TrueNorth maps AI investment cycle shifts to power semiconductors

As NVIDIA architects a fundamental shift in how power reaches its chips inside data centers, the semiconductor investment cycle is rotating toward a category most investors have barely thought about: power delivery.

The catalyst is NVIDIA’s transition to an 800 VDC (volt direct current) power distribution architecture, a move away from the traditional 54V systems that have powered data centers for decades. Large-scale deployment is expected around 2027, and it’s designed to support megawatt-scale IT racks, the kind of infrastructure that next-generation AI factories will demand.

Why 800 volts changes everything

NVIDIA’s new architecture converts power directly from 13.8 kV AC utility lines to 800 VDC, cutting out intermediate conversion steps that waste electricity and generate heat. The result is reduced copper usage, fewer conversion stages, and better overall efficiency.

The racks these systems support will exceed 1 MW of IT load each. To put that in perspective, a single rack drawing a megawatt could power roughly 750 average American homes.

Traditional silicon-based power components struggle at 800V levels. That’s where gallium nitride (GaN) and silicon carbide (SiC) devices come in, wide-bandgap semiconductors that can handle higher voltages, switch faster, and operate at higher temperatures than conventional silicon.

Texas Instruments has already demonstrated a two-stage conversion technology achieving peak efficiency of 97.6% with power density exceeding 2,000 W per cubic inch.

The money is following the watts

Projected AI infrastructure spending across major cloud providers runs into the hundreds of billions. Historically, the bulk of that spending went toward GPUs and core compute. The 800V transition signals a structural reallocation toward power systems and energy management.

Key partners positioned to benefit from this shift include Navitas Semiconductor, Texas Instruments, STMicroelectronics, and Infineon. These companies supply the GaN and SiC devices that will form the backbone of 800V power delivery systems.

From cyclical recovery to structural demand

Power semiconductors have traditionally been a cyclical business, rising and falling with automotive production, industrial output, and consumer electronics demand. The AI data center buildout changes the character of that demand into a structural, supply-constrained expansion. Every new data center built to NVIDIA’s specifications will require these components from day one.

The risk is timing. The 2027 deployment timeline means revenue ramps for power semiconductor suppliers are still a couple of years out. And there’s always the possibility that competing architectures or unexpected technical hurdles delay adoption.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.