For most of the last decade, the business of Bitcoin mining sounded simple enough. Find cheap electricity, fill a building with ASICs, keep the machines running, and hope the Bitcoin they produce is worth more than the power, hardware, financing, and maintenance needed to create it. The machines got faster, the sites got larger, and entire companies were built around squeezing a little more profit out of every megawatt.
In 2026, that equation is being rewritten by an industry that does not care about Bitcoin at all. Artificial intelligence companies need enormous amounts of power for GPU clusters, and they are willing to sign contracts that can stretch for fifteen or twenty years. That has forced miners to look at the same substations, grid connections, cooling systems, and data center campuses they spent years building and ask a question that would have sounded strange during the last crypto boom: is this electricity now worth more running AI than mining Bitcoin?
For some companies, the answer is already yes.
Access without medium partner: Bitcoin is getting affected by AI
IREN has begun decommissioning Bitcoin mining hardware and reallocating power and data center capacity toward AI Cloud Services. TeraWulf has repurposed mining infrastructure and signed a twenty year lease with Anthropic for roughly 401 megawatts of critical IT load. Hut 8, a company long associated with Bitcoin mining, is now developing enormous purpose built AI campuses, including Beacon Point in Texas, where 704 megawatts of IT capacity are under long term lease with a combined base term contract value of $19.6 billion.
The shift is easy to describe as miners abandoning Bitcoin for AI, but that misses the more important story. The ASICs were never the rarest asset. The real prize was the electricity around them.
Bitcoin mining was always a power business wearing a computing costume
A Bitcoin mine looks like a computer business because the most visible objects are rows of machines. Underneath those machines is an infrastructure business. An ASIC only earns money when the company can feed it electricity cheaply enough and continuously enough to stay competitive with miners everywhere else in the world.
That requirement forced mining companies to become unusually skilled at things most software companies rarely think about. They learned how to find land near substations, negotiate utility access, secure interconnection agreements, build electrical distribution, manage large heat loads, work with local governments, and operate industrial facilities around the clock. They also learned how to move quickly into places where hundreds of megawatts might be available before somebody else claimed them.
Those skills suddenly overlap with one of the hardest problems in artificial intelligence. AI companies can raise billions of dollars and order GPUs, but the GPUs are useless without a place to plug them in. Large clusters need substations, transformers, cooling, networking, fiber, backup systems, permits, land, and a grid capable of delivering enormous quantities of power reliably.
That is why the mining industry’s old infrastructure is being repriced. Bitcoin miners thought they were securing electricity for hashing. AI companies look at the same access to power and see the foundation of a long term compute business.
IREN is the clearest example because the ASICs are actually being removed
It is important to separate two stories that are often blended together. Some former Bitcoin mining companies are building new AI campuses from the ground up, while others are physically replacing mining capacity with AI infrastructure. IREN belongs in the second category.
In its fiscal 2026 filing, IREN said it had begun decommissioning Bitcoin mining hardware and reallocating power and data center capacity toward AI Cloud Services. The company said it aims to substantially complete that transition by December 31, 2026. As of June 30, it still reported about 23.2 exahashes per second of installed Bitcoin mining capacity representing roughly 380 megawatts of data center capacity, but it also said mining should represent a declining share of its business as those sites are repurposed.
The financial incentives are not subtle. IREN disclosed billions of dollars of AI related contracts and financing, including a large Microsoft agreement and additional cloud commitments involving customers such as Perplexity and Together AI. Its fiscal 2026 results also showed AI Cloud Services revenue rising sharply while the company recorded major impairment charges related in part to mining hardware being decommissioned as capacity shifted toward AI.
That is more than a diversification experiment. It is a company accepting that hardware purchased for one of the most famous computing workloads of the last decade can become less valuable than the power and data center capacity surrounding it.
Hut 8 shows how much a megawatt can be worth before a GPU even arrives
Hut 8 tells a different version of the story. Its Beacon Point campus in Nueces County, Texas is not an old Bitcoin mine being converted into AI. Hut 8 explicitly describes it as a purpose built AI data center and says it is not proposed or designed for cryptocurrency mining. That distinction matters because the value of Beacon Point comes from the infrastructure expertise and power development capabilities Hut 8 built over years in the digital asset industry, not from reusing old ASIC halls.
The numbers are enormous. Hut 8 announced a second fifteen year lease for 352 megawatts of IT capacity in July, doubling the customer’s commitment at Beacon Point to 704 megawatts. Each 352 megawatt lease carries a $9.8 billion base term value, bringing the campus total to $19.6 billion before renewal options. Hut 8 says those options could raise the potential campus level contract value to $50.2 billion.
Hut 8 also says its contracted AI data center portfolio now totals 949 megawatts of IT capacity supported by 1,330 megawatts of utility capacity. That is a remarkable transformation for a company whose name was once almost synonymous with Bitcoin mining. The most valuable thing it learned from mining may have been how to secure, finance, and operate large blocks of power.
The latest Anthropic infrastructure agreements make the connection even clearer. Anthropic has signed a $35 billion cloud computing deal with Lambda, and recent reporting links part of that buildout to capacity associated with Hut 8 in Texas and NVIDIA hardware. The chain now stretches from a company with Bitcoin mining roots, to data center infrastructure, to NVIDIA GPUs, to a cloud provider, to one of the world’s largest AI labs.
TeraWulf is signing contracts that Bitcoin mining could never imitate
TeraWulf is making a similar calculation, but its most striking number is not megawatts. It is time.
In July, the company signed a twenty year lease with Anthropic for approximately 401 megawatts of critical IT load at its Justified Data Campus in Kentucky. TeraWulf said the agreement is expected to generate roughly $19 billion of contracted revenue over the initial term, with additional extension options available later. The company has also described high performance computing leasing as its primary growth driver and has already repurposed parts of its mining infrastructure at other sites.
Twenty years changes the nature of the business. Bitcoin mining revenue can move violently with Bitcoin’s price, network difficulty, the block subsidy, transaction fees, hardware efficiency, and energy costs. A large mining operation can look extremely profitable in one market and far less attractive after a price decline or another difficulty increase.
A long term data center lease is a different type of asset. It may require far more capital to build and much higher standards to operate, but it can replace daily exposure to Bitcoin mining economics with contracted payments from an enterprise customer. Investors may give up some upside in exchange for visibility, financing options, and a business whose economics are not recalculated every time the Bitcoin network changes.
More than $70 billion in AI contracts means this is not a side project
By March 2026, public Bitcoin mining companies had announced more than $70 billion in cumulative AI and high performance computing contracts, according to CoinDesk citing CoinShares research. The same analysis argued that some listed miners could eventually earn most of their revenue from AI rather than Bitcoin.
The timing explains why. CoinDesk reported that the average listed miner had spent roughly $79,995 to produce a Bitcoin in the previous quarter while Bitcoin was trading around $70,000 at the time. That figure should not be treated as a universal cash cost for every operator because depreciation, financing, energy hedges, fleet age, and accounting treatment vary significantly between miners, but the direction was clear: mining margins were under pressure.
AI offered a way to monetize the same underlying infrastructure differently. Instead of spending a megawatt on an ASIC that competes against the global Bitcoin network for a probabilistic reward, a company could put that megawatt behind GPU infrastructure and sell compute under a multi year contract.
The electricity did not literally become more expensive. Its opportunity cost changed.
Why one megawatt can earn more when it feeds GPUs
An ASIC is one of the most specialized machines in modern computing. It performs SHA 256 hashing extraordinarily efficiently, which is exactly why it dominates Bitcoin mining. That specialization also gives it almost no alternative use when another workload becomes more profitable.
A GPU data center sits higher in the economic stack. The same infrastructure can support model training, inference, coding agents, enterprise applications, scientific workloads, cloud services, and whatever new AI product arrives next year. A customer may pay for the computing service rather than the raw electricity, which allows more revenue to be layered on top of every megawatt.
That does not automatically make AI more profitable in every location. AI infrastructure is much more expensive to build. CoinDesk’s March analysis cited infrastructure costs of roughly $8 million to $15 million per megawatt for AI compared with around $700,000 to $1 million per megawatt for Bitcoin mining. The return can be more stable and attractive, but the upfront capital required is far larger.
This is why the best positioned miners are not simply companies with cheap warehouses. They are companies with power rights, financing access, strong sites, fiber, utility relationships, and enough engineering capability to turn a mining location into infrastructure a hyperscaler is willing to trust.
A Bitcoin mine is not automatically an AI data center
The easiest version of this story makes conversion sound like replacing one rack with another. In practice, the differences are enormous.
Bitcoin mining is remarkably tolerant of simple infrastructure. If a group of ASICs goes offline, the rest of the network continues operating. Mining traffic does not require the kind of high speed east west networking used in large distributed training clusters. Many mines are built around huge volumes of air cooling and can operate in remote locations as long as they have power and basic connectivity.
Modern AI infrastructure can demand direct liquid cooling, extremely high rack densities, redundant power paths, advanced networking, high capacity fiber, physical security, backup systems, sophisticated building controls, and far tighter uptime guarantees. A frontier AI customer signing a twenty year contract expects a different facility from a miner that can curtail load whenever economics weaken.
This is why the grid connection is valuable but not sufficient. A mining company may own the right raw ingredients and still need billions of dollars to turn them into an AI campus.
Bitcoin and AI also want different kinds of electricity
Bitcoin has one advantage that AI cannot easily copy: geographic flexibility. A miner can operate near remote hydroelectric power, stranded energy, or another cheap source far from a major city because the Bitcoin network does not care where the hashes are produced.
AI has more constraints. Fiber quality matters. Equipment logistics matter. Skilled workers matter. Latency can matter for inference. Large customers may demand redundant connectivity, strict service levels, and easier access for technicians and replacement hardware.
That means some mining sites will never be attractive AI locations. In those places, Bitcoin may remain the better buyer of electricity precisely because it can monetize power that hyperscale data centers cannot use efficiently.
The shift is therefore not “AI replaces Bitcoin mining everywhere.” The more accurate story is that AI is repricing premium mining infrastructure, especially sites with strong grid access, expansion capacity, fiber, and locations suitable for hyperscale development.
The most important asset may now be the substation
The identity of a mining company changes when investors stop asking about ASIC efficiency and start asking about utility capacity.
How many megawatts are controlled? When can they be energized? Is the interconnection agreement secure? Does the site have expansion land? Can it support liquid cooling? How much fiber is available? Who is the tenant, how strong is the tenant’s credit, and how long is the lease?
Those questions sound more like infrastructure finance than cryptocurrency.
The shift is visible in how companies describe themselves. Hut 8 now calls itself an energy infrastructure platform. TeraWulf emphasizes digital infrastructure and high performance computing. IREN describes a three layer AI cloud stack consisting of data centers, compute, and software.
The ASIC fleet used to be the centerpiece. Increasingly, the substation is the strategic asset.
NVIDIA keeps appearing because the AI boom needs places to put the chips
NVIDIA’s role in this transition is easy to underestimate. Selling more GPUs eventually becomes difficult if customers cannot find enough powered data center space to install them.
That creates an incentive for NVIDIA to become involved deeper in the infrastructure chain. The company backs cloud providers, works with data center operators, supports financing arrangements, and participates in capacity deals that help turn electrical infrastructure into places where NVIDIA hardware can actually run.
The Anthropic, Lambda, and Hut 8 relationship is a good example of how the AI economy now works. One company develops infrastructure, another supplies hardware, another packages that hardware into cloud services, and the model company buys the compute.
The bottleneck is no longer only the chip. It is the entire physical system surrounding the chip.
Bitcoin miners happen to own some of the pieces that take longest to build.
This does not mean Bitcoin mining is dying
The strongest version of the headline is also the easiest one to exaggerate.
Bitcoin does not stop working because one mining company removes ASICs. The network adjusts mining difficulty roughly every two weeks so that blocks continue arriving near the target interval even when global hashrate changes. If less efficient miners leave, the remaining miners can capture a larger share of the reward after difficulty adjusts.
There are still valid questions about concentration. If smaller operators leave faster than large infrastructure owners and mining pools, the distribution of hashrate can become less diverse. CoinDesk reported that Bitcoin’s hashrate fell during the first quarter of 2026, the first first quarter decline in six years, while mining economics and the AI pivot were receiving more attention.
But AI is not simply stealing security from Bitcoin one megawatt at a time. It is changing which locations Bitcoin can economically justify using. Mining may become even more concentrated around electricity that is cheap, flexible, remote, or otherwise difficult for an AI data center to monetize.
Bitcoin can still be an excellent buyer of power that other industries do not want.
The problem for miners is that AI now wants the premium sites.
The grid is becoming the real battlefield
Once Bitcoin mining and AI compete for the same high quality electrical infrastructure, the story becomes larger than either industry.
Data centers can require hundreds of megawatts from grids that were never designed for this pace of demand growth. Utilities need new substations and transmission. Communities have to decide how costs are allocated, how water will be managed, what happens during grid stress, and whether the economic benefits justify the infrastructure burden.
Hut 8 has already addressed some of those questions at Beacon Point. The company says required infrastructure costs are being prefunded, the project will use a closed loop cooling design with water sourced outside Nueces County, and battery systems are intended to reduce voltage impacts and help support grid reliability.
Those commitments will matter more as AI campuses grow. A long term contract can make a project financially attractive without automatically making it good for the surrounding grid or community.
The next phase of the AI boom will therefore be negotiated as much in utility offices and state regulators’ hearing rooms as in semiconductor labs.
Bitcoin miners may have been power developers all along
In retrospect, Bitcoin mining gave these companies an unusual education.
They learned how to locate electricity, finance sites, cool dense hardware, negotiate with utilities, build quickly, and monetize a continuous industrial load. Bitcoin was the application that paid for those lessons.
Now a second application has arrived with deeper pockets and longer contracts.
AI labs do not want the ASICs. They want the grid connection, the land, the transformers, the fiber, the cooling, the permitting, and the ability to deliver hundreds of megawatts years faster than a greenfield project starting from nothing.
That is why the industry’s identity is changing so quickly. A company can sell or retire the miners and still keep the most valuable part of what it built.
The machines were replaceable.
The power was not.
AI did not make Bitcoin worthless. It changed the value of the megawatt.
Some miners will continue mining Bitcoin for years. Some sites will never be attractive to hyperscale AI. Some operators will run both businesses and move capital between them as economics change.
But the best power sites now have another bidder, and that bidder can offer something Bitcoin mining rarely can: a fifteen or twenty year contract backed by a giant technology company.
That changes every capital allocation decision.
Should the next hundred megawatts power ASICs and remain exposed to Bitcoin price, mining difficulty, and future halvings? Should they host Blackwell GPUs under a long term lease? Should the company remain a miner, become a cloud operator, or simply become the landlord underneath the AI economy?
There is no universal answer, but there is a new price on the infrastructure.
Bitcoin miners spent years learning how to turn electricity into money. AI has reached the point where, at some of the best sites, it can offer them more money for the electricity than for the Bitcoin they could mine with it.
That is why the ASICs are going dark.
Sources
- IREN fiscal 2026 Form 10-K, U.S. Securities and Exchange Commission
https://www.sec.gov/Archives/edgar/data/1878848/000187884826000052/iren-20260630.htm - Hut 8, “Hut 8 Fully Commercializes 1 GW Beacon Point AI Data Center Campus with Second 352 MW IT Lease,” July 20, 2026
https://www.hut8.com/news-insights/press-releases/hut-8-fully-commercializes-1-gw-beacon-point-ai-data-center-campus-with-second-352-mw-it-lease - TeraWulf Form 8-K, July 6, 2026, U.S. Securities and Exchange Commission
https://www.sec.gov/Archives/edgar/data/1083301/000110465926080583/tm2619468d1_8k.htm - CoinDesk, “Bitcoin miners are becoming AI companies and selling their BTC to fund the transition,” March 27, 2026
https://www.coindesk.com/markets/2026/03/27/bitcoin-miners-are-becoming-ai-companies-and-selling-their-btc-to-fund-the-transition - Hut 8 Beacon Point project information
https://www.hut8.com/data-centers/beacon-point