Lektra's EdgeStart puts 8-GPU AI servers in buildings with solar power
Lektra, Inc. has begun taking orders for EdgeStart, a self-contained cabinet with an 8-GPU NVIDIA server that building owners can power with their own solar arrays and batteries. Lektra installs and operates the system remotely, sells the computing capacity to AI customers and pays the owner 80 percent of the revenue, according to the company.
- Lektra is taking orders for EdgeStart, an 8-GPU NVIDIA server cabinet powered by a building's own solar and batteries.
- Lektra installs, operates and sells the compute through Lektra Cloud, and pays the owner 80 percent of revenue.
- EdgeStart RTX costs $310,197 with estimated earnings of about $9,700 a month; the B300 costs $840,951 with about $36,950 a month.
- Financing is offered through Blue Street Capital at 9 percent over 48 months with 20 percent down, subject to credit approval.
- Lektra says the earnings figures are estimates and not guaranteed.
How EdgeStart works
EdgeStart is aimed at commercial and industrial building owners, solar and storage sites, and energy developers who have underused space on their premises. Each cabinet holds an 8-GPU NVIDIA server, an online UPS, a metered power distribution unit, a firewall and a switch. Lektra says this is the same class of hardware AI companies rent from the largest clouds.
Lektra handles installation, remote operation, customers, billing, monitoring and support. It sells the computing capacity for inference and other GPU workloads through Lektra Cloud. Owners get a live dashboard showing usage, energy and earnings, and can start with one cabinet and add more later.
The company points to a problem on the solar side: many buildings with panels produce more power than they use, and in some markets utility tariffs and interconnection rules mean exported solar earns only a few cents per kilowatt-hour. Lektra says EdgeStart, using a patented process, offers a more efficient way to turn that power into income.
Two models, prices and estimated earnings
Lektra is launching two versions. The figures below are the company's own.
- EdgeStart RTX: eight NVIDIA RTX PRO 6000 Blackwell GPUs, priced at $310,197, drawing about 7 kW at peak. Lektra estimates owner earnings of about $9,700 a month in the first year.
- EdgeStart B300: eight NVIDIA B300 GPUs, priced at $840,951, drawing about 14.5 kW at peak, with estimated owner earnings of about $36,950 a month.
Prices cover the installed hardware and shipping. Owners can pay cash or finance through Blue Street Capital at 9 percent over 48 months with 20 percent down, subject to credit approval. Estimated monthly payments are about $6,200 for the RTX and about $16,700 for the B300.
The earnings estimates assume 730 contracted hours a month at current market rates, the 80 percent owner share, power at $0.10 per kilowatt-hour, and the cost of fiber and data storage. Lektra's release states that the illustrative revenue figures are not guaranteed earnings or investment returns, and that results depend on factors such as market demand, GPU utilization and electricity costs.
Part of the EdgeScale product line
EdgeStart is the entry point to Lektra's EdgeScale data center product suite, which also includes Edge Pro for larger commercial deployments and Edge Max for multi-megawatt installations. According to Lektra, its EdgeScale data centers, which operate at 20 MW or less, are deployed at eight sites today.
Founder and CEO Karl Andersen framed the product as an alternative to concentrated AI infrastructure. "AI infrastructure doesn't have to be concentrated behind the walls of a massive data center," he said. The release notes that large, centralized data centers can take years to develop because of construction, permitting, land acquisition, utility coordination and the need for significant reliable power, and that projects often face community concerns about electricity rates, water use, land use and grid capacity.
What it means for building owners
Lektra calls participating owners energy hosts. The process starts with a site assessment on lektra.com, followed by an engineering review and an earnings model. The host then buys the system with cash or financing, and Lektra installs, tests and brings it online. EdgeStart can be ordered now at lektra.com, and buyers can also request a site assessment.
For owners with surplus solar or battery capacity and spare space, the offer is a way to earn from AI computing rather than exporting power at low rates. The economics depend on the assumptions Lektra lists, including contracted hours and market rates, so the actual return will vary. Lektra says initial configurations support select NVIDIA GPU architectures, with additional hardware options and deployment configurations planned.
