Microsoft signs $9.7B deal with BTC miner IREN

Microsoft The $9.7 billion contract with a Texas miner reveals the new math that is pushing the crypto infrastructure toward AI, and what it means for the networks left behind.

IREN The announcement of November 3 collapses two transactions in a single strategic pivot. The first is a five-year, $9.7 billion cloud services contract with Microsoft, while the second is a $5.8 billion equipment deal with Dell to source. Nvidia GB300 systems.

The combined $15.5 billion commitment converts approximately 200 megawatts of critical IT capacity at IREN’s Childress, Texas campus from potential Bitcoin contracted GPU hosting mining infrastructure for Microsoft’s AI workloads.

Microsoft included a 20% prepayment, about $1.9 billion in advance, signaling the urgency around a capacity constraint that the company’s CFO has marked as extending at least until mid-2026.

The structure of the deal makes explicit what the miners have calculated in silence. At the current price of the hash forward, each megawatt dedicated to hosting AI it generates about $500,000 to $600,000 more in gross revenue per year than the same megawatt hashing Bitcoin.

That margin, an increase of about 80%, creates the economic logic that drives the reallocation of the most significant infrastructure in the history of crypto.

The math of revenue that breaks down

Bitcoin mining at 20 joules per terahash efficiency generates about $0.79 million per megawatt-hour when the hash price is $43.34 per petahash per day.

Even at $55 per petahash, which requires sustained Bitcoin price appreciation or tax activity, mining revenues only scale to $1.00 million per megawatt-year.

The AI ​​host, by contrast, is based at about $1.45 million per megawatt-year Scientific Core disclosed contracts with CoreWeave. This equates to $8.7 billion in cumulative revenue on approximately 500 megawatts over a 12-year period.

The crossing point where Bitcoin mining matches the AI ​​host economy is between $60 and $70 per petahash per day for a 20 joule-per-terahash fleet.

For the majority of the mining industry, which manages equipment from 20 to 25 joules, the price of the hash would need to rise from 40% to 60% from the current levels to make Bitcoin mining as lucrative as the contracted GPU hosting.

That scenario requires either a strong Bitcoin price rally, sustained rate pressure, or a significant drop in the network’s hashrate, none of which operators can bank on when Microsoft offers guaranteed, dollar-denominated revenues starting immediately.

Because Texas won the bid

IREN’s Childress campus is located on the ERCOT grid, where wholesale energy prices were expected to average $27 to $34 per megawatt-hour in 2025.

These numbers are lower than the US national average of nearly $40 and significantly cheaper than those in PJM or other eastern grids, where data center demand has driven capacity auction prices to regulatory limits.

Texas is benefiting from rapidly expanding solar and wind, keeping baseline energy costs competitive. But ERCOT’s volatility creates additional revenue streams that amplify the economic case for flexible computing infrastructure.

Platforms of revolt demonstrated this dynamic in August 2023 when it collected $31.7 million in demand response and reduction credits by closing mining operations during peak price events.

The same flexibility applies to hosting AI if the contract structures are structured as a pass: operators can limit operations during extreme price events, collect ancillary service payments, and resume operations when prices normalize.

PJM’s capacity market tells the other side of the story. Data center demand has pushed capacity prices to administrative caps for the delivery years ahead, signaling tight supply and multi-year queues for interconnection.

ERCOT operates an energy-only market without capacity building, which means interconnection times are shortened and operators face fewer regulatory hurdles.

IREN’s 750 megawatt campus already has the energy infrastructure in place; Converting from mining to AI hosting requires swapping out ASICs for GPUs and upgrading cooling systems rather than securing new throughput.

The implementation timeline and what will happen to the miners

Data Center Dynamics has marked IREN’s “Horizon 1” module in the second half of 2025: a 75-megawatt, liquid-cooled chip-directed installation designed for Blackwell-class GPUs.

Reports have confirmed that the phased implementation will extend until 2026, scaling to around 200 megawatts of critical IT load.

This timeline aligns precisely with Microsoft’s capacity crisis in mid-2026, making third-party capacity immediately valuable, even if hyperscale buildouts eventually arrive.

The 20% prepayment works as scheduling insurance. Microsoft is locking in delivery milestones and sharing some of the supply chain risk inherent in sourcing Nvidia’s GB300 systems, which remain supply-constrained.

The prepayment structure suggests that Microsoft is assessing the certainty of potentially cheaper capacity expectations in 2027 or 2028.

If the 200 megawatts of IREN represent the advance of a wider reallocation, the growth of the hashrate of the network moderates when the capacity comes out of Bitcoin mining. The network recently surpassed one zettahash per second, reflecting the steady increase in difficulty.

Removing even 500 to 1,000 megawatts from the global mining base, a plausible scenario if Core Scientific’s 500 megawatts combined with IREN’s pivot and similar moves by other miners, would slow hashrate growth and provide marginal relief on the hash price for the remaining operators.

The difficulty adjusts every 2,016 blocks based on the current hashrate. If the network’s aggregate capacity decreases or stops growing rapidly, each remaining petahash earns a little more Bitcoin.

High-efficiency fleets with hash rates below 20 joules per terahash benefit more because their cost structures can support lower hash rate levels than older hardware.

Treasury pressure makes it easier for miners to successfully pivot capacity to multi-year, dollar-denominated hosting contracts.

Bitcoin mining revenue fluctuates with price, difficulty and fee activity; operators with thin budgets often face forced sales during downturns to cover fixed costs.

Core Scientific’s 12-year contracts with CoreWeave de-link cash flow from the spot Bitcoin market, converting volatile revenue into predictable service fees.

Microsoft’s IREN contract achieves the same result: financial performance depends on uptime and operational efficiency instead of whether Bitcoin trades at $60,000 or $30,000.

This de-linking has second-order effects on the Bitcoin spot market. Miners represent a structural source of selling pressure because they must convert some mined coins into fiat to cover electricity and debt service.

The reduction in the mining base eliminates that incremental sale, marginally tightening Bitcoin’s supply-demand balance. If the trend scales to several gigawatts in the next 18 months, the cumulative impact on miner-driven sales becomes material.

The risk scenario that reverses the trade

The hash price does not remain static. If the price of Bitcoin recovers sharply while the growth of the hashrate of the network moderates due to the reallocation of capacity, the price of the hashprice could climb above $60 per petahash per day and approach the levels where the rival mining AI hosting economy.

Add in a spike in fees from network congestion, and the revenue gap narrows further. Miners who have locked capacity into multi-year hosting contracts cannot easily pivot back, as they have committed to hardware purchase budgets, site designs and customer SLAs around the GPU infrastructure.

Supply chain risk lies on the other side. Nvidia’s GB300 systems remain limited, liquid cooling components make turnaround times measured in quarters, and substation work can delay site preparation.

If Childress’ implementation of IREN slips past mid-2026, the revenue guarantee from Microsoft will lose some of its immediate value.

Microsoft needs capabilities when its internal constraints bite hardest, not six months later when the company’s buildouts are online.

The structure of the contract introduces another variable. The figure of $1.45 million per megawatt-year represents service revenue, and margins depend on SLA performance, availability guarantees, and if energy costs pass cleanly.

Some host contracts include take-or-pay commitments that protect the operator from curtailment losses but limit ancillary services.

Others leave the operator vulnerable to ERCOT price fluctuations, creating margin risk if extreme weather drives energy costs above pass-through thresholds.

What Microsoft actually bought

IREN and Core Scientific are not outliers, but rather the visible side of a re-optimization that is playing out throughout the publicly traded mining sector.

Miners with access to cheap energy, ERCOT or similar flexible networks, and existing infrastructure can launch hyperscalers on a capacity that is faster and cheaper to enable than greenfield data center construction.

The limiting factors are cooling capacity, direct-to-chip liquid cooling requires a different infrastructure than air-cooled ASICs, and the ability to ensure GPU supply.

What Microsoft bought from IREN wasn’t just 200 megawatts of GPU capacity. It bought the certainty of delivery during a limited window when every competitor faces the same bottlenecks.

The prepayment and five-year term signal that hyperscalers value speed and reliability enough to pay premiums over what future capacity might cost.

For miners, this first represents an arbitrage opportunity: redistribute megawatts toward the highest use case while the hash price remains suppressed, then reassess when the next Bitcoin cycle or the fee environment changes the math.

Trading works until it doesn’t, and the timing of that reversal will determine which operators have captured the best years of AI infrastructure scarcity and which are shut down just before the mining economy recovers.