Public Bitcoin Miners are Dumping Bitcoin for AI, a Historic Mistake
Bitcoin Magazine Public Bitcoin Miners are Dumping Bitcoin for AI, a Historic Mistake There is no doubt about it, this is the age of AI. Corporations are cutting their workforces in half to invest cash flow into hardware, while the stock market remains near all-time highs, mostly thanks to FAANG. OpenClaw, a self-hosted AI agent, has more stars on GitHub than Linux and React, while even Jack Dorsey is taking harsh measures to restructure Block in the face of digital, artificial intelligence. But how much of this AI wave is hype, and how many of the companies that build its infrastructure will actually capture the profits? Public Bitcoin miners in the United States have made their choice, a variety of them committing capital to building out AI datacenters, and some even making full rebrands, distancing themselves from the orange coin. While there’s a full range of AI-related pivots and statements made by public Bitcoin miners on the matter, a couple stand out as the most radical. Cypher Mining, estimated to be worth around six billion dollars — placing it among the biggest in the country – announced a full rebrand away from Bitcoin and on to the AI hype train. In their most recent investment report titled “Rebrands to Cipher Digital to Reflect Strategic Shift Toward HPC,” the company explained why they “Divested 49% Stake in Alborz, Bear, and Chief Mining Sites”. Bitfarms Ltd, another large public miner valued at over a billion dollars, also made a full pivot to AI. The CEO, Ben Gagnon, went as far as saying “We are no longer a Bitcoin company,” as reported by Coindesk, though they did keep the ‘Bit’ in the name. Some of these public companies are expecting more dollar returns from digital intelligence than those they get from Bitcoin, at least in the short to mid term, while other are others might consider it a diversification or an opportunity too large to miss. Kent Halliburton — Co-Founder & CEO of Sazminingexplained to Bitcoin Magazine in an exclusive interview that “The average cost to mine a bitcoin right now is about $87,000. The spot price of bitcoin is about $70,000. So most of the industry is underwater, and the public miners are using that as their excuse to pivot.” Sazmining is a private Bitcoin miner that specializes in frontier energy sources, with operations mostly outside of the United States. Halliburtonalso noted that “$87,000 is an industry average — it includes guys running old-gen rigs on grid power in Texas. At our sites in Paraguay and Ethiopia, our clients are producing bitcoin on an energy cost basis of $50,000 to $64,000, on 100% renewable energy. That’s a 10 to 30 percent discount to spot. The profitability is right there.” It just requires a longer investment horizon or cheaper energy, neither of which appears to be actionable for American public miners who have dollar-denominated quarterly reports to target. On the topic of cheaper energy, however, Halliburton suggests that public U.S. miners had the chance to be competitive but have failed to take advantage of their resources. He minced no words on the topic, saying that these public companies “had the power contracts, the land, the infrastructure — everything you need to mine bitcoin cheaply — and they’re handing it to Microsoft and Google in exchange for lease checks. They went from securing the Bitcoin network to securing rack space for hyperscalers, and they’re calling it a strategy. Meanwhile, they’ve dumped over 15,000 bitcoin off their balance sheets to fund the transition”. Of the biggest public Bitcoin miners, IREN Limited began its pivot to AI cloud services in April 2025, announcing a$9.7 billion, five-year agreement with Microsoft for 200 MW of critical IT load using NVIDIA GB300 GPUs. TeraWulf has executed multiple Google-backed HPC expansions through Fluidstack, securing 10 year agreements for over 200 MW. Cipher Digital completed its full rebrand to an HPC landlord with 600 MW of contracted capacity, including a 15-year, 300 MW lease with AWS and a 10-year, 300 MW lease with Fluidstack backed by Google. Hut 8 signed a 15-year, 245 MW lease with Fluidstack, also backed by Google, eyeing future possible extensions and a right of first offer for over 1,000 MW. Core Scientific has expanded its HPC focus to 270 MW through partnerships with CoreWeave, which serves Microsoft and OpenAI workloads. Riot Platforms is strategically evaluating an AI hosting expansion by partnering with AMD on an operational 10-year, 25 MW lease and assessments for 600 MW of AI/HPC at its Corsicana site, though no hyperscaler agreements have been announced. MARA Holdings is diversifying into AI through a joint venture with Starwood Capital’s Starwood Digital Ventures, targeting 1 GW of near-term IT capacity expandable to over 2.5 GW for hyperscale and AI workloads, with Starwood leading financing and tenant sourcing, but without named hyperscaler contracts yet. CleanSpark is pursuing a pivot to AI by acquiring Texas land and power for AI/HPC, including 447 acres in Brazoria County for 300–600 MW potential and an Austin County site contributing to 890 MW aggregate, funded by Bitcoin sales, with tenant discussions ongoing but no hyperscaler leases disclosed. So the AI gold rush is here, there’s no doubt about it, many of these public miners apparently see an opportunity to build out the infrastructure of — what is without a doubt— a profound technological trend. But history has not been kind to those who build the infrastructure of a new era, not in the long term anyway. It tends to be a very high-risk, medium-reward kind of bet. How many of the companies that built the railroads — for example — are still around today? Or, without going back that far, can you name any company that built out internet fiber lines in the late 90’s and 2000’s? There is a long list of railroad bankruptcies from the late 1800’s, which even led to a financial crisis in what’s called the Panic of 1873, many overleveraged in debt to fund build-outs for which there was not enough demand yet. After the panic, J.P. Morgan led a consolidation of bankrupt railroad companies, resolving debt disputes and bringing their real estate assets under new ownership. It was they who ended up capturing the upside of the railway build-out. And just around the corner of the century, the dot com bubble of the 2000’s left a graveyard of fiber line infrastructure companies that were also, in the end, bought out by hyper scalers like Google and Meta during the post crash consolidation, for pennies on the dollar. While both the railway and fiber line build-outs overall helped scale commerce for the world in incredible ways — demonstrating the overall wisdom of the markets — most individual companies involved did not survive the process, and venture capitalists looking at the AI boom today are aware of this dynamic. The Capex vs Revenue AI Gap Various investor groups are starting to question where the returns on this massive infrastructure spending will come from. In an October 2025 report titled “AI: In a bubble?”, GoldmanSachs took a argued that, while the investments so far could be supported by big tech revenue, the valuations of some of these companies were starting to get “frothy”. David Chan at Sequoia has been pointing out the growing gap between AI-driven revenue and capital expenditures (Capex) since 2023, leading to a widely reported number of a $600 billion gap between them. Capex spending commitments in 2026 are north of $700 for the hyper scalers, but where are the returns? OpenAI’s $20 billion annual recurring revenue (ARR) is impressive for a new company, but that represents “roughly 3% of the projected 2026 hyperscaler capex total” as reported by FuturumGroup, who noted that “Anthropic’s $9 billion run rate, while showing 9x year-over-year growth, occupies a similar position. The entire cohort of pure-play AI vendors – including Cohere ($150 million ARR), Mistral (~$400 million), Perplexity ($148 million annualized), and others – likely accounts for less than $35 billion in projected combined 2026 revenue.” Skepticism about where the value of AI will actually be captured has also been aired by VC’s like Chamath Palihapitiya. He was a prominent investor in Groq, a company building custom silicon for the AI age, which was licensed by NVIDIA in a $20 billion deal last year, and was a Facebook insider through the company’s rise to become a hyperscaler. If he has his doubts about the profitability of building the railroads of artificial intelligence, then perhaps there’s something worth giving a very close look at. Palihapitiya also argued in a recent All In Podcast that corporations might soon start to realize they are exposing their trade secrets to cloud AI, preferring instead to self-host. Building out in-house GPU farms might seem like a bit of a side quest, but can you really risk your trade secrets with AI providers who train on user data? After all, new versions of models trained on that data will have it in their knowledge base, exposed to the world. And even if corporate agreements not to train on corporate data become the norm, a very high trust relationship would be formed, posing a systemic risk to certain corporations, a risk that the data might get leaked or seen by the wrong insiders inside the cloud AI provider companies. There are also questions about whether the market fundamentally wants cloud AI for the same reasons. Would you hire a personal assistant if you knew the data you share with them would end up on the internet? Probably not, but that’s what’s happening with AI. In fact, the U.S. Southern District of New York recently ruled that users do not have client-attorney privilege when getting legal help from AI chatbots, and thus, sensitive discussions with AI could be legally subpoenaed and used against the clients in a court of law, a sign of the risks involved with trusting AI blindly. Some speculate that new kinds of terms and agreements will need to be formed to support this use case. But the legal case points to a fundamental element of the demand for AI: people want humanoid intelligence, digital or otherwise, that they can trust. AI Loyalty and Trust Ah, “Trust”, that ubiquitous, almost supernatural word that does so much work to carry the weight of the world. But what is trust? Fundamentally, it is predictability, one person’s confidence that another human, system, or AI will behave in a certain way, in a reliable, predictable, and positive way towards one’s interests. AI, when hosted in the cloud, however, can not give such assurances; the data is fundamentally leaving the user’s machine to be processed by “the cloud,” and what happens up there is beyond us mortals. In fact, “the cloud” has legal risks that might prevent it from being loyal to you as a user in certain scenarios. Hence, perhaps the public’s fascination with OpenClaw. In recent weeks, a new open source project in the AI world has taken the tech industry by storm. 289,000 stars on GitHub, more than Linux has gotten despite supporting the software infrastructure of the world, more than React, one of the most popular web development languages in the world. And it’s only been live for what, weeks? How could this be? Why do people like it so much? Well, arguably two reasons. It feels more like a human assistant than a chatbot; it updates itself, remembers what you are interested in, journals, and develops around your preferences. But most important of all, you can host it on your machine. People were buyingMac minis in droves to run OpenClaw, pairing it up with Claude Max API token plans of about $200 a month. Some argue this is a revolution in self-hosting, even though the above setup is still dependent on the cloud. But what’s actually happening here is that OpenClaw appears loyal, it remembers you, it is “in your home” in your PC. It’s not a chat interface whose context window will eventually become too much for it to manage, ending in a small death, replaced by a new chat tab. OpenClaw is not a chatbot; it’s an AI entity of sorts that users create a relationship with. And good relationships are built on trust. So what does all of this have to do with public Bitcoin Miners? Well, perhaps self-hosted AI is the future, Chinese AI models are increasingly leaner and can run on machines far from the cutting edge, arguably pressured into innovation by sanctions on specialized AI hardware like high-end Nvidia chips. Open source tools of all kinds that manage and host models locally are regularly launched and improved, and if history is any guide, the mass production of AI hardware will lead to the commoditization of powerful computers that will make it to end users’ homes, and can handle AI. In fact, Apple, the FAANG that has had the worst AI products deployed to date, may end up becoming one of the biggest winners of the AI race. Why? Because their user hardware is excellent. Recent Macs don’t have a distinction between RAM and VRAM, an issue all other computers dependent on GPUs, such as Nvidia, have. This limits the size and speed of models that can be self-hosted. Instead, all RAM is unified in the latest Mac machines, letting users run powerful models locally that don’t easily run on non-Apple hardware. Self-hosted AI is the future. And thus, public Bitcoin miners, in the pursuit of mid-term fiat gains, might have just fallen for a trap. The same trap the giants of the dot-com bubble fell for. The same trap that the titans of the industrial era, who built the railroads, fell for. The infrastructure that runs the future does not necessarily capture the gains. This post Public Bitcoin Miners are Dumping Bitcoin for AI, a Historic Mistake first appeared on Bitcoin Magazine and is written by Juan Galt.
