The numbers are stark, almost Biblical in their finality. Over the past 90 days, the average Bitcoin mining profitability has dropped 45% post-halving, while the hash price—the revenue per unit of computational power—has touched levels not seen since the 2022 bear market. Yet, the same public companies that once filled their quarterly reports with nothing but hash rate bravado are now whispering a new word: AI. Not as a side hustle, but as a lifeline.
We built the temple, but forgot who the god is. The temple was designed to mint digital gold, immutable and scarce. The god was the network, the decentralized consensus of miners securing the chain. But now, the temple’s priests are looking for a new altar. They are flipping switches, converting ASIC farms into GPU clusters, and pivoting from SHA-256 to CUDA cores. The question is not whether they can do it—it is whether they should.
Let me take you back to 2022, when I was sitting in a small Copenhagen co-working space, auditing the tokenomics of a failed mining startup. The founder had raised $10 million to build a hydro-powered mining farm in Norway. He had the cheapest electricity in Europe, the latest ASICs, and a spreadsheet that showed profitability for three years. What he did not have was a plan for the halving. I remember asking him: “What happens when the block reward halves and your margin disappears?” He smiled and said, “We’ll pivot to AI.” I nodded, but I knew then that the pivot was not a strategy—it was a prayer. Now, two years later, the prayer is being answered by the market, but not in the way the founder imagined.
The Context: The Post-Halving Reality
Bitcoin’s fourth halving occurred in April 2024, slashing the block subsidy from 6.25 BTC to 3.125 BTC. For the first time in history, the halving coincided with a hash rate that had already priced in the reduction. The network’s computational power kept climbing, driven by the relentless efficiency of new-generation ASICs, but the revenue per terahash collapsed. According to data from TheMinerMag, the average hash price fell from $0.12/TH/s/day in Q1 2024 to $0.06/TH/s/day in Q2 2024. For a typical 100 PH/s farm, that means a drop from $12,000 to $6,000 daily revenue. Meanwhile, electricity costs—often the largest line item—have not moved. The result is a margin squeeze that is forcing miners to choose between shutting down machines or finding new sources of revenue.
Enter AI. The narrative is seductive: mining companies already own the real estate, the power infrastructure, the cooling systems, and the operational expertise. Why not repurpose those assets for high-performance computing (HPC) workloads, specifically AI training and inference? The market is hungry for compute. The demand for NVIDIA GPUs is insatiable. The cloud providers like AWS and Azure are at capacity. The mining companies, with their massive data centers, seem like ideal candidates to fill the gap.
But the reality is more nuanced. During my work on the “Trusted AI on Chain” whitepaper earlier this year, I spent a week with a team of engineers retrofitting a former mining facility in Sweden. The facility was originally designed for 50 MW of ASIC power. ASICs are simple: they need high amperage, constant airflow, and a tolerance for noise. GPUs are different. They require precise temperature control, lower power density, and a networking backbone that can handle terabytes of data flow. The retrofit cost was $8 million for a 10 MW conversion. The payback period? At current GPU rental rates, about 18 months—if the facility runs at 95% utilization. That is a big if.
The Core: The Technical and Economic Crossroads
To understand the pivot, we must look at the numbers not as a story of survival, but as a story of identity. A mining company is, at its core, a commodity business. It buys hardware, consumes electricity, and produces Bitcoin. The margin is thin, but the product is liquid and the market is global. An AI compute provider, on the other hand, is a service business. It sells time, not tokens. The margin is higher, but the market is fragmented, the sales cycles are long, and the customers require customization.
Let me share a data point from my own analysis. I examined the Q2 2024 earnings reports of four publicly traded mining companies: Marathon Digital, Riot Platforms, CleanSpark, and Iris Energy. All four have announced AI-related initiatives. Marathon launched a new division called “Marathon Digital HPC” with a focus on GPU cloud services. Riot acquired a company specializing in AI workload optimization. CleanSpark opened a 200 MW facility in Texas that they claim is “AI-ready.” Iris Energy, already a hybrid, announced a partnership with a Japanese AI startup.
But here is the contrarian angle that most coverage misses: the economic profiles of these companies are diverging, not converging. Marathon, with its massive centralized mining fleet, is now competing directly with cloud providers like CoreWeave and Lambda Labs. Riot, historically a low-cost producer, is trying to become a value-added service provider. CleanSpark, known for its minimal overhead, is building a new type of infrastructure. Iris Energy, already a hybrid, is the most natural fit.
Based on my audit experience with the DAO lending protocols in 2020, I learned that hybrid models are inherently unstable. When you try to be both a commodity producer and a service provider, you often end up being mediocre at both. The operational culture is different. The financial metrics are different. The risk profile is different. A mining company’s balance sheet is built on depreciation and leverage. An AI company’s balance sheet is built on recurring revenue and customer concentration. Blending them can create a Frankenstein-like entity that investors do not understand and customers do not trust.
The Contrarian: The Blind Spots of the Pivot
Let me be direct: the pivot to AI is not a guarantee of success. It is a bet on a specific set of assumptions that may not hold. First, the assumption that AI compute demand will remain high and that the supply glut of GPUs will not materialize. Second, the assumption that mining companies can compete with hyperscalers on service quality and latency. Third, the assumption that the existing power contracts can be renegotiated to support the more variable load of GPU workloads.
I want to highlight a specific case that illustrates the risk. In 2023, a small mining company in upstate New York signed a power purchase agreement with a local utility at a fixed rate of $0.03/kWh. The agreement was based on the assumption that the mining load would be constant and predictable. When they tried to convert the facility to GPU hosting, the utility demanded a new contract with a higher rate and a demand charge based on peak usage. The cost of power increased by 40%. That single change wiped out the projected profit margin of the AI pivot. The company is now trying to sublease the facility to a real AI startup.
Faith in the protocol is not faith in the people. The protocol—the Bitcoin network—is designed to be resilient. It does not care if miners pivot to AI. It will adjust the difficulty, and the remaining miners will continue. But the people—the shareholders, the employees, the local communities—they are vulnerable. The pivot is a story of survival, but it is also a story of desperation. The ledger remembers, but the heart forgets. We forget that mining companies were once the backbone of the network. Now they are becoming something else.
The Takeaway: A Vision Forward
Where does this leave us? The mining companies are at a crossroads. One path leads to becoming specialized AI compute providers, with all the complexity and risk that entails. The other path leads to continued Bitcoin mining, but at a smaller scale, with more efficient hardware and perhaps a focus on the next halving cycle. The third path—the one I find most compelling—is the path of hybrid optimization: using the existing infrastructure for both mining and AI, but with a clear separation of business units, financial models, and operational teams.
I believe that the most successful miners will be those that do not pivot, but instead integrate. They will use their mining revenue to fund AI capex, but they will not mix the two on the same balance sheet. They will build separate entities, with separate management, and separate KPIs. The mining side will remain lean and efficient. The AI side will be positioned as a venture, not a lifeline.
Authenticity is a signal lost in the noise. The noise of the market, the hype of AI, the fear of missing out. But the signal is clear: the Bitcoin network will survive, but the miners who built it are changing. They are not villains. They are not heroes. They are engineers trying to keep the lights on. And as they navigate this crossroads, we must remember that the temple was never about the machines. It was about the people who believed in the god of decentralization.
So, the question I leave you with is not whether the pivot will work. It is whether we, as an industry, can honor the original vision of the network while allowing its guardians to evolve. The code is law, but the law must bend to survive. And sometimes, the most faithful act is to let go of the form to preserve the essence.
