From Hashrate to GPU Racks: How Bitcoin’s Mining Giants Are Quietly Becoming America’s AI Cloud

IT 6 Minutes

There is a number that explains almost everything happening right now in cryptocurrency’s most capital-intensive industry. It is $27.7.

That figure — the hashprice, or what Bitcoin miners earned per petahash of computing power per day in June 2026 — represents an all-time low. It is the compressed margin of an industry squeezed from every direction at once: a halving that cut block rewards in half, a surge in network difficulty as newer machines came online, and a Bitcoin price that, for all its autumn resilience, has not moved enough to rescue the economics of bulk mining at scale.

The result is a transformation that may be the most significant infrastructure story in crypto this year — and one that most retail investors have largely missed. Across publicly listed Bitcoin miners, the share of revenue coming from something other than mining Bitcoin has gone from a curiosity to a majority. And the thing they’re pivoting to is artificial intelligence.

The Hashprice Collapse That Changed the Calculation

Bitcoin mining economics have always been cyclical, but 2026 delivered a confluence of pressures that compressed margins to a degree the industry had never seen. When the network’s fourth halving reduced the block subsidy from 6.25 BTC to 3.125 BTC in April 2024, miners absorbed a 50% cut in their primary revenue stream overnight. The hope — shared by most in the industry — was that a rising Bitcoin price would compensate. In some months it has. In June 2026, it didn’t.

According to the CoinShares Bitcoin Mining Report for Q2 2026, hashprice hit $27.7 per petahash per day — the lowest on record. For large public miners operating at scale, that figure translates directly to a business that no longer makes sense at the same margins as before. Capital that once bought ASIC miners is now buying something else: GPU clusters for AI workloads.

The shift, once it began, moved quickly. Listed public miners collectively cut spending on new Bitcoin mining hardware by approximately $1.5 billion in the first half of 2026 compared to the same period a year earlier, according to CoinShares data. At least 35 exahashes per second of public-miner hashrate is scheduled to leave the Bitcoin network — from companies including IREN and Cipher — as facilities are repurposed or decommissioned.

What a “Neocloud” Actually Is — and Why Miners Keep Becoming Them

The term neocloud has become industry shorthand for a new category of cloud provider: companies that rent GPU computing capacity to AI developers, bundling raw graphics processing units with the data center space, networking, and software that AI training and inference requires. Unlike hyperscalers such as Amazon Web Services or Microsoft Azure — which build their own infrastructure — neoclouds often lease capacity to AI startups, research labs, and enterprise clients who need GPU access without the capital expenditure of building it themselves.

Three companies have emerged as the clearest examples of the neocloud model in crypto’s orbit: CoreWeave, Nebius, and IREN. CoreWeave’s roots in crypto date to 2016, when it acquired GPUs to mine Ethereum — a connection the company now treats as ancient history, but one that explains why it had the institutional knowledge to build large-scale GPU infrastructure before AI made it valuable. When Ethereum ended proof-of-work mining in September 2022, CoreWeave pivoted entirely. Its revenue for the quarter ended June 30, 2026 reached $2.575 billion, though the company is not yet profitable: depreciation, amortization, and net interest expense of roughly $2 billion against that revenue produced a net loss of $626 million for the quarter.

The critical distinction that the industry often glosses over: Bitcoin-mining machines themselves do not become AI GPUs. The pivot involves repurposing data center space, power infrastructure, and cooling systems — assets that transfer — while the mining hardware itself is sold, wound down, or scrapped. What miners bring to the AI cloud is land, power, and the organizational capacity to operate large facilities. The GPUs come separately, at significant capital cost.

The Deals That Rewrote the Revenue Tables

The scale of the contracts being signed in this space is difficult to overstate. Two deals in particular have defined the conversation in 2026.

In November 2025, IREN announced a five-year GPU cloud services agreement with Microsoft valued at approximately $9.7 billion, with a 20% prepayment built into the structure. To fulfill the contract, IREN planned roughly $5.8 billion in GPU and equipment purchases — a capital commitment that dwarfs anything the company spent during its Bitcoin-mining years. By the fiscal year ending June 30, 2026, IREN had begun shutting down Bitcoin-mining equipment in earnest and shifting power and data center capacity toward AI cloud services, with substantial completion of the pivot targeted for December 31, 2026. For the first time in a single quarter, IREN’s AI cloud revenue exceeded its Bitcoin mining revenue — a 51%-to-49% split that marks a genuine inflection point.

Meanwhile, TeraWulf signed a 20-year infrastructure lease with Anthropic — the AI safety company behind the Claude family of models — covering approximately 401 megawatts of critical IT load at a planned campus in Kentucky. Unlike IREN’s managed GPU cloud service, TeraWulf’s arrangement is a lease of infrastructure: Anthropic will operate its own equipment inside the facility. Initial capacity is expected in the second half of 2027, with full capacity by early 2028. In Q2 2026, TeraWulf already derived 71% of its revenue from AI and high-performance computing leases — the highest non-mining share of any major public miner.

And at Core Scientific, the transformation has gone furthest of all: 87% of Q2 2026 revenue came from sources other than Bitcoin mining, primarily colocation services for third-party AI operators. Bitcoin mining, once the company’s entire identity, now accounts for roughly 13% of what it earns.

Why the Economics Work — and Where They Break Down

The core appeal of AI hosting contracts for former miners is straightforward: predictability. An AI infrastructure lease pays fixed rent set in advance, insulated from Bitcoin price swings, network difficulty adjustments, or the quarterly lottery of block rewards. For a company running hundreds of megawatts of power-hungry infrastructure, the ability to underwrite that cost against a known revenue stream is transformative for both operations and financing.

The sector has more than 4 gigawatts of power contracted to AI workloads, according to Q2 2026 CoinShares data, though the gap between contracted and actual billing remains wide: only around 550 megawatts was actually billing as of mid-year. The total backlog across the sector exceeds $100 billion, but annualized colocation revenue remains under $800 million — a reminder that signed contracts and collected revenue are very different numbers.

The risks are real. CoreWeave’s quarterly loss demonstrates that building out AI cloud infrastructure requires enormous upfront capital — GPUs, networking, power upgrades — before a dollar of that contracted revenue arrives. Companies that overextend on commitments before securing the equipment and financing to meet them face genuine distress. And the AI cloud market is not a captive one: hyperscalers are expanding their own GPU capacity aggressively, and the price of rented AI compute has been falling across the industry as supply catches up with the demand surge of 2024 and 2025.

What This Means for Bitcoin’s Network

The departure of major public miners from the Bitcoin hashrate has implications for the network itself. When at least 35 exahashes per second exit the network — as currently scheduled — the result is a reduction in total mining difficulty, which in turn makes it temporarily easier for remaining miners to earn block rewards. This is not catastrophic; Bitcoin’s difficulty adjustment algorithm is designed exactly for this kind of shift. But it does mean that the composition of the mining industry is changing: large, publicly listed companies with access to capital markets are increasingly stepping back, while smaller, more nimble operators and mining pools with lower cost structures fill the gap.

It also raises longer-term questions about hashrate concentration and the geographic distribution of Bitcoin’s security. If the most capital-intensive mining operators are redirecting power and data center assets toward AI, the profile of who secures the Bitcoin network a decade from now may look quite different from today.

What This Means for You

If you hold shares in any of the publicly listed Bitcoin mining companies — Core Scientific, IREN, TeraWulf, Riot, Hut 8, or Cipher — you are already invested in an AI infrastructure thesis, whether or not you intended to be. The revenue tables published in Q2 2026 earnings reports make that plain. Evaluating these companies purely on hashrate, mining efficiency, or Bitcoin holdings is increasingly the wrong lens.

For crypto investors more broadly, the pivot reflects a structural truth about the industry’s maturation: the most valuable asset in the crypto mining buildout was never the mining software. It was the power infrastructure, the data center real estate, and the operational expertise to run large-scale compute facilities reliably. Those assets turn out to be exactly what AI needs in 2026. The companies that built them for Bitcoin are now leasing them to the companies building the next generation of artificial intelligence — and the contracts they’re signing suggest that both sides consider it a very good deal.

The question for the next twelve months is whether the backlog converts to revenue fast enough to cover the capital that has already been deployed. The deals are real. The facilities are being built. Whether the economics justify the investment — at CoreWeave’s scale of losses, at IREN’s $5.8 billion equipment commitment, at TeraWulf’s 20-year lease horizon — is a question that will only be answered by time and by the continued appetite of AI companies to pay for compute they cannot yet build fast enough themselves.

Sources:

Unknown's avatar

Published by jamrockbrown

Published

Leave a Reply