Bitcoin or AI: Which path will miners choose next?

The mining industry is entering what many analysts now call its most severe profitability crunch to date. Operating costs have jumped sharply as miners adopt power-hungry cooling systems and next-generation ASIC hardware. In this climate, only companies with deep pockets can survive—and even those players are increasingly reallocating part of their fleets toward artificial-intelligence infrastructure.

With the next bitcoin halving in 2028 looming, the pressure is only set to intensify. Survival will hinge more than ever on bitcoin’s price and whether on-chain activity can grow enough to lift fee revenue. Without that shift, the network risks consolidating into a handful of large corporations and government-backed entities capable of enduring a prolonged period of thin or negative margins.

A perfect storm

For the first time in bitcoin’s history, miners are facing what experts describe as a systemic crisis. Analysts at TheMinerMag highlighted how extreme current conditions have become in recent weekly reports.

Back in early Q3, hashprice hovered around $55 per PH/s per day. Following November’s market pullback, it collapsed to roughly $34.21 per PH/s per day. Even the most efficient miners—those with modern equipment and low power costs—are now operating near break-even. Their median Q3 “hashcost” is estimated at about $44 per PH/s per day.

What will happen after the next halving? Scheduled for April 2028, it will slash block rewards from 3.125 BTC to just 1.5625 BTC. Satoshi Nakamoto assumed transaction fees would eventually offset reducing rewards. Reality tells a different story: in November, fees accounted for less than 0.7% of miner revenue, far below historic peaks of 5–15%.

April 2024 briefly sparked hope when speculation around the Runes protocol pushed fees on the halving block to a record $2.4 million. But fees soon crashed back to multi-month lows—raising new doubts about long-term network security if incentives for miners continue to weaken.

What makes matters stranger is the combination of falling profits and rising hashrate. On October 16th, Hashrate Index reported a new all-time high of 1.16 EH/s, with mining difficulty close behind at nearly 156 T.

Higher costs and shrinking returns are forcing miners to rely less on current profitability and more on long-term optimism about bitcoin’s future—and on gaining market share now while weaker operators’ fold.

Only miners with cheap power, substantial capital reserves, and the latest hardware will likely make it through.

According to BlockEden.xyz, large-scale operators with electricity below $0.05/kWh still maintain healthy margins of 30–75%. Examples include:

• Marathon Digital (MARA): Estimated production cost per BTC around $39,235, with total costs between $26,000–28,000.
  
  
• Riot Platforms: Enjoys $0.025–0.03/kWh power rates in Texas.
  
  
• CleanSpark: Produces 1 BTC for roughly $35,000.
  
  

At current bitcoin prices, only miners paying $0.05–0.07/kWh can run the newest ASICs profitably—making home mining (typically $0.12–0.15/kWh) effectively dead. Older S19 machines are nearing retirement as the far more efficient S21 family takes over.

Marathon CEO Fred Thiel expects the market to purge unsustainable operators. By 2028, he predicts miners will need to generate their own power or formally partner with energy providers: “The days of plugging straight into the grid are numbered.”

Thiel also notes that rising difficulty reflects manufacturers activating their own hardware. Bitmain, Canaan and Bitdeer have been deploying unsold ASIC inventories themselves. One striking example came in October 2024, when Chinese auto firm Cango unexpectedly acquired 32 EH/s—likely via Bitmain transferring about $256 million in unsold S19 XP units through subsidiary Antalpha.

The rising side gig: AI

In September, bitcoin-mining profitability dropped more than 7%, prompting miners to step up diversification efforts. Many analysts argue that the rising hashrate is partly driven by companies squeezing out more cash flow to fund GPU-rich AI data centres.

The industry’s debt load ballooned from $2.1 billion to $12.7 billion in just a year, which VanEck attributes to hardware upgrades and AI-related commitments.

By mid-2024, Bernstein estimated that miners had secured roughly 6 GW of power for future expansion—a figure expected to double by 2027. Their ability to quickly integrate with large power grids makes them appealing partners for AI players.

“Bitcoin data centers already have robust power, cooling and operational infrastructure,” the analysts said.

Forecasts now suggest that by late 2027:

• Up to 20% of bitcoin mining capacity may shift to AI workloads
  
  
• The top five U.S. miners will control roughly 25% of global hashrate
  
  

Why miners are drifting toward AI:

• AI generates 2–5x more revenue per kWh than bitcoin mining
  
  
• Big tech firms like Microsoft and OpenAI desperately need GPU capacity
  
  
• Mining facilities already have cheap power and industrial cooling setups
  
  

Over 2024–2025, a wave of acquisitions and capital injections accelerated the pivot. In July, AI powerhouse CoreWeave bought mining operator Core Scientific for $9 billion. In August, Google expanded its stake in TeraWulf to 14% through $3.2 billion in financial guarantees. TeraWulf soon announced a further $3 billion raise for new data centres.

IREN put $193 million into AI computing, while Bitfarms said it would wind down bitcoin mining entirely by 2027 to focus on AI infrastructure.

Even Riot Platforms—long known for hoarding mined bitcoin—abandoned its HODL strategy in April 2025, selling all 463 BTC mined that month plus 12 reserve coins for $38.8 million to support operations. It continued selling in subsequent months.

Despite the shift, Riot’s Q3 numbers were strong: $104.5 million in net profit on a record $180.2 million in revenue, mining 1,406 BTC vs. 1,104 BTC the year before. It now holds 19,287 BTC (around $2.2 billion) and $170 million in cash.

VP Josh Kane summed up the company’s new direction: “We no longer see bitcoin mining as the end goal. It’s a means to maximise the value of our megawatts.”

Since late 2024, Riot has boosted capacity from 20.1 EH/s to 33.1 EH/s, with 126 MW at its Corsicana site earmarked for a new AI data-centre buildout. Mining infrastructure makes such pivots fast—operators can switch from hashing to AI workloads within minutes.

Survival of the fittest

ASIC efficiency has improved dramatically in recent years, with liquid cooling becoming essential rather than optional. The 2024–2025 generation represents a major leap forward, though improvements are slowing in line with Moore’s law.

Top models have gone from 31 J/TH in 2020 to 11–13.5 J/TH today—a roughly 65% gain over five years. As chips move into the 3–5 nm range, efficiency improvements have slowed to about 20–30%.

New models promise even fiercer competition. The S23 Hydro, expected in early 2026, aims for 9.5–9.7 J/TH at 580 TH/s. Bitdeer is developing 3–4 nm SEALMINER chips targeting 5 J/TH by 2026, while Jack Dorsey’s Block is working with Core Scientific to deploy open-source 3 nm hardware.

Cutting-edge cooling systems, however, raise entry costs substantially. Hydro cooling adds $500–1,000 per unit; immersion setups cost $2,000–5,000 upfront. These systems pay for themselves through 20–40% higher returns since they enable heavy overclocking.

Immersion cooling has become especially important. Air-cooled miners are loud—around 75–76 dB—and require heavy airflow management. Immersion eliminates noise and boosts performance by roughly 40%. Leading hydro models like the S21 XP Hyd and MicroBT M63S+ output water at 70–80°C, enabling heat reuse in agriculture, district heating or industrial processes.

The high capital requirements now form a steep barrier to entry, limiting smaller operators and reducing the potential for greater decentralisation.

The road ahead

The next two to three years could reshape bitcoin mining entirely. A bear market would force capitulation, free up hashrate and allow well-capitalised newcomers to expand. A sustained bull market wouldn’t fundamentally change the trend—it would simply soften it. Even at higher bitcoin prices, competition for cheap, “green” power and advanced cooling will stay fierce.

If miners keep shifting resources toward AI, bitcoin could face:

• slower hashrate growth
  
  
• lower difficulty
  
  
• reduced decentralisation
  
  
• greater vulnerability to state-linked mining groups or hostile syndicates
  
  

Researchers continue exploring incentive models—from game-theoretic “selfish mining” tweaks to reward mechanisms tied to hashrate—but practical solutions remain limited.

In reality, there are only a few ways to increase the share of miner revenue coming from fees:

• more capable Bitcoin Layer 2 solutions, including programmable smart contracts and new opcodes for BTC-native finance
  
  
• global expansion of micropayment systems such as Lightning
  
  
• mainstream adoption of bitcoin for everyday payments
  
  

Without progress on those fronts, miners may increasingly find themselves drawn not toward bitcoin, but toward the booming world of AI computing.