AI Power Delivery

last updated 2026-08-31

Physics / mechanism

AI power delivery refers to the power electronics chain that moves electrical energy from a datacentre’s supply into AI accelerator packages. The design problem is set by the current drawn at the low core voltages of large digital logic: as accelerator power rises, delivering it at conventional distribution voltages implies proportionally larger currents, and therefore larger conduction losses and copper cross-section in busbars, connectors and board planes. Raising the distribution voltage upstream and converting closer to the load reduces those currents for the same delivered power, which is why the current wave of datacentre infrastructure work is described as moving to higher-voltage power delivery ref.

The engineering variables are therefore the choice of intermediate bus voltage, the number and placement of conversion stages, the switching devices and topologies used at each stage, and the thermal and mechanical integration of the final conversion stage near or within the accelerator package. Each additional conversion stage adds loss and volume, so the trade is between transmission loss at low voltage and conversion loss at high voltage.

The sources available treat AI power delivery as an investment category rather than describing a specific device architecture, so the mechanism detail above is limited to what the higher-voltage framing supports.

Competitive landscape

The single available source positions higher-voltage power delivery alongside interconnect as one of two strands of a continuing datacentre-infrastructure funding wave, itself sitting beside AI hardware proper and a re-emerging edge silicon segment ref. No comparison of competing conversion topologies, semiconductor material systems or vendors can be drawn from the sources supplied.

Evidence base

Frontier (open questions)

Synthesised 2026-08-31 from 1 KB sources by the resynth pipeline; citations are KB source slugs.

Recent mentions

Frontier questions