Can Bitcoin Stay Decentralized in the AI Era?

What interests me about Bitcoin is the idea that money can work without a small group of companies deciding how everyone else gets to use it. You can hold your own keys, verify the rules, and take part in securing the network.

But somebody still has to buy the mining equipment, find electricity, and pay to keep the machines running. Bitcoin depends on businesses making those decisions, and those businesses face the same pressures as any other industry.

That’s what I keep thinking about as AI companies invest in computing infrastructure. If a few companies gain a much larger advantage in access to power, chips, and financing, how much of Bitcoin’s mining industry could eventually end up in their hands?

How this could happen

The worst case is simple: a few companies use their access to money and power infrastructure to buy mining equipment or acquire existing miners. They keep expanding, and smaller operators struggle to keep up. Eventually, much of Bitcoin’s mining power belongs to the same handful of owners.

An AI company would still need the right machines. Bitcoin uses specialized SHA-256 chips called ASICs. Owning a large GPU cluster does not give a company an equivalent amount of Bitcoin mining power. The hardware does a different job, and replacing it would require substantial investment.

But a company entering mining would not have to invent the entire business from scratch. It could buy equipment from existing manufacturers, partner with operators, or acquire miners outright. Access to capital and suitable power infrastructure would give it a head start, even though neither guarantees success.

There are already bottlenecks in this industry. Cambridge’s 2025 mining report found that electricity accounted for more than 80% of miners’ cash-based operating expenses, and that the three largest SHA-256 ASIC manufacturers accounted for more than 99% of the hardware market. The electricity figure excludes major costs such as buying the machines, but it shows how much access to affordable power matters.

My concern is that AI investment could widen the gap between companies that can secure these resources and everyone else. Mining ownership could become more concentrated through ordinary acquisitions and expansion, with no coordinated plan to attack Bitcoin. The software would still be open to anyone. Making a living as an independent miner could become much harder.

There is more than one kind of control

It helps to separate the businesses involved. Chip manufacturers decide what equipment to produce and whom to sell it to. Power providers and site owners affect where that equipment can run. Mining operators own or manage the machines. Pools coordinate mining work and, in many setups, choose which transactions go into the blocks their members try to mine.

These relationships can give a company influence beyond the machines it owns. A miner might own its ASICs but depend on one hosting provider or financier to keep operating. Buying a mining business could bring its equipment, contracts, and operating expertise under a new owner all at once.

Still, a pool’s share of mining is not the same thing as ownership of that hashrate. A pool can serve many independent miners, and miners can change pools. One company can also spread its machines across several pools. Pool statistics tell us something useful, but they do not tell us who owns the whole operation.

Full-node validation is another layer, with a different purpose. People running full nodes check that blocks follow Bitcoin’s rules. Their independence matters even when mining ownership is concentrated, because producing a block and deciding whether to accept it are separate decisions.

What would 51% actually mean?

Acquiring a majority of Bitcoin’s mining power would be an enormous undertaking. The relevant measure is hashrate, the rate at which machines perform Bitcoin’s mining calculations. It is separate from the compute figures used to describe AI systems.

If existing honest miners contribute hashrate H and a new entrant adds A, the entrant’s share is:

Adding machines equal to 51% of the existing network would therefore give the entrant about 34% of the enlarged network. To pass 50% by adding capacity alone, it would need more hashrate than the entire remaining honest network. That means obtaining the machines, deploying them, and keeping them powered while other miners can continue expanding.

A sustained mining majority could censor transactions, reorganize transaction history, and reverse its own payments to spend those coins again. Those are serious powers. The Bitcoin developer guide explains how an attacker’s hashrate affects its ability to replace transaction history.

It would not give the attacker permission to create arbitrary bitcoin or spend other people’s coins without valid signatures. Full nodes reject blocks that break those rules, regardless of how powerful the miner is.

I think this distinction is easy to lose in discussions about 51% attacks. Miners propose blocks and influence which transactions get confirmed. Full nodes independently check whether those blocks are valid. A mining cartel could exercise considerable power over payments while remaining unable to change the rules for people who enforce them themselves.

Why concentration does not guarantee a takeover

A more concentrated mining industry would not automatically mean someone was attacking the network. Large operators could keep mining honestly, and independent nodes could keep enforcing the protocol. But for a person whose transactions were being censored, the fact that the blocks were technically valid would be little comfort.

There are also reasons an owner would hesitate to attack. Buying mining equipment creates a financial interest in Bitcoin continuing to work. Damaging confidence in the network could reduce the value of future mining revenue and the equipment itself. The white paper discusses this incentive to mine honestly. It offers some protection, although it is less reassuring when an attacker has motives beyond earning mining revenue.

AI companies face their own tradeoffs. If serving AI customers is more profitable, buying mining operations may make little business sense. A contract to use a cloud provider’s GPUs also does not necessarily give a lab the right to replace them or take over the site. We should be careful about treating an AI lab, its cloud provider, and the owner of its data center as a single organization.

Nor does expanding AI infrastructure have to hurt independent miners. Investment could bring more electricity and manufacturing capacity into the market. Miners that can use remote or interruptible power may retain advantages of their own. Whether AI investment makes mining more concentrated depends on who gets access to the resulting capacity and on what terms.

What I’d want to know

The connection between the two industries is already visible. In 2024, Core Scientific agreed to provide roughly 200 megawatts of infrastructure to CoreWeave, modifying existing sites to host AI hardware. That deal shows why powered sites matter to both industries. The conversion still required investment; it tells us little by itself about the cost of going from AI hosting into Bitcoin mining.

What remains uncertain is how this overlap will affect ownership over time. I would want to know who owns the mining machines, who supplies them, who controls the sites and power contracts, and who finances the operators. I would also want to know how easily a miner can change any of those relationships without going out of business.

Those questions give us a better starting point than adding up AI companies’ compute budgets and assuming the largest number wins. They also make it possible to notice a loss of independence before any one company gets close to a majority of Bitcoin’s hashrate.

Bitcoin gives people a way to verify the rules for themselves. Whether enough independent people and companies can afford to keep producing blocks is a question about the world around Bitcoin. The AI buildout is changing that world quickly, and I don’t think we yet understand what that will mean for mining.