Reason to trust
How Our News is Made
Strict editorial policy that focuses on accuracy, relevance, and impartiality
Ad discliamer
Morbi pretium leo et nisl aliquam mollis. Quisque arcu lorem, ultricies quis pellentesque nec, ullamcorper eu odio.
Lekker Capital CIO Quinn Thompson argues on X that collapsing mining economics, combined with a growing shift by public miners toward AI and high-performance compute, could turn corporate BTC treasuries into a fresh source of market supply.
“A large underappreciated headwind for Bitcoin is the disaster that which is mining economics. The only way this heals is through a decline in hashrate, which is being spearheaded by the AI compute first movers like CORZ, WULF, CIFR, IREN, etc.,” Thompson wrote.
The chart Thompson shared, frames the problem visually. It shows aggregate bitcoin holdings across major listed miners climbing sharply through 2024 and 2025 before rolling over in 2026. Thompson’s argument is not that the AI pivot is bearish in structural terms.
On the contrary, lower hashrate and less uneconomic competition could improve mining industry health over time. His point is that the transition itself is expensive, and that capex-heavy AI buildouts may force miners to liquidate BTC that had previously been treated as strategic treasury.
“While helpful to long-term health and sustainability of the network economics, it presents a dilemma for prices in the near-term as Bitcoin miners hold almost 80,000 Bitcoin on their balance sheets. As these companies pivot away from BTC mining, they 1) need capital to fund the AI buildout capex requirements and 2) have no reason to hold any BTC on their balance sheet (not that they should have before either),” he argued.
Bitcoin Miners Pivot To AI
The 2025 filings and public data make that argument more concrete. Core Scientific’s fourth-quarter results showed the business mix tilting away from mining and toward AI-related infrastructure: self-mining revenue fell to $42.2 million from $79.9 million a year earlier, while colocation revenue rose to $31.3 million from $8.5 million. Management said the decline in hosted mining reflected the “continued strategic shift” to high-density colocation. For full-year 2025, Core generated $402.5 million of proceeds from selling digital assets and ended the year with 2,537 BTC on its balance sheet.
TeraWulf offers an even cleaner read-through. The company said that in 2025 it “solidified HPC hosting as its primary growth engine,” signed more than $12.8 billion in long-term customer contracts, and built a platform with 522 critical IT megawatts under contract. Yet the legacy mining business was still being monetized as that buildout took shape: fourth-quarter digital asset revenue was $26.1 million, versus $9.7 million in HPC lease revenue, and the company’s year-end digital asset rollforward shows 1,496 BTC mined, 1,500 BTC disposed of, and only 3 BTC left on the balance sheet at Dec. 31, 2025.
Cipher and IREN show two other versions of the same trend. Cipher said it increased its focus on HPC in 2025 and signed two HPC tenants for a combined 600 MW of data center capacity. It also sold bitcoin for approximately $214.7 million during the year. By year-end, Cipher had classified $94.9 million of Black Pearl mining rigs as held for sale after signing a sublease to transition the site to an HPC tenant. IREN, by contrast, has already taken the treasury issue largely off the table: with roughly 99,900 GPUs installed or on order as of Dec. 31, 2025, it said it “typically liquidate[s] all the Bitcoin we mine daily” and therefore held no bitcoin on its balance sheet at year-end.
MARA matters for a different reason. It is not yet as far along as Core, TeraWulf, Cipher or IREN in converting mining sites into a full AI/HPC business, though it had deployed its first ten AI racks at Granbury by November 2025 and later announced a Starwood partnership for AI and HPC infrastructure. But MARA is the treasury heavyweight in the group, and its own 2025 disclosures moved in Thompson’s direction: the company said it began selling bitcoin in the second half of 2025, sold about 4,076 BTC for $413.1 million during the year, and still ended 2025 with roughly 53,822 BTC.
That is the tension in Thompson’s thesis. A miner-led shift into AI can reduce hashrate pressure and improve the long-run economics of bitcoin mining. But the bridge from mining to AI is capital-intensive, and the 2025 filings show that bridge is already being funded with BTC sales, miner disposals and site conversions. For bitcoin, that means an industry adjustment that may be constructive later can still look like overhang now.
At press time, Bitcoin traded at $72,322.
