By Why Bitcoin Miners Are Betting on AI Data Centers
In its fourth-quarter 2025 earnings call, IREN, a company that spent years building itself into one of the world's larger Bitcoin miners, told investors plainly that Bitcoin is no longer a "long-term strategic focus." That's not a hedge or a diversification footnote. It's the industry's most literal public admission yet of something visible across nearly every major listed mining company right now: Core Scientific, IREN, Riot, TeraWulf, and Hut 8 are in the middle of converting the infrastructure they built to mine Bitcoin into AI data centers, backed by real contracts worth tens of billions of dollars, and they are selling their Bitcoin holdings to help pay for it. This isn't a niche corner of the crypto industry quietly experimenting. It's happening at the largest, most established, publicly traded names in Bitcoin mining, in full view of their own shareholders.
The math that made this an easy decision
The economic logic behind the shift is blunt once you see the comparison. CoreWeave, the AI cloud provider that doesn't mine Bitcoin at all but leases infrastructure from former miners, demonstrated that 10 megawatts of Nvidia H100 GPUs can generate revenue roughly equivalent to 100 megawatts of Bitcoin mining. For a mining company sitting on gigawatts of power capacity and cooling infrastructure it already built, the calculation is close to arithmetic: swap SHA-256 hashing for tensor operations and multiply margins by a factor of ten. Nvidia's own supply constraints reinforce the advantage for whoever moves first. CEO Jensen Huang confirmed in early 2026 that the company's Blackwell architecture was sold out through the middle of the year, with a backlog of 3.6 million units, meaning the miners who secured early GPU allocations, IREN and Core Scientific among them, hold a real hardware advantage over competitors trying to enter later. There's also a financing logic that has nothing to do with Bitcoin's price at all: building a 200-megawatt liquid-cooled GPU site is a hyperscaler-grade construction project, 18 to 30 months of permitting, transformer procurement, water rights, and substation upgrades, the kind of capital expenditure that institutional debt markets will only finance against a contracted future revenue stream, not against a balance sheet full of an asset as volatile as Bitcoin.
There's a squeeze on the mining side of the business making that comparison sharper every quarter, too. Bitcoin's average electricity cost per coin reached roughly $46,426 in March 2026, and by one industry estimate the all-in production cost, including hardware and overhead, hit around $74,300 per coin in January, a number that eats deeply into margin whenever Bitcoin's price sits anywhere close to it. US wholesale electricity prices were projected to average around $51 per megawatt-hour in 2026, continuing to climb as AI data center demand itself competes for the same grid capacity mining rigs depend on. A miner facing rising power costs and a fixed, halving-driven block reward has a straightforward incentive to redirect that same power toward a customer willing to pay a premium for it instead.
The deals, and they are large by any standard
IREN's own pivot is the most complete of the group. In late 2025 it secured a five-year, $9.7 billion AI cloud agreement with Microsoft covering 76,000 Nvidia GB300 GPUs across 200 megawatts at its Childress, Texas campus, projected to generate $1.94 billion in annual recurring revenue at roughly 85% EBITDA margins, with Microsoft providing a $1.9 billion customer prepayment that gave IREN immediate construction capital. Core Scientific, once one of North America's largest Bitcoin miners before a 2022 bankruptcy driven by high power costs and a weak Bitcoin price, is now building six AI data centers leased to CoreWeave over twelve years, with roughly $10 billion in contracted revenue behind the arrangement, funded in part by a $3.3 billion junk bond sale in April 2026. Hut 8 signed a fifteen-year, $9.8 billion lease for a 352-megawatt Texas facility built to Nvidia's own reference architecture. TeraWulf has locked in $12.8 billion in contracted AI and HPC revenue, with 27% of its revenue already coming from that side of the business rather than mining. Riot Platforms moved later than the others, spending most of 2025 defending its position as a pure Bitcoin miner before signing a ten-year lease with AMD in January 2026 for 25 megawatts at its Rockdale, Texas site, expandable to 200. Across the sector, CoinShares estimates listed miners have signed roughly $70 billion in AI and HPC contracts combined, and projects that AI could account for as much as 70% of listed miner revenue by the end of 2026, up from around 30% at the start of the year.
The private key for every Bitcoin wallet on Earth is on this website, even Satoshi's. But even if you try for a million years, you'll never find a funded one.
Try the key collider nowThey're selling the Bitcoin to pay for it
The pivot isn't being funded purely by new debt. Public Bitcoin miners collectively sold an estimated 32,000 BTC in the first quarter of 2026 alone, a striking figure for an industry whose entire original business model was accumulating the asset rather than offloading it, liquidating treasury holdings to help cover the up-front capital a multi-year AI infrastructure contract demands before its revenue starts arriving. Core Scientific sold $175 million worth of Bitcoin the month before its April 2026 bond sale and now holds, according to its own CFO, under 1,000 bitcoin, a remarkable reversal for a company that built its original business entirely around accumulating it. IREN has gone further than any of its peers: it holds zero Bitcoin in treasury, a choice its own executives describe as deliberate strategy rather than financial necessity, betting the company's future entirely on contracted AI revenue rather than Bitcoin's price. Mining companies aren't selling Bitcoin because they've lost faith in it. They're selling it because a fifteen-year AI infrastructure lease requires capital today, and a four-year halving cycle that keeps cutting their mining revenue in half can no longer reliably fund a construction project of that scale on its own.
Not everyone is moving at the same speed
The pivot isn't universal, and the holdouts are informative. Marathon Digital, the largest publicly traded Bitcoin miner by hashrate, has been notably more cautious than its peers, acquiring some AI-relevant facilities and announcing exploratory partnerships without committing to anything at the scale of IREN's or Core Scientific's contracts, a strategic direction the company itself describes as still evolving. Bitfarms took the opposite approach to Riot's late-but-massive shift, rebranding part of its infrastructure business as Keel Infrastructure specifically to signal a deliberate move toward AI hosting rather than treating it as a side project. Smaller players are finding narrower footholds too: Bitdeer secured a $43 million annual recurring revenue AI cloud services agreement, a fraction of the billions involved in IREN's or Hut 8's deals but still a real, contracted business the company didn't have two years ago. What separates the companies moving fastest from the ones holding back seems to come down to how much spare power capacity and site infrastructure they already had sitting underused, and how comfortable each management team was making a multi-decade bet on AI demand continuing at its current pace.
What this is actually doing to Bitcoin's own security
The honest answer requires two things to be true at once, and most coverage of this story picks one and ignores the other. Bitcoin's network hashrate peaked at 1,160 exahashes per second in October 2025 and fell to roughly 850 EH/s by February 2026, a drop of about 27%, before partially recovering to around 1,020 EH/s. The first quarter of 2026 marked the first quarterly hashrate decline in Q1 since 2020, ending five straight years of double-digit growth. On March 21, 2026, mining difficulty dropped 7.76% in a single adjustment, one of the sharpest on record, and a lower difficulty does mechanically lower the computational cost of a hypothetical attack on the network, at least temporarily. That's the real, measurable part of the story. The part that gets lost is that over 1,000 exahashes per second, even after the decline, remains an almost incomprehensible amount of computational commitment, and difficulty adjusts specifically to protect the economics of whichever miners remain, which is exactly the mechanism that's already stabilized the network partway back up. There's also a genuine case that this could make Bitcoin's security more resilient rather than less: publicly listed US miners have historically represented over 40% of global hashrate, and a reduction in their dominance, even one driven by them chasing AI contracts rather than any ideological commitment to decentralization, could leave mining more geographically spread out rather than less, with Paraguay, Ethiopia, and Oman all newly appearing in global hashrate rankings as the balance shifts. There's a countervailing trend worth naming honestly, though: mining pool concentration has risen in parallel, with roughly six pools now controlling around 99% of total hashrate, a centralization vector that exists independently of which individual companies own the machines plugged into those pools. Geographic diversification and pool concentration are pulling in opposite directions at the same time, and it's not obvious which effect dominates the network's real-world resistance to coordinated attack. One industry analysis put the actual stakes about as precisely as the situation deserves: this doesn't mean the network is in danger, but it does mean anyone writing institutional risk assessments for Bitcoin ETF allocations now has to model an asset whose underlying security budget is partially funded by companies that see their own long-term future in a different business entirely.
The stock market already rendered a verdict
Mining company shares significantly outperformed Bitcoin itself through 2026, and the reason why matters more than the fact itself: investors are rewarding these companies for the AI contracts, not for their Bitcoin holdings or mining capacity. That's the same pattern this blog has already documented in a different corner of the market, institutional capital choosing an asset for reasons that have little to do with the ideology that originally built it. A pension fund buying IREN stock today is underwriting a Microsoft AI data center lease with a Bitcoin-mining company's name still attached to it, in much the same way an ETF investor buying Bitcoin exposure is underwriting a custodian's balance sheet rather than holding a key. The label survives. The actual business underneath it has, in several of these companies, largely moved on.
There's a structural reason this pivot is unlikely to reverse quickly even if Bitcoin's price rallies hard from here, and it changes a dynamic the market has relied on for over a decade. Historically, a rising Bitcoin price pulled idle hashrate back online as mining became profitable again, a natural feedback loop that helped stabilize the network. A five-year hyperscaler lease doesn't work that way. It can't simply be paused the next time Bitcoin rallies, because the contract, and the debt raised against it, assumes that capacity keeps serving AI customers for the life of the agreement regardless of what Bitcoin's price does in the meantime. As of early September 2026, Bitcoin's hashrate had spent 316 consecutive days below its October 2025 peak, the longest stretch of its kind in a decade, a direct consequence of power that used to flow back into mining the moment price recovered now sitting locked into contracts that don't care what Bitcoin is worth this quarter.
For more on the parallel shift in who holds Bitcoin itself and why, see why Bitcoin ETFs brought in a different asset class, not more believers. For the mechanics of what a hashrate decline and difficulty adjustment actually change about Bitcoin's attack resistance, see how mining pool concentration affects network security, and for the environmental side of the same energy story, see Bitcoin's energy debate. For how mining itself actually works, and what these companies are stepping away from, see how Bitcoin mining works. For a real-time example of a miner running the AI pivot and continued Bitcoin accumulation at the same time, see MARA's $100 million Bitcoin purchase amid its AI expansion.