A 525-acre site in Nueces County, Texas, was built to mine bitcoin. This week it became part of one of the largest computing contracts the AI industry has seen. Anthropic has agreed to spend $35 billion buying capacity from Lambda, an Nvidia-backed cloud provider, the Wall Street Journal reported late Monday, and some of that capacity will run through Hut 8’s Beacon Point campus, a former mining operation with access to up to one gigawatt of power.
The structure of the deal is unusual. Nvidia holds the lease on the facility, Lambda installs Nvidia chips inside it, and Anthropic buys the resulting compute. But the more interesting story is why a bitcoin miner is in the middle of it at all, and what that says about where the electricity for AI is actually going to come from.
Miners had the one thing AI could not build fast enough
Getting hundreds of megawatts connected to the grid from scratch can take years. Interconnection queues in the United States are long, transformers are back-ordered and utilities are wary of committing capacity to projects that may never be built. Bitcoin miners spent the last decade solving exactly that problem for themselves. They found cheap power, secured grid connections and built large industrial sites around them. When bitcoin economics weakened, those sites became the most valuable thing the miners owned.
CoinDesk reported in March that publicly traded miners had already signed more than $70 billion in AI and high-performance computing contracts. The pattern is now well established. TeraWulf agreed in July to supply Anthropic with about 401 megawatts at a former industrial site in Kentucky. Bitdeer signed a 16-year, $4.7 billion lease in August for 121 megawatts at its Tydal campus in Norway. Microsoft’s five-year, $9.7 billion deal with IREN for roughly 200 megawatts of Texas capacity dates to last year.
Hut 8’s numbers show how lopsided the shift has become. The company reported quarterly revenue of $74.93 million in June. Its two 15-year leases at Beacon Point cover 704 megawatts of IT capacity and carry $19.6 billion in contracted value, more than 260 times that quarterly figure. Hut 8 has not confirmed how much of that capacity is tied to Lambda and Anthropic, or whether Nvidia is the undisclosed tenant behind both leases.
Repurposed power is still power
From a climate standpoint, there is a case that converting mining sites is the least bad option. The power infrastructure already exists. The grid connection is already approved. No new land is cleared and no new transmission lines are strung to serve a project that could have gone anywhere. In Texas, where much of the mining capacity sits, some of that electricity comes from wind and solar that would otherwise be curtailed.
But repurposing does not reduce demand. It relocates it. A gigawatt is a gigawatt regardless of whether the load is hashing blocks or training a language model, and the AI load runs flatter and harder than mining ever did. Miners could switch off when power prices spiked, and were often paid to do so. An AI training run cannot pause because the grid is tight on a hot afternoon.
That is where the broader numbers become uncomfortable. A Bloomberg analysis in August counted 99 proposed gas plants tied to data centre projects that, if run at industry-standard rates, would emit about 318 million metric tons of carbon dioxide a year, enough to lift US power sector emissions by 20 percent. Amazon’s planned facility in Pecos County, Texas, has permission to release 33 million tons a year on its own. The miners’ sites came with grid connections, which is precisely why they are being bought up first. The next wave of capacity will not.
A grid designed for average conditions
The weather is the other variable. A study by climate analytics firm First Street found that 79 percent of global data centre capacity sits in regions with elevated exposure to acute hazards such as flooding, extreme wind and wildfire, and that 54 percent is in markets facing chronic heat or drought stress. Coastal Texas, with its hurricane exposure and summer heat, is not an obvious exception, and the state’s grid has faced repeated questions about its ability to serve new industrial load during summer peaks.
None of this will slow the deals. The commercial logic is too strong, and companies such as Anthropic have compute needs that dwarf what any single site can offer. What it does mean is that the AI industry’s energy problem is being solved, for now, by inheriting infrastructure that was built for a different purpose and a different load profile. That inheritance runs out. The interesting question is what gets built when it does.
Hut 8 is expected to respond to questions about which tenants sit behind the Beacon Point leases, and Anthropic’s compute footprint will be worth tracking as the Texas summer gives way to a grid planning season. For more coverage of AI and climate, visit Mylistingo.
