Look at the chart. WGMI, the Valkyrie Bitcoin Miners ETF, has dropped 34% from its peak. Meanwhile, the hash price is stable, and Bitcoin itself hasn't cratered. The narrative is breaking, and the data is sending a different signal.
For the uninitiated, 2025 saw a tectonic shift. Bitcoin miners—TeraWulf, CleanSpark, Hut 8—stopped talking about hashprice and started pitching themselves as ‘energy-first AI data center REITs’. The hook was simple: AI labs need gigawatts of power, and miners already have the land, the substations, and the grid interconnects. TeraWulf signed a $19 billion lease with Anthropic—a figure larger than its entire market cap. CleanSpark followed with $6.6 billion. The stocks doubled. The narrative was perfect.
Let me be clear: this is not a technology upgrade. It is a resource arbitrage.
The core of the argument rests on one fragile assumption: that compute will remain scarce enough for model labs to keep signing 10-year leases. The entire valuation thesis—moving from a ‘hash price multiple’ to an ‘infrastructure cash flow multiple’—depends on that scarcity holding. The data, however, is already showing cracks.
First, the market moves. Based on my audit of the WGMI flows and the individual stock reactions, the behavior is textbook ‘buy the rumor, sell the news’. When the TeraWulf lease was announced, the stock initially surged, but within weeks, the entire sector retraced. The ETF is down 34% from highs. This is not a market that believes the story. The code does not lie, only the narrative.
Second, the counter-argument is emerging directly from the AI side. Open-source models—Llama, Qwen, Kimi K3—are closing the gap with closed-source behemoths. If an open-source model matches GPT-5 on key benchmarks, the demand for proprietary training compute plummets. Whales do not whisper; they shake the ledger. A single model breakthrough can collapse the scarcity premium that justifies a $19 billion lease.
Here is the contrarian angle that most analysts miss: the lease value is being priced as income today, but it is a 20-year stream of future payments. The market is capitalizing the entire contract value into the stock price immediately. If in year three, Anthropic decides to use a smaller model, or goes bankrupt, or compute becomes cheaper, that 20-year stream collapses. Pegs break, principles remain, portfolios vanish.
The data also reveals a hidden execution risk. Based on my 2017 ICO due diligence, I learned to look at the team, not the press release. The technical stack for running a Bitcoin ASIC farm is fundamentally different from managing a 100MW GPU cluster for AI training. The cooling requirements, the networking latency, the uptime SLAs—none of this exists in the miner playbook. These companies are signing leases to build data centers they have never operated. The market is pricing a REIT, but the operational reality is a construction project with technology risk.
Further, the competitive landscape is not static. The miners are not competing against each other; they are competing against Equinix, CoreSite, and every energy company with a spare plot of land. A utility that builds a new 500MW substation can offer the same ‘power lease’ to an AI lab. The miner's advantage—‘already built’—erodes quickly as AI labs become more willing to build their own facilities. Audits reveal the skeleton, not the soul. A lease is a contract; it is not a moat.
The signal I am watching now is the Q3 earnings calls. The market will stop caring about the lease headline and start demanding AI revenue, AI margin, and AI operational metrics. If the companies report only Bitcoin mining income and say ‘transformation is in progress’, the stocks will get cut in half again. Volatility is the tax on ignorance.
To the traders still holding: check the token flows. Empery Digital sold its Bitcoin position to buy miner stocks. That is a fund rotating from a macro asset to a micro infrastructure bet. It is smart capital, but it is also a warning. When the smartest money in the room is already positioned, the marginal buyer is gone. The only question is who sells last.
So here is my takeaway: The miner-to-AI narrative is a high-leverage bet on compute scarcity. The data is already showing that scarcity is an assumption, not a law. The market is beginning to price that risk. Watch the open-source model benchmarks, watch the next earnings call, and trace the wallet, not the headline. Trace the wallet, ignore the tweet.