Karman Boosts NexGen AI Server Capacity 60% in Montreal
Software orchestration allows 16 Blackwell B300 servers to run within 10-server power budget
AI infrastructure provider Karman has deployed its AI-enabled power orchestration software at a NexGen Cloud data center in Montreal, Canada, boosting nameplate compute capacity by 60 percent without increasing provisioned grid power.
Key details
During the pilot project, Karman's orchestration software enabled 16 Nvidia B300 servers to operate safely within the electrical budget originally provisioned for ten servers. The deployment achieved 60 percent more nameplate compute capacity and token throughput while preserving full inference quality and system reliability across sustained high-power workloads.
Karman’s platform utilizes Edge AI hardware—specifically a custom version of the Nvidia Jetson Orin Nano platform providing 100 times the processing power of traditional meter-based solutions—to deliver real-time visibility for grid operators and data center managers. The software identifies, coordinates, and manages distributed energy resources across millions of endpoints, smoothing power draw spikes and allowing data centers to squeeze additional compute out of existing electrical allocations. Following the successful trial in Montreal, NexGen Cloud announced it is expanding the software rollout across its European cloud facilities.
Why this matters
Power grid constraints and long utility interconnection queues have become the primary bottleneck for expanding AI cloud infrastructure. By optimizing power delivery at the server rack and cluster level, software orchestration allows operators to significantly scale token output and GPU density without waiting years for new utility power drops or costly substation upgrades.
Context
As hyperscalers and neoclouds race to deploy megawatt- and gigawatt-scale AI clusters, power availability is restricting revenue growth. The Montreal pilot demonstrates an emerging industry trend toward software-defined power management and dynamic load balancing, complementing hardware solutions like high-voltage direct current architectures and on-site microgrids to maximize hardware utilization in power-constrained markets.
Risks and open questions
While dynamic power orchestration allows higher server packing density, operating clusters closer to power thresholds increases reliance on real-time software control systems. Open questions remain regarding how extreme multi-gigawatt workloads, sudden hardware failovers, or unexpected power spikes could impact system stability when operating without traditional physical headroom buffers.
What happens next
NexGen Cloud will roll out Karman’s power orchestration software across its European cloud data center footprint to unlock latent compute capacity across its existing European infrastructure. Industry observers will monitor whether utility regulators and grid operators recognize software-defined power capping as a standard method for increasing site capacity.
Source: Data Center Dynamics Published on AI Usage Global, author: AUG Bot
