Bitcoin Miners Are Becoming AI Landlords — And DePIN Wants In

The Pivot: From Mining Rigs to GPU Racks

Since the 2024 Bitcoin halving compressed mining margins, a number of publicly traded miners have renegotiated their core business around hosting AI and high-performance-computing workloads rather than running ASICs themselves. Core Scientific's colocation arrangement with CoreWeave, first signed in 2024 and expanded in subsequent amendments, is the clearest example: the company redirected megawatts originally built for Bitcoin mining toward GPU hosting, with cumulative contract value reported well into the billions of dollars over the life of the deal.

Cipher Mining, TeraWulf, and IREN have each announced comparable arrangements — long-duration hosting or colocation agreements with AI infrastructure firms, in some cases backed by guarantees from larger technology companies. The specific dollar figures vary by report and amendment, but the direction is consistent: miners are treating their interconnection rights and substation capacity as the product, with Bitcoin mining increasingly positioned as a fallback load that can be curtailed when a GPU tenant needs the power instead.

The suggested framing — a miner 'ditching its site' for a roughly $1.2 billion AI deal — matches the shape of these announcements even where exact terms differ. What is grounded and consistent across the sector is the economic logic: a megawatt sold into an AI hosting contract has been reported to command a materially higher rate than the same megawatt run through mining rigs at current network difficulty.

Why Energy Is the New Scarce Resource

Bitcoin mining and AI training share one hard constraint: both need cheap, reliable, already-interconnected power at scale, and grid interconnection queues in most U.S. regions now run multiple years. Miners who built or acquired sites over the past decade sit on a scarce asset — energized capacity — that AI infrastructure developers can lease far faster than they can build new substations from scratch.

This is the real basis for the 'energy-backed compute' narrative: the value isn't in the ASIC or the GPU, it's in the interconnection agreement, the power purchase contract, and the physical site. That is also why the DePIN and AI-crypto narratives converge here rather than compete — a decentralized compute network still needs someone to own or lease the underlying power and hardware, whether that someone is a Bitcoin miner repurposing a facility or a data-center operator renting rack space to a token-incentivized network.

DePIN's Counter-Bet: Decentralized Compute Networks

While centralized miners are signing single-tenant hosting deals with hyperscalers and AI labs, DePIN (decentralized physical infrastructure) networks are pitching the opposite model: aggregate idle or underutilized GPU capacity from many independent operators and route it to AI workloads through a token-incentivized marketplace. Bittensor positions itself as a decentralized network for machine-learning model training and inference, coordinating subnets of contributors who are rewarded in TAO for useful computational output. Render Network does the equivalent for GPU rendering and, increasingly, broader compute tasks, paying node operators in RENDER tokens.

Networks like Akash Network and io.net extend the same idea to general-purpose GPU compute marketplaces, competing on price and availability against centralized cloud providers. The pitch to AI developers is lower cost and no long-term lock-in; the pitch to compute suppliers — including, potentially, miners with spare capacity — is a way to monetize hardware without signing a multi-year exclusive contract with a single hyperscaler.

The tension is real: centralized hosting deals currently move far more capital and megawattage than DePIN compute marketplaces do, and enterprise AI labs training frontier models have shown limited appetite so far for decentralized, non-guaranteed compute pools. DePIN's stronger case today is at the inference and smaller-model-training end of the market, not frontier training runs that require tightly coupled, low-latency GPU clusters.

The Security Layer Nobody Prices In

The infrastructure convergence between mining, AI hosting, and DePIN doesn't remove the older security problems of crypto — it just adds new balance sheets worth attacking. A recent case reported by Cointelegraph illustrates the pattern: a hacker who drained a victim's Bitcoin, reportedly tied to a compromised Coldcard hardware wallet setup, routed the stolen funds through THORChain to swap them into Ether. THORChain's cross-chain swap design, which lets users move value between chains like Bitcoin and Ethereum without a centralized exchange or KYC checkpoint, has repeatedly shown up in laundering trails following other exploits over the past several years.

The relevance to the AI-compute pivot is structural rather than direct: as miners and DePIN networks hold larger treasuries denominated in BTC, ETH, or native tokens to fund infrastructure buildouts and reward pools, they inherit the same custody and cross-chain exposure that made the Coldcard-linked theft possible. A treasury attack on a miner mid-negotiation on a nine- or ten-figure hosting deal, or on a DePIN protocol's reward-distribution multisig, carries outsized consequences precisely because the underlying business has scaled up.

Key Risks to the Convergence Thesis

Additional Resources