The crypto mining gold rush just found a new vein, and it's running inference loops instead of SHA-256 hashes.

The Summary

The Signal

IREN isn't abandoning Bitcoin mining because it stopped working. They're abandoning it because they did the math. Bernstein's model shows mining revenue declining over time while AI cloud revenue scales to $3.7 billion. That's not a hedge. That's a full pivot.

The infrastructure play here is elegant. Bitcoin miners spent years optimizing for one thing: buying power cheap and using it 24/7. They built data centers in places with stranded energy, negotiated bulk rates, and got very good at cooling dense hardware. The Microsoft deal and GPU expansion means IREN is swapping ASICs for GPUs without changing the underlying economics.

"The same cheap power that made Bitcoin mining profitable at scale now makes AI inference profitable at scale."

What changed is the customer. Bitcoin mining is a commodity market with razor-thin margins and halvings that cut revenue in half every four years. AI cloud services sell to enterprises with multi-year contracts and predictable demand curves. IREN isn't chasing a better token. They're chasing a better business model.

This matters because IREN isn't alone. Every Bitcoin miner with cheap power and data center expertise is looking at the same spreadsheet. The buildout for Web4 needs compute infrastructure, and the companies best positioned to deliver it might be the ones who spent the last decade mining Web3 assets. The skillset transfers. The customer base doesn't.

The Implication

Watch for more miners to follow. If IREN's bet pays off, every public Bitcoin mining company will face the same analyst pressure: why are you still mining when you could be renting compute to Microsoft? The irony is that the infrastructure Web3 built to decentralize money might end up centralizing AI inference. Cheap power doesn't care what ideology it serves.

For anyone building agents or training models, this expands the cloud provider map. IREN and companies like it aren't AWS, but they might be cheaper and faster for specific workloads. The next six months will show whether ex-miners can compete on latency and reliability, or if they're just power arbitrageurs in new clothing.

Sources

RWA Times | CoinTelegraph