Ampt’s new Battery Converter helps AI data centres run more compute per megawatt of grid capacity as they transition to 800VDC architectures
The converter delivers regulated 800VDC from a DC bus shared by battery storage and grid or on-site power, providing active power conditioning that responds to GPU load changes in under one millisecond.
As the battery handles short peaks, GPUs can run without power capping, and the grid connection can be sized for average load rather than peak load. Each 2U rack-mountable unit provides 80kW, and units can be paralleled from a single rack to a full row or an entire GPU cluster. The Battery Converter is in partner pilots for data centres, with general availability in Q3 2027.
Why does millisecond response matter?
Data centres are transitioning to 800VDC power architectures for rack-scale AI systems, projected to support 1MW IT racks and improve end-to-end power efficiency by up to 5% compared with 54V systems, according to Nvidia.
AI workloads can swing a rack’s power draw between 30% and 100% in milliseconds. The grid requires a steady, predictable load, while the GPUs need a highly dynamic source. Nvidia’s white paper says energy storage “needs to fill the gap between those two conflicting requirements.”
How does the battery converter work?
In Ampt’s design, the battery, the grid, and any on-site generation all feed a common DC bus. The Battery Converter steps that bus down to a regulated 800VDC supply for the GPU racks.
When GPU load changes, the Ampt converter responds in under one millisecond, helping the battery supply or absorb the difference. At the same time, the grid or on-site generation sees steadier demand. Smoothing these peaks lets operators size upstream equipment closer to average demand, reducing equipment costs, while providing a steadier load that can help sites secure timely grid access.
The grid connects to the DC bus through a medium-voltage transformer and power conversion system and feeds batteries that operate up to 1500VDC. With a grid-forming power conversion system, the battery that absorbs GPU swings can actively support grid voltage and frequency.
These are the same transformers, power conversion systems and batteries used in utility-scale storage, enabling data centre operators to build on established supply chains and on the nearly two decades Ampt has spent deploying DC power conversion for utility-scale solar and energy storage systems.
How does the design improve efficiency?
In a power-limited AI factory, every kilowatt lost in conversion and distribution is capacity that cannot run GPUs, and operators pay for it twice: as electricity and cooling. To keep more power for compute, Ampt’s design reduces losses in the converter, storage, and distribution.
The Ampt Battery Converter reaches up to 99.2% peak efficiency. Stored energy stays DC, preventing DC-to-AC-to-DC conversions of AC-coupled storage, and the DC bus feeding the converter runs at 1100 to 1500VDC, depending on the battery’s state of charge, with the converters delivering a stable 800VDC to the GPU racks.
High voltage carries the same power at lower current, reducing conductor losses or allowing smaller conductors. Ampt’s Battery Converter is designed for first-generation 800VDC racks while delivering the benefits of higher distribution voltages that appear only on future roadmaps.
How does the converter scale?
Ampt Battery Converter units can be paralleled to any power level, enabling the same 2U, 80kW building block to serve a single gPUE rack, a full row or an entire data hall. Adding one unit provides N+1 redundancy, maintaining full power if a unit goes offline. Sites with existing or planned solar can also connect PV systems to the same battery DC bus through Ampt String Optimisers, with no additional inverter or transformer.
Aaron Gomolak, CEO of Ampt, said, “In AI factories, GPU loads swing in milliseconds, so storage has to work every minute of the day. With the Ampt Battery Converter, the same battery gives the grid a steady load and the GPUs the fast-responding power they need. For operators, that means reliable compute and a faster path to bringing new AI capacity online. For the communities around them, it means data centres that are easier on the grid.”
