The alarm came from a man who builds data centers for a living. Greg Friedman, CEO of Peachtree Group—a firm that has deployed over $3 billion in hospitality and data center real estate—stated plainly in a recent interview: the data center sector is in a bubble. He warned that the AI-driven construction frenzy is outpacing actual demand, and that this poses a potential risk to the crypto mining industry. The crypto media picked it up as a footnote. I read it as an echo of 2017.
Let me give you context. In late 2017, I was auditing a Sydney-based project that promised 1,000% APY. The whitepaper had all the right buzzwords—decentralized, trustless, revolutionary. Their tokenomics table showed 60% for the community. But after six weeks of modeling their liquidity pools against SEC securities laws, I discovered that 40% of their tokens were unvested, held by a single wallet cluster controlled by the team. I flagged it as a potential Ponzi. The exchange delisted the project before it could dump. That experience taught me two things: first, technical elegance does not equal security; second, when a builder tells you there's a structural flaw, you listen.
Friedman is a builder. His warning is not about Bitcoin's fundamental value or the disruptive potential of AI. It's about the physical layer—the concrete, power lines, and cooling towers that both AI training and Bitcoin mining depend on. The data suggests a mismatch between supply and sustainable demand.
The Core Mechanics: A Clash of Capital and Physics
First, let's quantify the problem. AI model training consumes massive amounts of electricity—a single GPT-3 training run required roughly 1,300 MWh, about the same as 130 average U.S. homes for a year. Today, the largest data center campuses are designed to handle 300–500 MW per facility. Bitcoin mining, by comparison, consumes around 150 TWh annually, with a single ASIC miner drawing 3–4 kW. Both sectors compete for the same finite resources: high-capacity power substations, cheap land near renewable sources, and skilled labor for facility management.
The bubble Friedman refers to is the speculative overbuild of data centers financed by cheap debt and exuberant AI hype. According to industry reports, hyperscale data center construction reached a record $150 billion in 2024, with nearly 40% of that in pre-leased space that remains unoccupied. A 2025 survey by JLL shows data center vacancy rates in primary U.S. markets have dropped below 3%, but secondary markets—like Phoenix, Ohio, and Northern Virginia—are building at a pace that will take 5–7 years to absorb. In the meantime, operators lock in long-term power purchase agreements at peak prices, passing the cost to tenants. When demand softens—and it will, because AI currently accounts for only 10% of total data center power consumption, with the rest coming from cloud computing, streaming, and enterprise workloads—those fixed costs become stranded assets.
For Bitcoin miners, this is a structural headwind. Most of the publicly listed miners (Riot Platforms, Marathon Digital, CleanSpark) operate under fixed-price or cost-plus power contracts. However, a growing number of new entrants are signing variable-rate agreements tied to the local wholesale electricity market. When AI-driven data center demand spikes, wholesale power prices rise. In Texas, for example, the ERCOT market saw winter peaks of $5,000/MWh in early 2025, compared to $50/MWh in calm periods. Miners with variable contracts faced a 100x cost swing. This is not a bug—it is a direct consequence of competing with AI workloads.
Let me show you the math. A typical modern ASIC miner (Bitmain S21) consumes about 3.5 kW and produces 200 TH/s. At $0.04/kWh, the daily electricity cost per unit is approximately $3.36. At Bitcoin price of $70,000 and network difficulty of 80T, the daily revenue per unit is roughly $12.80, yielding a margin of 74%. Now assume power cost doubles to $0.08/kWh due to AI demand. The margin collapses to 48%. That is survivable. But at $0.12/kWh, the margin drops to 22%, and the breakeven Bitcoin price rises to $55,000. Any sustained correction in Bitcoin price would push those miners into negative territory.
The most vulnerable are the small-to-medium miners who cannot negotiate long-term fixed contracts. They rely on co-location services provided by data center operators who serve both AI and crypto clients. When AI customers are willing to pay a premium for guaranteed uptime, the operator reallocates power capacity to them, leaving miners with intermittent service or higher charges. I have seen this pattern before. In 2020, I audited a Compound Finance governance contract and found a rounding error that could have allowed a whale to extract $2 million in arbitrage. The devs fixed it, but the root cause was the same: a conflict between incentive structures—in that case, between the borrow rate model and actual market supply. Here, the conflict is between AI's insatiable GPU demand and mining's fixed capital allocation.
The Contrarian: What the Bulls Are Right About
Every bubble has a kernel of truth. AI workloads are real and growing. McKinsey projects that data center power consumption will triple by 2030. This is not a made-up narrative. The question is not whether AI will need more compute, but whether the current capital deployment linearizes that growth curve or overshoots it.
Bitcoin mining, too, has demonstrated surprising resilience. During the 2022 downturn, despite the collapse of FTX and the Terra fallout, hash rate continued to climb. Miners diversified into high-performance computing (HPC) services, offering their underutilized GPU capacity for AI inference workloads. Hut 8 and Core Scientific have already signed multi-year contracts with AI startups generating $20–30 million in annual revenue from these side activities. If the bubble in standalone AI data centers bursts, some of that physical capacity may become available for mining at lower rates—a silver lining.
But Friedman's warning is not about the long-term viability of either industry. It is about the timing mismatch. The debt used to finance current construction comes due in 18–36 months. If the AI demand doesn't materialize fast enough, those facilities become liabilities. Miners who locked into multi-year variable-rate contracts with those operators will face renegotiation or default. In the absence of data, opinion is just noise. The data says: 40% of new data center projects have no binding tenant commitments. That is a red flag.
The Takeaway: A Call for Structural Auditing
Friedman's warning is a signal that the commodity cycle is turning. Miners should treat it as a trigger to re-evaluate their power contracts, diversify their geographic exposure, and stress-test their models against a 50% increase in electricity costs. The ones that survive will be those that audit their own operational dependencies with the same rigor that I apply to smart contract code.
Are you prepared for the silence when the data center lights go out? If not, your hash power is collateral for someone else's bet on AI infinite demand. That is not a strategy. That is a margin call waiting to happen.