners are converting existing mining sites into AI data centers, and doing it at a pace that's turned a niche pivot into a real construction category. Since July 20, four public miners have signed multibillion-dollar AI infrastructure deals:

Cumulative AI and HPC contracts across the public bitcoin mining sector now exceed $70 billion. These aren't paper deals — Hut 8's contracted AI portfolio alone has grown to $26.6 billion, and several of these companies have stopped describing themselves as miners in their own investor materials.

Why convert a mine instead of building fresh?

Because a mining site comes with something a raw parcel doesn't: an energized grid connection. TeraWulf has disclosed that bringing a site to AI-ready standard runs roughly $8 million to $11 million per megawatt — liquid cooling retrofit, electrical service upgrades, and redundant power systems — against the roughly $700,000 to $1 million per megawatt the original mining buildout cost. That's eight to fifteen times the price per megawatt.

The retrofit is expensive because the two workloads have nothing in common electrically. ASIC mining rigs are air-cooled and spread load thin across a warehouse floor. AI GPU racks can pull past 130 kilowatts each and need liquid delivered directly to the chip. Every mining conversion means ripping out the ASICs, re-running power distribution, and adding a cooling loop the site was never built for.

What the higher cost buys is time. A converted mining site can be delivering AI capacity in under a year. A new-build data center in a constrained market — Texas being the clearest example, where the state froze new grid interconnection approvals pending an audit — can spend multiple years waiting in the interconnection queue before it ever pours a foundation. Paying a premium per megawatt to skip that queue is, for a hyperscaler with a capacity deadline, the cheaper option.

Who bids this work?

It's a different trade mix than a greenfield data center. There's comparatively little civil and structural scope — the building, pad, and switchyard already exist. The scope concentrates in two places: electrical contractors doing switchgear and power distribution upgrades to carry GPU-class loads, and mechanical subs installing direct-to-chip or rear-door liquid cooling where the site only ever needed forced air. Reinforcing floors and racking for denser GPU loads is a smaller structural line item, not a new build.

Should a GC or sub go looking for this work?

If your market has retired or struggling crypto mining capacity nearby, yes — worth a call to the site owner or the hyperscaler now leasing it. This is a live, funded project type, not a forecast: four deals in three weeks, $70 billion in contracts, and staged buildout schedules already running through 2028. The retrofit scope is narrower than a ground-up data center, which is exactly why it's a fit for an MEP sub that couldn't touch the civil-heavy greenfield jobs going to the big data-center GCs.

Related: Texas froze new data-center grid approvals in August — the interconnection bottleneck behind why an already-powered mining site is suddenly worth a premium.

If you're bidding electrical or mechanical scope near a bitcoin mining facility, this is the deal type to watch for — the retrofit math above is what a hyperscaler is willing to pay to skip the queue.

Construction AI Brief publishes three times a week. Subscribe at constructionaibrief.com.