Bitcoin miners spent years building some of the most power-hungry computing infrastructure on Earth.

Now some of them are realizing that Bitcoin may no longer be the best customer for it.

The shift has accelerated dramatically in 2026. CoinShares estimates that publicly listed Bitcoin miners have announced more than $70 billion in cumulative AI and high-performance-computing contracts, while some miners could generate as much as 70% of their revenue from AI by the end of this year.

Then came a string of increasingly difficult-to-ignore moves.

Core Scientific signed a 15-year agreement tied to 529 MW of AI infrastructure with AMD. Bitdeer secured a potential $4.7 billion, 16-year AI data-center lease in Norway. And this week, Keel Infrastructure shut down all of its U.S. Bitcoin mining operations and began converting those sites toward AI and high-performance computing.

The obvious conclusion is that miners are diversifying.

The more interesting conclusion is that AI may be changing the economics of Bitcoin mining itself.

Bitcoin Mining Has Become an Electricity Business

Bitcoin mining is often described as a technology business.

At its core, it is an energy business.

Miners buy specialized machines called ASICs, connect thousands of them to electricity, and compete to process Bitcoin transactions and secure the network. The machines are expensive, but electricity is the recurring cost that determines whether the operation survives.

That makes the location of a mining facility incredibly important.

A miner with cheap power can potentially remain profitable while another operator with expensive electricity loses money using exactly the same hardware.

The problem is that the economics have become increasingly difficult.

CoinShares reported that the weighted average cash cost to produce one Bitcoin among publicly listed miners reached approximately $79,995 in Q4 2025. It also said hashprice, a key measure of mining revenue relative to computing power, fell to roughly $29 per petahash per second per day in Q1 2026. At that level, CoinShares estimated that roughly 15% to 20% of the global mining fleet was uneconomic under certain power-cost and hardware assumptions.

This creates a brutal situation.

A mining company can own a giant facility with hundreds of megawatts of electrical capacity and still struggle to make attractive returns from Bitcoin.

But that same electricity connection could be extremely valuable to an AI company.

And that is where the industry's strategy starts to make sense.

AI Doesn't Care About Bitcoin. It Cares About Megawatts.

The surprising thing about the AI boom is that one of its biggest bottlenecks isn't software.

It's power.

Large AI systems require enormous quantities of electricity and sophisticated data centers capable of supporting high-density GPU clusters. Getting the chips is only one part of the problem. Companies also need land, power, cooling, fiber connectivity, transformers, substations and buildings capable of supporting the equipment.

Building all of that from scratch takes years.

Bitcoin miners already spent years doing something remarkably similar.

They found cheap power.

They secured land.

They built electrical infrastructure.

They negotiated grid connections.

They constructed large data-center facilities.

They learned how to operate enormous computing loads continuously.

The machines are different, but the infrastructure underneath them overlaps.

That gives miners something increasingly valuable in the AI economy: time-to-power.

An AI company may be willing to pay a premium for a facility that can provide large quantities of electricity sooner than a completely new data center.

This is why the miner-to-AI transition isn't necessarily as strange as it first appears.

Bitcoin miners aren't abandoning computing.

They're changing what kind of computing happens inside their buildings.

Core Scientific Shows What the Pivot Can Look Like

Core Scientific is one of the clearest examples.

The company historically built its business around Bitcoin mining infrastructure. Today, AI and high-performance computing are increasingly becoming the center of its strategy.

In July, Core Scientific announced a 15-year infrastructure partnership involving 529 MW of capacity for AMD-related AI deployments. The agreements could generate more than $14 billion in base contracted revenue, while AMD received rights to reserve another 1.925 GW through 2028 under certain conditions.

The shift is already visible in the company's financial results.

In Q2 2026, Core Scientific's colocation business generated $136.7 million, representing 83% of its $164.2 million in quarterly revenue. Meanwhile, self-mining revenue fell 66% to $21.5 million.

That is a dramatic change in the business model.

The company isn't simply putting a few GPUs next to its Bitcoin miners.

It is increasingly behaving like a data-center infrastructure company that happens to have a history in Bitcoin mining.

And the market is beginning to value these companies differently.

Bitcoin miners traditionally get valued according to things such as Bitcoin production, hashrate, energy costs and the amount of BTC on their balance sheets.

AI infrastructure businesses can be valued around contracted revenue, power capacity, customer quality and the expected life of their data-center agreements.

Those are very different valuation frameworks.

The transformation could therefore change not only what these companies do, but how investors think about them.

Bitdeer's $4.7 Billion Deal Reveals the New Economics

Bitdeer offers an even more striking example.

In August, the company announced a 16-year lease agreement potentially worth about $4.7 billion for AI and high-performance computing capacity at its Tydal facility in Norway. The arrangement covers 121 MW of IT capacity and is designed for NVIDIA GPU-based workloads.

The deal is not yet effective and remains subject to closing conditions, so the headline $4.7 billion should not be treated as money already earned.

That distinction matters.

But the strategic significance is still enormous.

A Bitcoin miner is effectively taking infrastructure originally designed to turn electricity into Bitcoin and trying to turn it into long-duration AI data-center revenue.

Bitdeer's own disclosures show how broad this transformation has become. As of June, the company had deployed 4,248 GPUs, with a 95% utilization rate and approximately $76 million in annual recurring revenue from its AI cloud business. It also had a 21.7 MW Malaysian AI data-center lease planned around 128 NVIDIA GB300 NVL72 systems.

The company now presents itself across Bitcoin mining, data centers, AI cloud and hardware manufacturing.

That is not a side project anymore.

It's a second business.

Keel Just Demonstrated the Other Side of the Equation

Then there is Keel Infrastructure.

Its story is more painful.

Keel shut down its remaining U.S. Bitcoin mining operations on June 29 and is converting those sites toward high-performance computing and AI. Its Q2 revenue fell roughly 50% year over year to $30.4 million, while the company reported a net loss of about $65 million.

Keel also sold 1,085 BTC between April 1 and August 7 for approximately $75 million. Management said its intention was to liquidate its remaining Bitcoin position during 2026, although that remains a plan rather than a completed sale.

There is an important warning hidden here.

The AI pivot doesn't automatically create revenue.

As of August 7, Keel had not yet generated HPC data-center revenue from its Washington or Pennsylvania sites. The company is still in the development and conversion phase.

That means the market is currently dealing with two very different types of miner-to-AI stories.

Some companies have signed large contracts and are already earning AI-related revenue.

Others have shut down mining and are betting that AI customers will eventually arrive.

Those are not equivalent.

A data center without a customer is an expensive building.

The Bitcoin Network Could Feel the Consequences

This trend matters to Bitcoin itself.

If mining companies increasingly allocate their best power capacity to AI, fewer resources may be available for Bitcoin mining.

CoinShares has already described a widening split between miners that are becoming AI infrastructure companies and those remaining focused on Bitcoin. It expects the AI/HPC migration to push some Bitcoin mining toward cheaper, more flexible or intermittent power sources over time.

That could change the structure of the mining industry.

The most valuable power sites might increasingly go to AI workloads because AI customers can offer longer-term contracts and potentially more predictable revenue.

Bitcoin mining could then become the flexible load.

A facility might mine Bitcoin when electricity is cheap and redirect capacity elsewhere when a higher-value customer needs it.

That would make mining more like an energy arbitrage business.

This isn't necessarily bad for Bitcoin.

In fact, one possible consequence is greater decentralization of mining capacity if large industrial sites become less attractive relative to smaller operations using stranded, curtailed or otherwise cheap energy.

But there is another side.

If major public miners reduce their Bitcoin operations, the network could become more dependent on miners with extremely low energy costs, specialized power arrangements or access to unusual energy sources.

The economics of Bitcoin mining could therefore become more concentrated around energy efficiency rather than simply scale.

The Real Competition Is for Power, Not Bitcoin

This may be the most important insight.

Bitcoin miners aren't really competing with AI companies for computers.

They're competing for electricity.

AI data centers and Bitcoin mines can both consume enormous amounts of power, but they generate revenue in radically different ways.

Bitcoin mining converts electricity into BTC according to a globally competitive algorithmic process.

AI data centers sell computing capacity to customers.

If a company can make more predictable revenue by renting its electricity-backed infrastructure to an AI customer than by using the same capacity to mine Bitcoin, economics will push it toward AI.

That doesn't require anyone to hate Bitcoin.

It doesn't require a technological breakthrough.

It is simply capital allocation.

And this is where the AI boom could have an unusual second-order effect on crypto.

AI may effectively bid up the value of the physical infrastructure required to mine Bitcoin.

Power contracts become more valuable.

Grid connections become more valuable.

Permitted land becomes more valuable.

Large substations become more valuable.

Existing data centers become more valuable.

Even the ability to secure a connection to the grid can become a strategic asset.

Bitcoin miners accidentally spent years building infrastructure that AI companies now desperately want.

But Converting a Bitcoin Mine Into an AI Data Center Is Hard

There is a misconception that a Bitcoin mine can simply remove its ASICs and plug in NVIDIA GPUs.

It can't.

AI infrastructure has significantly different requirements.

Bitcoin mining is relatively flexible. If power disappears for an hour, the miner can simply stop hashing and restart later. AI workloads can require much more demanding uptime, cooling, redundancy and networking.

Galaxy Research has previously highlighted this difference, noting that AI data centers have more stringent redundancy requirements than Bitcoin mining facilities. Mining operations can be flexible in ways that AI workloads generally cannot.

Cooling is another major issue.

Modern AI GPUs can generate enormous heat densities, which can require advanced liquid-cooling systems and facility redesigns. High-speed networking also becomes crucial because AI accelerators often need to communicate with one another at extremely high bandwidth.

Then there is reliability.

A Bitcoin miner can accept some operational interruptions if the economics make sense.

An AI customer running mission-critical workloads may demand far higher uptime.

That means the conversion isn't simply a hardware swap.

It is an infrastructure upgrade.

And that costs money.

The $70 Billion Number Comes With a Big Asterisk

The headline figure is impressive: more than $70 billion in announced AI/HPC contracts across public Bitcoin miners, according to CoinShares.

But investors should be careful with the word announced.

A contract announcement does not necessarily mean that a data center is already operational.

Some agreements depend on financing.

Others depend on construction.

Some require permits or grid upgrades.

Some involve future capacity rather than currently available capacity.

And some contracts contain options that may or may not be exercised.

CoinShares itself warns that not all announced deals will translate into operational infrastructure and that the capital requirements for AI buildouts are substantial.

That is why the next stage of the story will be about execution.

Which companies can actually deliver power?

Which can build on schedule?

Which can attract credible AI customers?

Which can finance the construction without taking on dangerous amounts of debt?

And which companies simply slapped "AI" onto their investor presentation because the market was willing to pay a higher multiple for it?

Those questions are going to separate the real infrastructure businesses from the AI-washing stories.

Bitcoin Mining May Become the Backup Business

The strangest possible outcome is that some companies that started as Bitcoin miners may eventually treat Bitcoin as their secondary business.

Imagine a company with 1 GW of available power.

During periods when an AI customer is using the facility, most of the capacity goes toward GPUs.

But when capacity is temporarily unused, the company can switch some electricity toward Bitcoin mining.

Bitcoin becomes a flexible buyer of power rather than the primary customer.

That could actually make Bitcoin mining more economically interesting.

Instead of building a massive facility solely to mine BTC, operators could build infrastructure for a higher-value customer and use Bitcoin mining to monetize excess electricity.

The two businesses would no longer necessarily compete.

They could complement each other.

The irony is remarkable.

AI could reduce the amount of dedicated Bitcoin mining infrastructure while simultaneously creating a new reason to mine Bitcoin around the edges of the AI economy.

What This Means for Bitcoin Miners Going Forward

The industry is probably going to split into three groups.

The first will become AI infrastructure companies.

These firms will increasingly measure themselves in megawatts, GPU capacity, contracted revenue and data-center utilization rather than Bitcoin produced.

The second group will remain pure-play Bitcoin miners.

Their advantage will need to come from extremely cheap energy, efficient ASICs, operational excellence and perhaps access to stranded or flexible power sources.

The third group will become hybrids.

They will use Bitcoin mining as a flexible energy-consuming business while gradually allocating their most valuable infrastructure to AI and other high-performance computing workloads.

None of these strategies is guaranteed to win.

AI demand could disappoint.

Power prices could rise.

Data-center construction could face delays.

GPU economics could change.

Bitcoin could enter another major bull market and make mining dramatically more attractive again.

But the underlying trend is difficult to dismiss.

When a company can earn potentially long-duration infrastructure revenue from AI customers, it has a powerful reason to reconsider using every available megawatt to produce Bitcoin.

Conclusion

The Bitcoin mining industry may be undergoing its most important transformation since the industrialization of mining itself.

The machines are still there. The power is still there. The data centers are still there.

But the customer is changing.

Core Scientific is signing enormous AI infrastructure contracts. Bitdeer is building an AI business alongside its mining operation. Keel has gone as far as shutting down its U.S. Bitcoin mining operations while it converts those sites for AI and high-performance computing.

The reason is straightforward: AI is creating a new market for the scarce resource Bitcoin miners spent years accumulating, namely reliable access to large amounts of electricity.

That doesn't mean Bitcoin mining is dying.

It means the economics are changing.

And the next generation of Bitcoin miners may not think of themselves primarily as Bitcoin companies at all.

They may think of themselves as power companies with the option to mine Bitcoin.

FAQ

1. Why are Bitcoin miners moving into AI?

Bitcoin mining profitability has been pressured by network difficulty, energy costs and Bitcoin price fluctuations, while AI companies are searching for large amounts of data-center capacity and power. Existing mining infrastructure can sometimes be adapted to serve that demand.

2. Are Bitcoin miners abandoning Bitcoin?

Some are reducing their dependence on Bitcoin, but the industry is not moving uniformly. Companies such as Keel have shut down U.S. mining operations, while others continue mining while building AI capacity.

3. How much money has been committed to AI by Bitcoin miners?

CoinShares says publicly listed miners have announced more than $70 billion in cumulative AI and HPC contracts. That figure represents announced contracts, not necessarily revenue already generated.

4. Why are Bitcoin mining facilities useful for AI?

Both industries require substantial electricity and large-scale computing infrastructure. Mining companies may already have power connections, land, buildings, cooling systems and grid infrastructure that can reduce some of the time required to develop new AI facilities.

5. Can Bitcoin mining facilities easily become AI data centers?

No. AI workloads generally require more demanding cooling, networking, redundancy and uptime requirements. Converting a mining facility can require significant construction and capital investment.

6. What happened with Core Scientific?

Core Scientific signed a 15-year agreement involving 529 MW of AI infrastructure with AMD, potentially generating more than $14 billion in base contracted revenue. Its Q2 results showed colocation revenue already making up the majority of its quarterly revenue.

7. What did Bitdeer do?

Bitdeer signed a potential $4.7 billion, 16-year lease for 121 MW of AI data-center capacity in Norway. The agreement remains subject to closing conditions, so the headline value should not be interpreted as guaranteed realized revenue.

8. Why did Keel shut down U.S. Bitcoin mining?

Keel said it was redirecting its U.S. power portfolio toward AI and high-performance computing. Its Q2 revenue fell 50% year over year, illustrating the financial pressure behind the transition.

9. Will AI reduce Bitcoin's hashrate?

It could put downward pressure on Bitcoin mining capacity if miners redirect their most valuable power resources to AI. The actual impact will depend on Bitcoin's price, mining profitability, new capacity and how much infrastructure is converted. CoinShares expects the AI/HPC migration to contribute to a decline in Bitcoin mining's share of some miners' revenue.

10. Could AI actually help Bitcoin mining?

Yes. Some operators could use Bitcoin mining as a flexible way to monetize power when AI capacity is idle or electricity is unusually cheap. That would make mining complementary to AI infrastructure rather than purely competitive with it.

11. What is hashprice?

Hashprice measures the expected Bitcoin mining revenue generated by a unit of computing power over a given period. Miners use it alongside electricity costs and hardware efficiency to estimate whether operating their machines is economically viable.

12. Is Bitcoin mining still profitable in 2026?

It depends heavily on the miner's electricity price, hardware efficiency, financing costs and Bitcoin's market price. CoinShares estimated that at a hashprice around $30 per PH/s/day, certain older or less efficient machines were losing money, representing an estimated 15% to 20% of the global fleet under its assumptions.

13. Does this mean Bitcoin is losing to AI?

Not necessarily. AI and Bitcoin are competing for electricity and infrastructure, but they serve completely different markets. The trend primarily shows that some companies believe AI customers currently offer a more attractive return on scarce power capacity.

14. What is the biggest risk in the miner-to-AI strategy?

Execution. Announced contracts still require financing, construction, permits, power delivery and customers to become operational revenue. CoinShares specifically cautions that not every announced AI deal will necessarily translate into functioning infrastructure.

15. Could Bitcoin miners become data-center companies?

Some already are moving in that direction. Core Scientific, Bitdeer and several other public miners increasingly operate or develop AI/HPC infrastructure alongside or instead of traditional Bitcoin mining.

16. What should investors watch next?

Watch actual energized megawatts, contracted customers, utilization, construction progress and AI-related revenue rather than headline contract values alone. Those figures will reveal which miners have genuinely transformed their businesses and which are still making promises about future capacity.

Key Takeaways

• More than $70 billion of AI/HPC contracts have been announced across public Bitcoin miners, according to CoinShares, showing that the AI pivot has moved well beyond a handful of experiments.

• CoinShares estimates AI could generate up to 70% of some listed miners' revenue by the end of 2026, compared with roughly 30% at the time of its report.

• Core Scientific's Q2 colocation revenue already represented 83% of quarterly revenue, while its self-mining revenue dropped 66% year over year.

• Bitdeer's $4.7 billion AI lease is enormous but should not be treated as guaranteed revenue, because the agreement remains subject to closing conditions.

• Keel's U.S. shutdown shows the downside of the transition: it has redirected its sites toward AI but had not yet generated HPC revenue from those properties as of August 7.

• The scarce asset is increasingly electricity and grid access, not Bitcoin mining hardware. AI companies can potentially pay infrastructure operators for long-term access to power capacity.

• Bitcoin mining may increasingly become a flexible power customer, operating when electricity is cheap or AI capacity is temporarily unused.

• Not every AI contract is equal: investors need to distinguish signed agreements from energized capacity and actual recurring revenue.

• The mining industry may split into AI-focused operators, pure Bitcoin miners and hybrid companies, each with a very different economic model.

• The long-term winner may be the company that controls the best power infrastructure, regardless of whether that infrastructure ultimately runs ASICs, GPUs or both.

Disclaimer

This article is for educational and informational purposes only and is not financial, investment, trading, legal or tax advice. Bitcoin mining, AI infrastructure companies and digital assets involve substantial risks, including volatility, energy-price risk, technological changes, financing risk, execution risk, regulatory uncertainty and potential loss of capital. Announced contracts may not become operational revenue, and company plans can change. Conduct independent research and consult qualified professionals before making financial decisions.