[Market Analysis] U.S. Bitcoin mining companies race to pivot to artificial intelligence... History issues a warning about the "infrastructure trap"

U.S.-listed Bitcoin mining companies are competing to invest heavily in building AI data centers. Some companies have even rebranded to declare a departure from Bitcoin. However, history is full of cases where companies building infrastructure for new technologies failed to realize profits, and debates over whether this major transformation is a strategic choice or a repeat of past mistakes are intensifying.

“We Are No Longer Bitcoin Companies”

One of the largest U.S. mining companies, Cipher Mining, with a market value of about $60 billion (roughly 87 trillion Korean won), recently rebranded as “Cipher Digital,” officially announcing a shift to AI and high-performance computing infrastructure. The company has sold 49% of its core mining assets and signed a 15-year, 300MW lease agreement with AWS, as well as a 10-year, 300MW lease with FluidStack, securing a total of 600MW of committed capacity.

Ben Gagnon, CEO of Bitfarms, stated in an interview with CoinDesk, “We are no longer a Bitcoin company.” Although the company name still includes “Bit,” its business focus has completely shifted to AI.

Looking at the movements of major mining companies, IREN Limited signed a five-year, $9.7 billion contract with Microsoft in April this year; TeraWulf reached a 10-year, over 200MW contract supported by Google-backed FluidStack; Hut 8 signed a 15-year, 245MW lease agreement supported by Google; and Core Scientific partnered with CoreWeave, which handles workloads for Microsoft and OpenAI, expanding its HPC business to 270MW.

Riot Platforms signed a 10-year, 25MW lease agreement with AMD and is evaluating transforming its Coscicana site into a 600MW AI and HPC facility; MARA Holdings is working with Stroud Capital to advance a massive AI infrastructure project of up to 2.5GW.

Deteriorating Mining Profitability Accelerates Transformation

The direct background for this shift is the worsening profitability of mining. Kent Haliburton, co-founder and CEO of Sazmining, explained, “Currently, the average cost of Bitcoin mining is about $87,000, while the spot price of Bitcoin is around $70,000. Most of the industry is operating at a loss, and publicly listed miners are citing this as a reason for their business transformation.”

However, he did not hide his criticism of the corporate shift: “These companies once had all the power contracts, land, and infrastructure needed to mine Bitcoin at low cost. Now, they are exchanging these resources for lease income from Microsoft and Google. Those who once protected the Bitcoin network are now guarding racks for massive corporations, calling it a strategy. During this process, they have sold over 15,000 Bitcoins to fund their transformation.”

Overheated AI Investment… $600 Billion Gap Between Capital Expenditure and Returns

Skepticism about the AI infrastructure investment boom itself is also rising. Goldman Sachs warned in a October 2025 report that the scale of AI investments by tech giants is approaching levels difficult to justify with current returns, raising the possibility of a “bubble.”

David Chen of Sequoia Capital has pointed out since 2023 that there is a $600 billion gap between AI capital expenditure and actual returns. By 2026, the total equipment investment commitments by large-scale companies have exceeded $700 billion, yet pure AI companies’ total revenue remains insignificant. While OpenAI’s annual recurring revenue of $20 billion is impressive, it accounts for only about 3% of the overall equipment investment by these giants.

Venture capitalist Chamath Palihapitiya has publicly expressed concerns that AI infrastructure investments may not yield returns. He noted that companies might realize the risks of exposing business secrets to cloud-based AI and thus shift toward building their own AI infrastructure. In fact, a recent ruling by the U.S. Southern District Court in New York determined that legal consultations via AI chatbots do not qualify for attorney-client privilege, indicating that trust issues with AI are spreading into the legal field.

Warnings from History… The Railroad and Internet Bubble Parallels

Some industry insiders worry that this AI infrastructure boom resembles past episodes. In the late 19th century, the U.S. railroad construction boom led to competitive investments fueled by excessive debt, ultimately causing the Panic of 1873. Many bankrupt railroads’ assets were later acquired at low prices by J.P. Morgan and others.

A similar pattern played out during the early 2000s internet fiber infrastructure boom. Many fiber optic companies went bankrupt after the dot-com bubble burst, with their assets acquired at low prices by giants like Google and Meta. Infrastructure was laid, but profits ended up elsewhere.

Additionally, the surge in demand for self-hosted AI presents another variable. As exemplified by the recent open-source AI agent “OpenClaw” on GitHub, which has garnered more stars than Linux and React, the trend toward running AI locally on personal devices without relying on the cloud is gaining momentum. The latest Mac devices with unified memory architecture, which are more conducive to local AI operation, are also noteworthy. If the future of AI is not in the cloud but on personal devices, then the bets made by mining companies shifting to rent rack space to large enterprises may be misplaced.

In light of the historical lesson that “infrastructure builders do not always profit,” whether the large-scale AI transformation of U.S.-listed Bitcoin miners is a wise strategy or another historic mistake will ultimately be judged by time.