The Bitcoin mining industry is undergoing its most radical structural transformation since China's 2021 ban. Facing record-low hashprices, a 66% collapse in per-terahash revenue since October 2025, and BTC trading roughly 20% below average production costs, publicly traded miners are not merely d...
"We are no longer a Bitcoin company. We are an infrastructure-first owner and developer for HPC/AI data centers across North America." — Ben Gagnon, CEO, Bitfarms (now Keel Infrastructure)
The Bitcoin mining industry is undergoing its most radical structural transformation since China's 2021 ban. Facing record-low hashprices, a 66% collapse in per-terahash revenue since October 2025, and BTC trading roughly 20% below average production costs, publicly traded miners are not merely diversifying — they are abandoning their core identity. Bitfarms has formally announced it will rebrand as Keel Infrastructure and exit Bitcoin mining entirely. Cipher Mining has locked in $8.5 billion in AI hosting contracts. TeraWulf now counts Google as a 14% equity stakeholder. Morgan Stanley has initiated coverage of these companies not as crypto plays, but as data center infrastructure assets.
This is not a cyclical downturn response. It is a permanent capital reallocation driven by an inescapable economic reality: AI compute demand is growing exponentially while Bitcoin block rewards are shrinking on a fixed schedule. The miners who built North America's largest power-connected computing footprint are now repurposing that infrastructure for a customer willing to pay orders of magnitude more per megawatt-hour. The implications for Bitcoin's network security, decentralization, and long-term economic model are profound.
Bitcoin mining economics have deteriorated to levels that make continued pure-play mining irrational for most public operators. The data is unambiguous:
Only the most efficient operators — those running latest-generation S21-class ASICs with sub-$0.05/kWh power costs — can produce Bitcoin profitably, at estimated costs of $34,000–$43,000. Everyone else is underwater.
On February 24, Rosenblatt analyst Chris Brendler issued a note declaring that Bitcoin mining is "no longer profitable for all miners except the most efficient operations," adding bluntly: "With HPC economics still improving amid increased demand from the hyperscalers, we think all miners should now be actively transitioning from BTC to HPC if at all possible."
Bloomberg reported that its closely-watched mining revenue gauge fell to its lowest level on record in early February, as operators began unplugging equipment. A severe winter storm in Texas — home to the largest concentration of North American hashrate — forced additional shutdowns, compounding an already dire situation.
The strategic pivot is no longer theoretical. It is being executed at scale, with corporate identity changes that signal permanence:
Bitfarms → Keel Infrastructure. On February 6, 2026, Bitfarms announced it would rebrand as Keel Infrastructure, redomicile from Canada to the United States, and trade under the new ticker KEEL on Nasdaq and TSX. CEO Ben Gagnon stated that converting just the company's Washington state facility to GPU-as-a-Service "could potentially produce more net operating income than we have ever generated with Bitcoin mining." A shareholder vote is scheduled for March 20, with the transition expected to close around April 1. The company holds $698 million in liquidity and has begun repaying its $300 million Macquarie credit facility. Bitfarms has secured a fully funded $128 million deal with a major U.S.-based data center partner for equipment and materials, with completion targeted for December 2026.
Cipher Mining has appointed new board members reinforcing its shift toward large-scale data center development. The company's transformation is backed by hard contracts worth billions (detailed below).
TeraWulf has repositioned itself as a "digital infrastructure" company, with Google now holding a 14% equity stake — a level of Big Tech validation that no pure-play miner has ever achieved.
CoinShares projects that mining revenue will plummet from approximately 85% of total revenue in early 2025 to less than 20% by the end of 2026 for companies that have secured AI contracts. The remaining companies are marketing themselves as "compute" or "digital infrastructure" operators while maintaining the optionality to switch between mining and AI hosting depending on relative profitability.
The scale of committed capital flowing into former mining facilities is staggering:
| Company | Contract | Partner | Value | Term | |---------|----------|---------|-------|------| | Cipher Mining | AWS hosting agreement | Amazon Web Services | $5.5 billion | 15 years | | Cipher Mining | Barber Lake AI hosting | Fluidstack | $830 million | 10 years | | Cipher Mining | Additional AI hosting | Fluidstack | 56 MW expansion | 10 years | | TeraWulf | AI hosting agreement | Fluidstack (Google-backed) | $1.8B lease guarantee | 10 years | | TeraWulf | Expanded Google backstop | Google | $3.2 billion | Ongoing | | Bitfarms/Keel | HPC facility buildout | U.S. data center partner | $128 million | 2026 delivery |
Cipher Mining's AI hosting contracts now total approximately $8.5 billion in committed lease payments. Revenue from its AWS deal begins in August 2026. TeraWulf's Fluidstack agreement covers 200+ MW of capacity, backstopped by Google's $3.2 billion guarantee.
These are not speculative memoranda of understanding. They are binding, multi-year contracts with the world's largest technology companies — the kind of counterparty risk profile that Bitcoin mining has never offered.
Morgan Stanley's initiation of coverage on February 9, 2026, crystallized the new narrative. The bank rated Cipher Mining and TeraWulf as Overweight, with price targets implying 158% and 159% upside respectively. Marathon Digital received an Underweight rating — a direct penalty for its continued commitment to pure-play mining.
Morgan Stanley's thesis rests on a structural supply-demand imbalance: U.S. data center power demand is projected to surge by 74 gigawatts by 2028, against a projected shortfall of approximately 49 gigawatts. Bitcoin miners sit on the single largest portfolio of grid-connected, permitted power capacity outside of traditional utilities. They are, in effect, the last available source of rapid-deployment compute infrastructure for an AI industry that cannot build data centers fast enough.
The market reaction was immediate. Cipher shares rose 13.4% and TeraWulf gained 13% on the day of the report. Bitfarms stock surged 16% upon its rebranding announcement. Meanwhile, mining-focused names like Bitmine Immersion Technologies have declined 29% in 2026, and MARA Holdings is down 13%.
The investment calculus is straightforward: AI hosting contracts offer predictable, long-duration revenue streams with investment-grade counterparties, while Bitcoin mining revenue is volatile, declining on a fixed halving schedule, and subject to network difficulty that no individual operator controls.
The economic logic driving the pivot is irrefutable. But the consequences for Bitcoin's network deserve scrutiny.
When publicly traded miners — which collectively represent a significant share of global hashrate — reallocate power capacity from ASICs to GPUs, the network loses hashrate from its most transparent, regulated, and geographically distributed operators. What remains is increasingly dominated by private operations in jurisdictions with minimal regulatory oversight and the lowest electricity costs — often subsidized or state-adjacent power in Central Asia, Russia, and parts of Africa.
This concentration effect is the mirror image of what happened after China's ban: hashrate migrated to North America, improving the network's geographic distribution. Now, North American capacity is migrating away from mining entirely, potentially reversing those gains.
Furthermore, the mining difficulty adjustment mechanism ensures that the network continues to function regardless of hashrate levels. But "functioning" and "secure" are not synonyms. A lower, more concentrated hashrate raises the theoretical cost-to-attack ratio and introduces geopolitical dependencies that Bitcoin's original design sought to eliminate.
The counterargument is that difficulty adjustments will make mining profitable again for remaining operators, attracting new entrants. This is mechanically true. But the new entrants are unlikely to be publicly traded U.S. companies subject to ESG reporting, energy disclosure, and securities regulation. The quality of hashrate — not just the quantity — matters for long-term network resilience.
The Bitcoin mining industry's AI pivot represents one of the largest capital reallocation events in crypto history. An entire sector built on the premise that converting electricity into BTC was the highest-value use of compute infrastructure has discovered, decisively, that it is not. AI training and inference workloads now command multiples of what any blockchain consensus mechanism can pay per megawatt-hour.
For investors, the message is clear: the companies with the highest probability of generating returns in this cohort are those moving fastest away from their original business model. For Bitcoin maximalists, the message is more uncomfortable: the network's security budget — already under structural pressure from the halving schedule — is now competing for physical infrastructure against the most capital-intensive technology buildout in human history.
The miners aren't hedging. They're leaving. And the hashrate they take with them isn't coming back.