The biggest trade of 2025 isn’t a new token. It’s a power cord.
Over the past seven days, a strange signal emerged: TeraWulf, a bitcoin miner, signed a $19 billion lease with Anthropic—a figure that dwarfs its own market cap. CleanSpark followed with $6.6 billion. Hut 8 got rebranded by Benchmark as a “power-first data center REIT.” The initial reaction was euphoria. WGMI ETF doubled. But then the narrative snapped. The ETF dropped 34% from its peak. The market didn’t sell the news. It sold the condition.
This is not a technology shift. It’s a narrative inversion. Bitcoin miners aren’t becoming AI companies. They’re becoming energy landlords. They’re renting out what they already own: gigawatt-scale electricity access, built-out facilities, and grid connections. The code—bitcoin mining—breaks. The story doesn’t.
Context: The Mining-to-Host Migration For years, bitcoin miners lived on a razor-thin margin: the difference between the hashprice and the cost of power. Every halving squeezed that delta. Then AI labs showed up looking for compute. Not just any compute—massive, contiguous, power-hungry capacity. Miners realized their electricity contracts were the real asset. They stopped selling hash and started selling kilowatts.
I saw this pattern before. During the 2022 LUNA death spiral, trust shifted from algorithms to social consensus. I spent three weeks mapping wallet interactions in the USDe launch, tracking emotional resilience instead of TVL. The same thing is happening here: miners are betting that the narrative of computing scarcity will hold longer than the technical reality. “Code breaks. Stories don’t.”
But the story has a hidden antagonist.
Core: The Scarcity Assumption Under Stress The entire miner-to-AI thesis rests on one fragile assumption: computing power will remain scarce enough for AI labs to sign ten-year leases at premium rates. The data supports this—for now. TeraWulf’s lease implies a $1.9 billion annual rent, far above its mining revenue. Empery Digital sold its bitcoin holdings to buy miner stocks, signaling a structural shift from asset ownership to infrastructure equity.
Yet the market is smart. It already started discounting the uncertainty. The WGMI ETF’s drop isn’t random—it’s a direct repricing of the scarcity bet. Investors are asking: what happens when open-source AI models catch up? If Llama, Qwen, or Kimi K3 reach GPT-5 performance, demand for training compute could crater. The ten-year lease becomes a liability.
Based on my experience dissecting SEC filings during the ETF narrative inversion—500 pages of S-1s to find hidden regulatory signals—I learned that the biggest risks hide in plain language. The miner leases contain no granular details on termination clauses, price adjustments, or performance penalties. That’s not oversight. It’s deliberate ambiguity. The market is now pricing in that ambiguity.
“Don’t buy the chart. Buy the chaos.” The chaos is the differentiation among miners. Hut 8 gets a REIT rating; TeraWulf gets a $19B headline. But inside, the execution gap is widening. I’ve audited mining operations during the WASM Wars—interviewing 40 engineers across Arbitrum, Optimism, and zkSync. Technical superiority never won. Narrative cohesion did. The miners who can articulate a clear operational plan—not just a press release—will survive.
Contrarian: The Open-Source Paradox The contrarian angle is not that AI hype will die. It’s that open-source AI may accelerate the end of computing scarcity faster than anyone expects. The article’s source notes that “miners betting on computing scarcity are taking a highly leveraged trade.” I agree—and I see the leverage as both financial and narrative.
Consider this: if open-source models achieve parity, the primary customer for miner power shifts from a handful of deep-pocketed labs to a fragmented market of researchers and startups. Those customers won’t commit to ten-year leases. They’ll want spot pricing. Miners built for steady rental income would face a pricing collapse. The narrative flips from “AI infrastructure play” back to “commodity electricity seller.”
“Code breaks. Stories don’t.” But stories can break too—if the underlying reality diverges far enough. The current story is that miners are unique because they have pre-built, low-latency power. That’s true today. But traditional data center operators like Equinix and CoreSite are retrofitting. Energy companies are building new substations. The moat is time, not technology.
My time at NeuralLedger Labs taught me that first-mover advantage in infrastructure is fragile. We built a decentralized identity protocol in four months—beta launched, $50k seed—but scalability failures killed it. The same applies here: miners have the power, but do they have the operational rigor to maintain GPU clusters? Bitcoin ASICs are rugged. AI GPUs need precision cooling, low latency, and constant uptime. That’s a different skillset.
Takeaway: The Next Narrative Signal The market is now in a verification phase. The next signal won’t come from press releases. It will come from quarterly earnings calls. The companies that report actual AI revenue—not just lease commitments—will separate from the pack. Watch for key metrics: AI segment revenue as a percentage of total, power utilization rates, and client concentration.
I recommend tracking two specific data points: the open-source model benchmark scores relative to GPT-5, and the net inflows/outflows of the WGMI ETF. If open-source surpasses proprietary models within six months, the scarcity narrative collapses. If miner earnings show real AI revenue exceeding 30% of total, the narrative solidifies.
“Don’t buy the chart. Buy the chaos.” The chaos right now is the differentiation among miners. Some will execute. Most will not. The ones that do will become the infrastructure REITs of the AI era—priced on AFFO, not hashprice. The ones that don’t will revert to penny stocks. The story is written. The code is yet to compile.
