Anthropic's six-year capacity contract, Texas audits for new data centers, and Runware's portable pod place compute expansion alongside the sites, resources, and local conditions needed to host it.
Artificial Intelligence··Evening
The first site in a six-year contract
Anthropic agreed to buy $10 billion of computing capacity from Volta Infra Holdings over six years. According to THE DECODER's account of Bloomberg reporting, the first deployment is 133 megawatts, to be handed over in two phases through March 2027. The capacity will sit at a hydropower-fed site in Tydal, Norway, operated by bitcoin miner Bitdeer Technologies, with Nvidia Vera Rubin processors in the racks. Those details connect the contract's large monetary value to a specific location, operator, power source, and initial delivery size. Volta was founded in early 2026 by former managers of asset manager Brookfield. It raised $300 million in a round led by Andreessen Horowitz and Altimeter Capital, with Nvidia and Michael Dell participating, at a $2.4 billion valuation. Volta also says it assembled a $5 billion financing pool to help customers buy AI chips and secured 1 gigawatt of power for near-term data centers. Anthropic has issued no announcement of its own, so the publicly available terms rest on the Bloomberg report carried by THE DECODER.[1]
The information Texas wants from each site
Texas has placed a different process in front of new capacity. Governor Greg Abbott directed that every new data center project in the state be audited by the Public Utility Commission of Texas and grid operator ERCOT. The review covers a project's on-site and off-site electricity demand, water use, noise mitigation, light controls, use of tax incentives, and ownership. Siting compute is therefore being considered together with grid access, surrounding resource use, and public support. ERCOT's interconnection queue gives the decision its scale: requests rose from 233 gigawatts in January 2026 to 474 gigawatts in early August, more than doubling in under six months. ERCOT says about 90% of those requests come from data centers, while the full queue exceeds five times the system's total peak demand. These numbers describe connection applications rather than capacity already built and operating. The TechCrunch report gives no duration for the audits and no date for the halt on new projects to end. The condition Texas has imposed is a project-by-project examination of electricity, water, local effects, incentives, and ownership before new data centers proceed.[2]
What portability changes about siting
Runware's Sonic Inference Pod, unveiled the same day, puts inference capacity into a modular unit that can be transported as one piece. The company says its closed-loop cooling needs no water, that pods can be installed in days rather than the months or years required for a conventional facility, and that capacity can grow by adding units. It reports 10 pods operating across the United States, Europe, and Asia-Pacific, with 160 sites available. Higgsfield AI and Wix are named as customers. Runware argues that distributed compute placed closer to users provides faster inference; its claims of lower cost and higher inference quality are the company's account, without independent measurements in the report. This third development adds a smaller, portable deployment form alongside large fixed capacity bought by contract and new sites placed under audit. In all three cases, compute still needs a physical location, electricity, and an operating arrangement. The difference lies in how those components are packaged: the Tydal contract binds capacity to a named operator and hydropower-fed site, Texas subjects project information to review, and Runware proposes distributing water-free pods among available sites.[3], [1], [2]
Related columns
For more information on this topic, you can read the related columns.
