Legrand embeds Karman AI platform into rack PDUs and DC power shelves to unlock stranded data center capacity

French industrial group Legrand has agreed to embed the Karman AI power orchestration platform into its data center power distribution product lines, the companies announced on 6 October 2026^[1]. The deal is the first commercial partnership signed since Karman rebranded from Utilidata on 17 September 2026^[1].

What gets embedded and where

Legrand will integrate Karman across its intelligent AC rack PDUs and DC power shelves, which are marketed for high-density and AI data centers^[1]. The affected product lines sit under Legrand's data center brands - Raritan, Server Technology, and ZPE Systems. The first Karman-enabled Legrand products are expected to reach market in mid-2027.

The core claim from both parties is that the integration will allow data center operators to generate 50% or more AI tokens from the same provisioned power^[1], without adding new grid capacity.

How the platform works

Karman's platform is built on a custom version of the Nvidia Jetson Orin Nano, which the company says delivers 100 times the processing power of conventional meter-based solutions^[1]. The hardware sits at the rack level and captures power usage at microsecond resolution, enabling:

• Real-time profiling of power draw tied to instantaneous workload demand
• Predictive controls that let operators schedule more compute tasks within the same power budget
• Control signals sent to servers in under 2 milliseconds - fast enough to track the sharp swings characteristic of AI workloads

The argument is that software-only power tools operating at second-scale resolution leave a material share of provisioned capacity idle. By embedding intelligence directly in the PDU or power shelf, Karman converts that static buffer into dynamically allocated headroom.

The stranded-power problem

The partnership is framed around a structural inefficiency in how data centers are provisioned. Operators contract for a power envelope large enough to cover peak demand, but average utilisation sits well below that ceiling. A large portion of provisioned power in data centers goes untapped, according to Karman CEO Josh Brumberger - a condition that becomes commercially significant when GPU-hour pricing is the unit of competition.

Legrand CTO Swen Anderson described the integration as delivering "additional value to data center operators" by unlocking that stranded power. The platform can also connect to millions of endpoints and coordinate them to monitor and react to real-time grid conditions^[1], which positions it as a potential tool for demand-response programmes as utilities push back on large-load interconnection requests.

Background: Karman's pivot from grid to data center

Karman - formerly Utilidata - originally deployed its Nvidia-based platform inside smart meters to bring real-time visibility and power-flow control to the distribution grid. Over the past year the same technology has found its most urgent application inside data centers, where power availability has become the binding constraint on growth. The September rebrand formalised that pivot, with the company adopting the name of its platform as its corporate identity.

In June 2026, the firm raised $60.3 million in Series C funding led by Renown Capital Partners, with participation from Quanta, Nvidia, and existing investor Keyframe Capital^[1]. Nvidia's presence on the cap table is notable given that Karman's hardware runs on a custom Nvidia module and that the primary use case is maximising throughput on Nvidia GPU clusters.

The mid-2027 product availability date is the next concrete milestone to watch. Whether the 50% token-throughput claim holds across varied customer environments - Karman's own documentation notes results will vary by power infrastructure and server configuration - will determine how quickly the integration moves from partnership announcement to standard specification in high-density builds.