Lambda is betting $33 billion on Alabama. The race now is who fills the building

The Alabama bet

Lambda plans to build a large-scale data centre campus in Lowndes County, Alabama. The Red Clay Center Campus represents an estimated $33.1 billion investment in Alabama, according to the company. Projected local tax revenue for Lowndes County is more than $2.8 billion through 2049, with Lowndes County Schools projected to receive more than $1.6 billion, and combined state and local tax revenue projected to reach $3.4 billion through 2049. The project is expected to create more than 1,000 construction jobs and support more than 225 full-time roles. m:…3842

Construction is expected to begin in the first quarter of 2027 and take up to five years, according to Lambda. The facility will deploy closed-loop cooling technology to reduce water usage, and Lambda will pay the full cost of power and grid infrastructure to protect Alabama Power ratepayers. Lambda also plans to partner with N-Drip to support local farmers in water conservation, and is investing $10 million through a Community Benefits Agreement with the Community Foundation. The package is designed to answer the two objections every giant data centre now faces: who pays for the grid, and who pays for the water. m:…3842

A new neocloud, built for inference

Blackfuel emerged from stealth this week as an inference-native neocloud provider, according to AMD's account. The company has secured more than $250 million in contracted revenue under multi-year customer agreements, and its first inference-only deployment is located in Barcelona, hosted by Digital Realty and built on liquid-cooled Dell PowerRack systems featuring Dell PowerEdge XE9785L servers. The Barcelona deployment utilises AMD Instinct MI355X GPUs and AMD ROCm open software. m:…9911

Blackfuel's platform is designed as an AI Token Grid, a network of accelerator clusters connected through a common software and commercial layer, and it is introducing Logical GPUs, which allow customers to reserve a share of a model's serving capacity for a fixed monthly fee. Its infrastructure uses open serving technologies including vLLM and SGLang, and its Token Platform plans to support models from Meta, Mistral, H Company, DeepSeek, Qwen, GLM, Moonshot AI, OpenAI, MiniMax, Cohere, Reflection AI, NVIDIA, Liquid AI, and Google at launch. Blackfuel plans to offer dedicated GPU clusters, managed inference, and reserved serving capacity. The pitch is inference as a reservable utility, not training as a rental. m:…9911

The ramp is where the money is made or lost

Aethir's warning this week is that building the capacity is the easy half. The company describes the GPU cluster ramp, the interval between a facility's go-live and reaching steady, paying utilisation, as the critical economic phase, because depreciation costs often outpace revenue. A fully costed GPU cluster breaks even between 55% and 65% utilisation when accounting for three-year depreciation, according to Aethir, and depreciation begins the day the hardware arrives at the facility. Monthly depreciation for an 8x H100 node is approximately $7,000 to $10,000. m:…8504

The utilisation gap is wide. Aethir says enterprise Kubernetes clusters average 5% GPU utilisation, while the Meta Research SuperCluster reports GPU utilisation between 83% and 85%, and Aethir's own GPU network maintains a constant utilisation rate of 95% or higher, according to the company. Its answer is to build data centres only when compute demand is already secured through its ACCELERATE program, and it notes NVIDIA requires a 24-hour soak test under representative inference load before a provider serves production traffic. Global AI capital expenditure is projected at $400 billion to $450 billion in 2026, rising toward $1 trillion by 2028, according to Aethir. The question for every builder in that wave is the same: who is the paying customer on day one. m:…8504

Lambda is also tightening the operational side. The company introduced Identity and Access Management to its GPU cloud platform, enabling role-based access control for users, groups, and service identities. To act on a resource in a workspace, a user needs both membership in that workspace and a role that allows the action. Audit logs for keys, identities, workspaces, and resources are recorded automatically and kept for six months, and the IAM features are free to use, according to Lambda. A $33 billion campus needs customers who trust the access controls as much as the hardware. m:…1440