GPU-agnostic
G12 works at the power shelf, not inside the accelerator, so it smooths any GPU or AI chip in the rack.
Technology
Smoothing at the source: inside the rack.
A power chip, firmware and an energy-storage module work together to absorb GPU power swings before they leave the rack.
How it works
Three parts, one job.
01
Power chip
Sits in the power shelf and moves energy between the rack bus and local storage. It does this fast enough to follow GPU load steps as they happen.
02
Firmware
Watches the load and decides, moment to moment, when to store energy and when to release it. The goal is to keep the draw the facility and grid see flat and predictable.
03
Energy-storage module
A compact module in the power shelf that charges in the dips and discharges at the peaks. It is sized for the swing, not for backup.
Reaction time
Closer to the load means faster.
Site batteries sit far from the GPUs, behind transformers and switchgear. By the time they react, the swing has already reached the grid. G12 reacts inside the rack, where the swing starts.
Safety certification plan
We are designing toward the safety and energy-storage standards that power-shelf makers and data centers need. We'll publish details as certification moves forward.
Want the details?
Detailed specs are shared with design partners under NDA. Talk to engineering.