AI just made Bitcoin mining investable again
Four ways miners are winning from it, and only one makes the news.
In the past fortnight two Bitcoin miners signed long-term leases with AI companies. CleanSpark took a 20-year deal worth $6.6 billion in Georgia. TeraWulf took $19 billion in contracted revenue, building 401 MW for Anthropic in Kentucky.
Every outlet covered them the same way: another miner pivots to AI that is good for shareholders (it is).
What I have not seen anyone price is what the grid gives up. A Bitcoin miner can power down in seconds. An AI data centre cannot power down at all without breaking its contracts. So every megawatt that converts is a megawatt the grid loses as a shock absorber, at the moment grid operators have started paying for exactly that flexibility.
That tension runs underneath everything I am watching in this industry. Only one of the four paths open to miners makes the news, and it is the one that sells the flexibility away.
AI is the single biggest megatrend in the world right now, and it has a problem that money alone can’t solve. It needs expertise. It needs people who already know how to secure power at scale, procure and permit sites, build them quickly, manage energy contracts, stabilize the grids they draw from, and keep hundreds of megawatts of dense computing cool.
Where does that expertise live? Turns out, mostly in Bitcoin mining. Miners have spent a decade solving exactly these problems, under conditions AI companies have never faced: brutal margins, hostile media, and a block reward that halves every 4 years. When your survival depends on finding the world’s cheapest electron, you get very good at energy.
AI companies know this, and they are prepared to throw money at it. Requests in ERCOT’s connection queue have grown almost 300% in a year, nearly all of it data centers. MARA’s CEO Fred Thiel, announcing a $1.5 billion power acquisition in April: “Power is the scarce input in AI.”
So opportunities are flowing to people in Bitcoin mining in abundance right now. Only 1 of the paths makes the news: the complete pivot. Underneath it, I’m watching at least 4 distinct ways miners are winning:
1. The complete pivot. The TeraWulf and CleanSpark path: become an AI data center company. It brings the headline deals and the headline multiples, and there is no way back to Bitcoin mining once the leases are signed. It is also the only one of the four that converts a flexible load into a firm one, which is what the grid loses in the trade.
2. The service provider. An AI data center is a firm load: it can’t switch off without breaking contracts. Bitcoin mining is the only large load that can power down in seconds. That makes flexibility a sellable service, in 2 directions. To the grid, which needs shock absorbers (Denmark has begun rewarding flexible consumption rather than merely tolerating it). And to the AI operators themselves: co-locate flexible mining behind the meter, and the data center can promise the grid a steadiness it couldn’t otherwise offer.
3. The expertise provider. Cooling, heat dissipation and reuse, site procurement and site builds, energy management: every one of these is a niche where mining teams are selling hard-won knowledge to AI developers, mostly in deals you’ll never read about.
4. The hedge. MARA’s path: it just agreed to pay up to $600 million for a Texas site with permitted grid rights approaching 2 GW - mine Bitcoin when it pays best, serve AI load otherwise. The optionality itself is the asset.
What I’m seeing is simple: Bitcoin mining spent 10 years doing proof of work on the hardest problems in energy, and a 2nd industry just started verifying it.
Industry Convergence = Exit Liquidity
The rest of this letter covers information relevant to investors including why we think Bitcoin mining companies now have stronger valuation uplift potential, and why we’re reconsidering our 2022 decision not to take equity positions in Bitcoin mining companies.
