When Mining Changes Its Nature
According to the latest report from CoinShares, the weighted average cash cost to produce one bitcoin among listed miners reached approximately $79,995 in Q4 2025. For many, this made it the most challenging quarter since the April 2024 halving. Following the halving of rewards (from 6.25 BTC per validated block to 3.125), revenues dropped, while electricity, hardware and financing costs remained high. Consequently, each extracted bitcoin yields less, margins are compressed, and only the most efficient miners can truly find breathing room. This is especially true as bitcoin fell from $126,000 at the end of 2025 to $73,400 today.
The sector has thus tipped into a world where mining is no longer automatically synonymous with accumulating bitcoin; mining can become a mere transitory use of an even more precious asset: access to electricity and connected real estate.
In this new landscape, the competitive advantage is no longer solely the most efficient fleet of ASICs. It is becoming the ability to monetize highly sought-after assets in other ways: land, power lines, transformers, substations, load engineering, grid connection, and high-energy-density operational discipline. This is precisely what the AI explosion has made valuable. TeraWulf states explicitly that access to power has become the primary constraint for the large-scale computing industry. Hut 8, meanwhile, builds its entire communication around a "power-first" model. In other words: what miners built for Bitcoin can, in some cases, be revalued much higher by AI.
AI Is Already Siphoning the Heart of the Industry
CoinShares estimates that cumulative contracts announced in AI and high-performance computing (HPC) across listed miners exceed $70bn. The same report estimates that some miners could derive up to 70% of their revenue from AI by the end of 2026, compared to approximately 30% at the time of publication. For several groups, HPC is no longer a diversification but the primary source of expected growth for the coming years.
Core Scientific perhaps best embodies this trend. The company, which until recently was perceived as a pure-play miner, signed high-density colocation contracts with CoreWeave representing over $10bn in potential revenue over twelve years and approximately 590 MW under contract. In Q1 2026, its colocation revenue reached $77.5m out of $115.2m in total revenue: in other words, about two-thirds of the group's income already comes from this AI-related activity rather than pure mining.
At TeraWulf, the mutation is just as spectacular. The company announced 522 critical MW under contract for over $12.8bn in revenue; then, in Q1 2026, it generated $21m in HPC revenue out of $34m in total revenue. Here again, the center of gravity has shifted: the most stable, visible, and market-valued activity is no longer mined bitcoin, but long-term rent signed with AI counterparties.
Around this core group, other firms are following the same path with variations. Hut 8 signed a 15-year lease with Fluidstack at River Bend, Louisiana, for 245 MW and a contract value of $7bn over the base term, with significantly higher potential in case of renewals. Cipher initially secured approximately $3bn in revenue over ten years with Fluidstack at Barber Lake, before extending the partnership to the site's entire 300 MW, bringing the initial contracted revenue to approximately $3.8bn. Finally, IREN is no longer content with just mining: CoinShares notes $17.3m in AI Cloud revenue in Q4 2025, or 9% of the total, with up to 200 MW of liquid-cooled GPU capacity under construction.
Even the market has already chosen its side. CoinShares notes that miners with secured HPC contracts trade at approximately 12.3x their 12m forward sales, compared to only 5.9x for pure miners. In other words, the stock market grants a valuation premium of more than double for AI exposure.
Transformation Is Funded by Debt and Bitcoin Sales
Such a conversion is not free. Transforming a mining site into an AI computing site is not just about replacing machines. It involves rebuilding or adapting the electrical chain, redesigning cooling systems, adding redundancy, reinforcing structures, securing the premises, and sometimes destroying existing assets. Above all, it requires raising financing amounts far beyond traditional mining capital. Bitcoin infrastructure costs approximately $700,000 to $1m per megawatt, compared to $8m to $15m for AI infrastructure. It is a different scale entirely.
The primary source of funding is therefore debt. IREN carries $3.7bn in convertibles. TeraWulf already showed $5.8bn in total debt after raising massive financing for its HPC campuses. Core Scientific, for its part, issued $3.3bn in secured notes to fund its development projects.
The second source of funding, even more symbolic, is the sale of bitcoin. Here too, the cultural break is immense. CoinShares estimates that listed miners have collectively reduced their reserves by more than 15,000 BTC from their peaks.
This detail may seem financial; it is, in reality, philosophical. During the previous cycle, the implicit promise of many listed miners was: "we accumulate BTC for you." In 2026, the message has become much more nuanced: "we arbitrage our BTC for infrastructure assets if those assets offer better future cash flows." The companies securing the network are therefore no longer necessarily net accumulators of bitcoin. They may become structural sellers, precisely because they find it more profitable to fund AI data centers than to refresh their fleet of mining machines.
What This Pivot Means for Bitcoin Itself
The most visible consequence appears in network metrics. Bitcoin's hashrate had surpassed 1 zettahash for the first time in late August 2025, before peaking around 1,160 EH/s in early October. Then the movement reversed. The first quarter of 2026 marked the first first-quarter decline in hashrate since 2020, with the network returning to around 1 ZH/s.
However, one must avoid a misunderstanding. A drop in hashrate does not automatically mean that Bitcoin "ceases to be secure." The protocol adjusts its difficulty. The least efficient miners shut down, survivors sometimes see their relative position improve, and the network continues to function. Conversely, microeconomic arbitrage changes the nature of the marginal security budget: if a portion of capital, sites, and teams turns away from mining in favor of AI, then the growth of power dedicated to Bitcoin becomes more dependent on the BTC price than before. It is no longer just a technological race; it is direct competition between two uses of the megawatt.
Everything Now Depends on the Price of Bitcoin
At this stage, the big question is whether this metamorphosis is cyclical or structural. If bitcoin rises toward $100,000, the hashprice recovers, mining margins breathe again, and the incentive to sacrifice mining too quickly for AI is reduced. Conversely, if BTC remains sustainably below $80,000, we could see further miner exits. Below $70,000, the capitulation scenario becomes even more plausible.
The mining industry entered the decade as a group of companies tasked with securing Bitcoin and accumulating reserves. It is exiting it as a group of infrastructure operators capable, depending on prices, of performing two trades: mining when bitcoin pays enough, and leasing to AI when it pays better. This may be a temporary phase. It may already be a new era.
