Finance

FTX-Backed Bitcoin Miner Pivots to AI in Hunt for Power

Marcus SterlingPublished 57m ago3 min readBased on 7 sources
Reading level
FTX-Backed Bitcoin Miner Pivots to AI in Hunt for Power
source:prnewswire.com

SWI Group has bought a majority stake in Genesis Digital Assets, a bitcoin miner once backed by FTX, to serve artificial-intelligence customers. The switch was reported on Oct. 12, 2026. The Wall Street Journal

Genesis Digital Assets is a Cyprus-registered bitcoin miner. Sam Bankman-Fried's hedge fund put $1 billion into the company, a bet detailed in January 2023 reporting. The Wall Street Journal The October 2026 account called that mining bet one of his "most reckless investments."

Why power is the deal

Power procurement explains the bid. Morgan Stanley analysts forecast 68 gigawatts of U.S. data-center power demand between 2026 and 2028, about 38 gigawatts more than projected supply. Reuters Breakingviews A gigawatt is a billion watts, roughly the output of a large power plant. Data centers still account for just 1% of the U.S. commercial real estate market. Reuters

A wider switch from crypto to AI

The Genesis shift fits a wider redeployment. Many bitcoin miners have begun repurposing parts of their operations into data centers to tap the AI boom. Bitfarms Ltd., a Canadian miner, plans to wind down its bitcoin-mining operations over the next two years and focus on AI data centers. Bloomberg Hyperscale Data plans to form Patriot BTC to separate its bitcoin-mining operations from its Michigan AI data center strategy.

What it means for the money

The broader context here is connection to the grid, not chips. Miners control energized land, substations, transmission access and operating permits. Hyperscalers, the large cloud providers, and neoclouds, newer AI-focused cloud firms, need ready buildings with power within a two to three year delivery window. A miner with curtailed load, or power that can be freed up, and an approved load study, or utility sign-off on how much power can be drawn, can shorten that wait. For lenders, that swaps merchant hashprice exposure, meaning income tied to volatile crypto-mining prices, for contracted lease or hosting cash flow from steadier tenants, but with heavier upfront spending on cooling, backup systems and fiber.

Looking at what this means for valuation, the trade is scarcity of deliverable megawatts. The 68 gigawatt demand figure against a 38 gigawatt shortfall frames why AI tenants will pay for used power sites. Miners price that option in enterprise value per megawatt, or company value divided by power capacity, while data-center REITs, or property owners, and hyperscalers price it in speed to revenue and avoided queue risk. The gap between those two pricing models drives M&A and majority-stake deals of the SWI-Genesis type.

In my view, skepticism is warranted on duration. Repurposing is not a like-for-like conversion. Bitcoin sites were built for interruptible load and low-tier uptime, meaning they could shut off easily and did not need to stay online every second. AI training and inference require Tier III equivalent concurrency, or backup systems that keep computers running during maintenance, closed-loop liquid cooling, and dense fiber backhaul, or high-capacity internet links. Retrofits can strand capital if utility upgrades slip or if tenants insist on new-build specifications. The Bitfarms and Hyperscale structures, full pivot versus corporate separation, reflect different answers to that spending and counterparty risk. Separation isolates volatile mining income from data-center financing. A full wind-down bets that contracted AI rents will clear the retrofit hurdle rate, or earn enough to justify the upgrade cost.

For investors tracking the FTX estate overhang, the lesson is recovery through asset reclassification. A distressed energy position bought for proof-of-work, or running computers to validate crypto, becomes an AI infrastructure position rented for leased capacity. The electrons are the same. The contract length, credit quality and capital intensity are not.