On a quiet Tuesday, the news hit the wires like a circuit breaker thrown open: TeraWulf, a publicly traded Bitcoin mining company, had signed a $19 billion agreement with Anthropic, the AI lab behind Claude. The deal, rumored to span ten years, essentially turns a former SHA-256 hash farm into a GPU playground for large language models. It’s the largest such pivot in mining history. And it made me stop cold.
Because here’s the thing: we built Bitcoin mining to be the backbone of a permissionless financial system. We filled warehouses with ASICs to secure a decentralized ledger that no single entity could control. Now, those same warehouses are being leased out to the most centralized actors in tech—private AI companies whose models are black boxes. The irony isn’t just thick; it’s corrosive. Are we building for humans, or just for nodes?
Let me give you the context that matters beyond the press release. Meta had already been in talks with Anthropic for a reported $10 billion compute deal. That negotiation tested the upper bound of what AI companies are willing to pay for raw processing power. Then TeraWulf jumped in with a $19 billion offer—almost double. Why? Because Bitcoin miners own something AI needs desperately: cheap, stranded energy and existing industrial real estate with power rights already secured. A typical Bitcoin mine draws 100 to 200 megawatts. That’s enough to run a small city. For an AI hyperscaler, it’s just a training run.
But the transformation isn’t simple. Bitcoin mining runs on ASICs—specialized chips that do one thing: compute SHA-256 hashes. AI training runs on GPUs—Nvidia H100s, B200s, the kind that cost $30,000 each and require liquid cooling, InfiniBand networking, and a completely different power profile. TeraWulf can’t just plug in Anthropic’s servers and turn the key. They need to retrofit substations, install closed-loop cooling towers, and hire an entirely new operations team that speaks CUDA, not stratum. Based on my time facilitating the Prague Consensus Workshop in 2017, I saw miners grapple with identity. Now, that identity crisis has a $19 billion price tag.
Here is my core insight: this deal is a bet that compute is fungible, but it is not. Bitcoin mining compute is a public good—it secures a global state machine that anyone can read and write to. AI compute is a private good—it trains a model owned by a single corporation. By renting out its infrastructure, TeraWulf is converting a permissionless asset into a permissioned one. The electrons are the same, but the sovereignty is not. We are witnessing the financialization of infrastructure trust, and the ledger is being written in fine print.
Consider the numbers. $19 billion over ten years means roughly $1.9 billion in annual revenue for TeraWulf. To compare, Bitcoin mining revenue for the entire public mining sector in 2024 was around $4 billion. TeraWulf alone was pulling in maybe $300 million. This deal triples their top line overnight—if they can deliver. But the margin story is more nuanced. AI compute leases typically come with strict Service Level Agreements (SLAs) that guarantee uptime, latency, and throughput. If TeraWulf fails to meet those metrics, they pay penalties. In Bitcoin mining, if a rig overheats, you just lose a few blocks. In AI training, a cluster failure mid-epoch can waste millions of dollars in lost model progress. Execution risk is the real variable, not revenue.
From a moral framing perspective, this deal also raises a question about compute access. Bitcoin’s mining network is permissionless: anyone with the right hardware and electricity can mine. AI compute, by contrast, is becoming increasingly concentrated in the hands of three or four cloud providers and now a handful of miner-turned-landlords. If TeraWulf dedicates 80% of its capacity to Anthropic, that capacity is locked out from everyone else. Small research labs, independent developers, open-source AI projects—they get the leftovers. We are privatizing the computational commons. Build for humans, not just nodes.
Now, the contrarian angle: maybe this is exactly what the miner community has been waiting for. The 2022 bear market was brutal—many miners went bankrupt or sold their rigs for scrap. AI compute offers a hedge against Bitcoin price volatility. A stable, dollar-denominated lease is a lifeline. TeraWulf’s shareholders are cheering. And from a technical standpoint, repurposing existing power infrastructure is arguably more environmentally efficient than building new gas plants. So isn’t this a win-win?
Not quite. The counter-intuitive truth is that this deal might weaken the Bitcoin network itself. If major miners divert their best sites—the ones with the cheapest power—to AI clients, the remaining hash rate shifts to higher-cost energy sources. That could concentrate mining in places with subsidized or low-grid electricity, increasing geographic centralization. Furthermore, the counterparty risk is now enormous. TeraWulf is betting its future on one client. If Anthropic stumbles (regulatory crackdown, model failure, funding dry-up), TeraWulf’s revenue collapses overnight. Diversification into AI is not diversification away from single points of failure.
I remember facilitating a workshop during the 2022 bear market, part of my Reclaim peer-support network in Prague. A developer in the room, burned out from a failed DeFi project, told me: “I thought crypto would set me free, but it just made me more dependent on the market.” Today, I hear that same voice in TeraWulf’s boardroom. They are trading dependency on Bitcoin price for dependency on an AI giant’s whim. Education is the ultimate yield. We need to understand that true resilience doesn’t come from switching revenue streams; it comes from building systems that are robust by design—protocols that distribute power, not concentrate it.
So where does this leave us? The TeraWulf-Anthropic deal is a landmark—not because it proves the convergence of crypto and AI, but because it reveals the fault line between open and closed compute. We have to ask ourselves: Are we building a future where compute is a public utility, or a luxury good guarded by landlords? The answer will determine whether blockchain remains a tool for inclusion or becomes just another real estate play for the connected few.
Build for humans, not just nodes.
Education is the ultimate yield.
And maybe, just maybe, the next mining rig you spin up should come with a question: Who gets to use its power?
