From mining to artificial intelligence
Conversion is reshaping the industry without weakening Bitcoin
Several large Bitcoin mining companies are converting part of their infrastructure into data centers for artificial intelligence through multiyear agreements with leading technology companies. Compressed margins and the decline in the price of bitcoin have reduced the profitability of mining, while growing demand for AI computing capacity has increased the value of existing land, substations, grid connections and energy contracts. The conversion concerns the sites rather than the machines installed within them, as ASICs, integrated circuits designed exclusively to perform the cryptographic calculations required by Bitcoin, cannot be used to train or run artificial intelligence models.
This transformation does not necessarily reduce Bitcoin’s security, which depends on the total computing power committed to the network and how that capacity is distributed. ASICs removed from converted sites can be sold or transferred to other facilities, while the protocol periodically adjusts mining difficulty to the capacity available. The migration of some infrastructure to AI may therefore alter the industry’s geography and composition without automatically causing an equivalent reduction in the computing power dedicated to the network.
Why Bitcoin needs mining
When a user sends bitcoin, the transaction is broadcast to the network and verified by nodes according to the protocol’s rules. Miners collect valid transactions, place them in a block and compete to add it to the blockchain by repeatedly performing cryptographic calculations in search of a result that satisfies the conditions established by the network. The first miner to find one proposes the new block, while the other nodes verify its validity before accepting it and continuing to build the chain.
This process, known as proof of work, allows the network to agree on a single ordering of transactions, prevents the same bitcoin from being spent more than once and makes altering the blockchain economically unattractive. To change a confirmed transaction, an attacker would have to reconstruct the block containing it and every block produced thereafter, generating an alternative chain with more accumulated computational work than the chain recognized by the network. The operation would therefore require a substantial deployment of ASICs and energy, with no certainty that the costs could be recovered. The higher the hashrate and the longer a transaction remains in the blockchain, the more work and expense such an attack would require.
In exchange for this activity, the miner that produces a block receives the transaction fees it contains and an amount of newly issued bitcoin determined by the protocol under a schedule that halves the reward every 210,000 blocks, or approximately every four years. In April 2024, the reward fell from 6.25 to 3.125 bitcoin per block. Mining therefore serves both as the mechanism that protects the network’s history and as the process through which new units enter circulation at a predetermined rate.
What mining farms are
In Bitcoin’s early years, mining could be performed with ordinary computers, but the growth of the network’s total computing power encouraged the development of ASICs, specialized machines that execute the protocol’s algorithm as efficiently as possible. Because a single machine now has an extremely low probability of producing a block independently, much of the hashrate is generated by industrial facilities that house thousands of ASICs and participate in mining pools, which coordinate the work of multiple participants and distribute rewards in proportion to the computing power contributed.
A mining farm is an industrial facility that requires a continuous energy supply, grid connections, transformers, substations, ventilation or cooling systems, internet connectivity and specialized technical staff. Since electricity is one of the main cost items, operators favor areas where energy is available at low prices or cannot easily be absorbed by other activities, in many cases relying on renewable sources. Agreements with electricity suppliers and grid operators may also allow them to reduce consumption rapidly when the power system is under greater strain.
Unlike conventional data centers, mining farms can switch ASICs off and on without losing data or interrupting customer services, allowing consumption to be adjusted in response to electricity prices and grid conditions. This flexibility does not eliminate the activity’s economic vulnerability, as profitability depends on the price of bitcoin, the block reward, mining difficulty, machine efficiency and the cost of energy. A deterioration in any one of these factors can quickly make less competitive facilities uneconomic.
Why artificial intelligence needs data centers
Artificial intelligence models are trained and run in data centers that bring together large numbers of GPUs and other specialized processors, connected through high speed networks and supported by systems capable of ensuring a continuous power supply and heat dissipation. Strong demand for AI computing capacity allows these facilities to offer better prospects for profitability than Bitcoin mining, particularly when contracts involve large technology companies and secure revenue over extended periods.
In this context, an already energized mining farm offers a significant advantage because it has industrial land, permits, supply contracts, substations and grid connections that would require substantial investment and lengthy development timelines to build from scratch. According to an analysis published in May 2026 by Bernstein, an international research and financial services firm, obtaining a new grid connection for one gigawatt of capacity in the United States can take an average of approximately four years, even in the states most supportive of data center development. The availability of operating infrastructure can therefore shorten a substantial part of the process and enhance the strategic value of existing sites.
Conversion nevertheless remains complex because ASICs cannot run artificial intelligence workloads and must be replaced with GPUs and entirely different components, whose supply is struggling to keep pace with demand from the largest technology companies. A mining farm is also designed to support an activity that can be interrupted, whereas an AI data center must provide continuous operation, redundant power, high capacity connections and more advanced cooling systems. Access to energy and a grid connection therefore represents an important advantage, but it is not sufficient to make every site technically and economically suitable for conversion.
Why miners are changing strategy
The halving is the mechanism built into the Bitcoin protocol that reduces the amount of bitcoin awarded to miners by half every 210,000 blocks, or approximately every four years. The April 2024 halving reduced the reward for each new block from 6.25 to 3.125 bitcoin, structurally compressing one of the industry’s main sources of revenue. Its effects were not fully apparent immediately because they were initially offset by the rise in the price of bitcoin. Pressure became more evident after the record high of $126,000 reached in October 2025, from which bitcoin subsequently lost almost 50%, reducing the value of mining rewards while energy and operating costs remained broadly unchanged.
The gap between the two industries has consequently widened, as mining must absorb a lower reward, the price correction, rising difficulty and persistently intense competition, while investment in artificial intelligence continues to support demand for computing capacity. In its Bitcoin Mining Report for the first quarter of 2026, CoinShares, a European company specializing in the management of crypto asset investment products, reported that hashprice, the daily revenue earned per unit of computing power, had fallen to its lowest level in five years, leaving an estimated 15% to 20% of older facilities operating at a loss. The report also projects that AI could account for as much as 70% of revenue by the end of 2026 among listed miners with active AI contracts.
The economic scale of the shift is evident in the contracts signed between mining companies and large technology groups. According to CoinShares, the largest agreements guarantee total revenue of more than $10 billion over periods that can exceed a decade, providing multiyear cash flows, financially sound counterparties and greater predictability than mining. Conversion therefore reflects a different balance between risk and return, allowing infrastructure exposed to bitcoin’s volatility to be monetized through long term contracts linked to growing demand for computing capacity.
Hashrate and network security
Hashrate is the total computing power employed by miners and is one of the main indicators of Bitcoin’s economic security. A higher level increases the amount of hardware and energy required to attempt to build a competing chain and alter the transaction history. Despite compressed margins and diversification by some operators, hashrate currently remains around 900 exahashes per second, a historically high level that has not declined in proportion to the number of projects announced in the artificial intelligence sector.
Security, however, depends not only on the total amount of computing power but also on how it is distributed among operators and mining pools. If compressed margins and conversion to artificial intelligence were to drive a large number of operators out of the market, a growing share of hashrate could become concentrated among the remaining large operators. This is not an immediate risk, but it is a dynamic worth monitoring, since greater concentration could increase the ability of a small number of participants to coordinate, censor transactions or influence their selection. Mining pools, however, are not necessarily the owners of the machines connected to them, and miners can redirect their computing power to other pools, mitigating this risk to some extent.
Conversion announcements are therefore not sufficient on their own to assess the implications for Bitcoin’s security. The relevant variables are the amount of computing power that actually leaves the network and the degree of concentration in the remaining hashrate, two distinct dynamics that can evolve independently of the number of companies moving into artificial intelligence.
Conversion does not eliminate hashrate
The conversion of a mining farm to artificial intelligence does not necessarily mean that the ASICs installed at the site cease contributing to the Bitcoin network. Although they cannot be used for AI workloads, these devices retain economic value for other miners and can be sold on the secondary market, transferred to facilities with lower energy costs or reactivated when conditions become more favorable. Conversion may therefore redistribute hashrate among operators and geographic regions rather than remove it permanently from the network.
If part of the hashrate is switched off, the number of attempts made every second to find a valid block declines and block production temporarily slows. Difficulty remains unchanged until 2,016 blocks have been completed, a process that takes approximately two weeks when the protocol’s average interval of ten minutes between blocks is maintained. At the end of the period, nodes independently recalculate the proof of work target based on the time elapsed: if blocks were produced more slowly than expected, difficulty decreases; if they were produced more quickly, it increases. The protocol thereby adjusts the probability of finding a block to the available computing power and progressively returns the network to its intended pace.
A particularly significant example occurred in 2021, when China, which at the time hosted a substantial share of global mining activity, progressively banned the industry and ordered facilities in the country to close. The simultaneous shutdown of numerous mining farms reduced global hashrate by almost 50% and temporarily slowed block production. The network nevertheless continued to operate and, following the difficulty adjustment, a substantial number of ASICs were transferred to other countries, including the United States and Kazakhstan. Their subsequent reactivation supported the gradual recovery of computing power, illustrating how hardware mobility and economic incentives can help rebalance the network even after a shock of considerable scale.
Conclusions
The conversion of some mining farms to artificial intelligence reflects the widening gap between two industries that compete for the same energy resources but are experiencing different economic conditions. Demand for AI infrastructure continues to support investment and prospects for profitability, while mining is contending with the effects of the halving, bitcoin’s decline from its highs, rising difficulty and energy costs. Existing land, permits, substations and supply contracts therefore represent a meaningful advantage, although conversion requires new capital, advanced chips and substantial upgrades to the facilities.
Conversion concerns the site rather than the mining machines, as ASICs cannot perform artificial intelligence workloads and must be replaced with entirely different hardware. The removed machines can nevertheless be sold or transferred to other miners, which means that converting a facility does not necessarily cause an equivalent reduction in total hashrate. When computing power is actually switched off, the difficulty adjustment preserves the regular pace of block production and strengthens the economic incentives of the remaining operators, without automatically restoring the level of security associated with the previous hashrate.
The industry’s transformation therefore does not represent an immediate threat to Bitcoin, but it makes both the amount and distribution of active computing power important to monitor. The decisive issue is not the number of companies announcing a conversion, but the share of hashrate that actually leaves the network, the ability to redeploy ASICs and any resulting concentration among a smaller number of operators, pools or geographic regions. Bitcoin’s resilience will continue to depend on the system’s ability to maintain adequate economic incentives and a sufficiently distributed base of miners.
