Overview

Ampt, a Fort Collins, Colorado-based power conversion specialist, has introduced a Battery Converter aimed squarely at the emerging bottleneck in AI infrastructure: the mismatch between what GPU racks demand electrically and what a grid interconnection can actually deliver. Announced on November 11, 2026, the product delivers regulated 800 VDC from a DC bus shared by battery storage and grid or on-site power, and it responds to load changes in less than 1 millisecond. The company positions the device as a way for operators to extract more compute from an existing or planned grid connection rather than waiting years for additional capacity.

The unit is currently in partner pilots for data centers, with general availability targeted for the third quarter of 2027.

The Millisecond Problem

The engineering rationale rests on a fundamental conflict. Data centers are migrating to 800 VDC power architectures for rack-scale AI systems, which NVIDIA projects will support 1 MW IT racks and improve end-to-end power efficiency by up to 5% compared with 54 V systems. But AI workloads can swing a rack's power draw between 30% and 100% within milliseconds. The grid, by contrast, needs a steady, predictable load. NVIDIA's own white paper acknowledges that energy storage "needs to fill the gap between these two conflicting requirements."

Response time is therefore the whole point. A converter that reacts in sub-millisecond time can let the battery supply or absorb the delta while upstream generation sees a smoothed demand profile. Slower power conditioning simply passes the transient through to the grid connection, forcing operators to provision for peak rather than average load.

Architecture and Technical Details

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 800 VDC supply for GPU racks. The grid connects to the DC bus through a medium-voltage transformer and power conversion system, feeding batteries that operate at up to 1500 VDC. The DC bus itself runs between 1100 and 1500 VDC depending on the battery's state of charge, while the converters hold a stable 800 VDC output regardless.

Each 2U rack-mountable unit delivers 80 kW (100 A), and units can be paralleled from a single rack to a full row or an entire GPU cluster. With a grid-forming power conversion system, the same battery that absorbs GPU swings can also actively support grid voltage and frequency.

The Efficiency Accounting

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, once as electricity and again as cooling. Ampt's design attacks losses at three points. The Battery Converter reaches up to 99.2% peak efficiency. Stored energy stays DC, avoiding the DC-to-AC-to-DC conversions inherent in AC-coupled storage. And the higher-voltage DC distribution bus carries the same power at lower current, reducing conductor losses or permitting smaller conductors.

Notably, Ampt describes the converter as built for the first generation of 800 VDC racks while delivering the distribution-voltage benefits that appear on other vendors' future roadmaps — a positioning that implicitly argues operators need not wait for a later architecture generation.

Scaling, Redundancy and Solar Integration

The modular approach is deliberately conventional: the same 2U building block serves a single GPU rack, a full row or an entire data hall, and adding one unit provides N+1 redundancy so that full power is maintained if a unit goes offline. Sites with existing or planned solar can connect PV systems to the same battery DC bus through Ampt String Optimizers, with no additional inverter or transformer — a nod to the company's heritage in utility-scale solar DC optimization.

Market Context

Ampt's core argument is supply-chain leverage. The transformers, power conversion systems and batteries in this architecture are the same components used in utility-scale storage today, so data center operators can build on established supply chains rather than bespoke data center power equipment. The company cites nearly two decades of deploying DC power conversion for utility-scale solar and energy storage systems, and describes itself as the number one DC optimizer company for large-scale systems, with sales and support across North America, Europe, Japan and Australia, plus representation in Asia and the Middle East.

Implications for AI Data Centers

The practical consequence is interconnection economics. If batteries absorb short peaks, GPUs can run without power capping, and the grid connection can be sized for average load rather than peak load. That smoothing lets operators size upstream power equipment closer to average demand, reducing equipment costs, while presenting a steadier load that may help sites secure timely grid access — increasingly the binding constraint on AI capacity expansion.

"In AI factories, GPU loads swing in milliseconds, so storage has to work every minute of the day," said Aaron Gomolak, CEO of Ampt. "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 centers that are easier on the grid."

Future Outlook

With pilot deployments underway and volume availability not expected until Q3 2027, the Battery Converter's real test will be whether sub-millisecond buffering holds up under the duty cycles of production AI training and inference fleets, where storage is cycled constantly rather than reserved for rare peaks. The broader signal is architectural: the 800 VDC transition is pulling utility-scale storage technology — grid-forming converters, high-voltage DC buses, DC-coupled batteries — directly into the data hall, collapsing a boundary that historically separated the substation from the rack.