Custom Meta Tags

Hero Banner

Introduction (LC)

Application Focus: Applying Power in New Contexts, When Design Assumptions Break

Rethinking power electronics for AI & Agriculture

Power electronics rarely moves into a new application as a blank-sheet exercise. More often, established technologies and architectures are pushed into a different operating context, but sometimes the assumptions underneath them begin to fail. In today’s increasingly digital and electrified world, this is visible in two very different markets: artificial intelligence (AI) infrastructure and electrified agriculture.

Intro cont (MM)

In both cases, the technical challenge is not simply choosing a better MOSFET, converter, or topology. It is understanding what changes when familiar power systems are scaled, integrated, or exposed to challenging operating environments.

AI: Familiar Architectures, New Constraints

In AI data centres, the basic power elements are instantly recognisable: AC distribution, uninterruptible power supply (UPS) systems, rectification, DC bus structures, intermediate bus converters, and point-of-load regulation.

What has changed from traditional data centres to AI ones is the scale, power density, and load profile. The International Energy Agency (IEA) estimates that data centres consumed about 415 TWh of electricity in 2024, around 1.5% of global electricity demand, and projects that this could reach about 945 TWh by 2030 in its base case ^[1]. Accelerated servers, largely driven by AI adoption, are projected to grow their electricity consumption at around 30% annually ^[2].

Electric tractor connected to a charging station in a farm setting, with agricultural machinery visible in the background.

Electrified agriculture exposes challenges – not hyperscale density, but energy availability, duty cycle, and operating environment.

Main body (LC)

This matters because the average rack assumption is no longer a useful design proxy for the highest-value workloads. Uptime Institute’s 2024 survey still found 4 kW to 6 kW racks to be the most common deployment, with a calculated average typical rack density of 8 kW, or 7.1 kW if a small number of above-50 kW sites are treated as outliers ^[3]. However, AI racks now sit well outside that historical envelope, with modern accelerated compute platforms pushing rack power from single-digit kilowatts into the 100 kW-plus range.

At these power densities, electrical and thermal constraints that were previously manageable become system-defining. The DC link is still a DC link, but current, copper, connector losses, busbar inductance, protection coordination, and thermal management can become system-level constraints. Transient response also becomes more severe, since GPU and AI ASIC loading can move rapidly between idle, inference bursts, and sustained training loads. Additional conversion stages improve regulation and fault isolation, but at AI-scale rack densities they also increase thermal load, bus complexity, and cooling demand. At 150 kW rack power, each percentage point of loss corresponds to roughly 1.5 kW of heat that must be removed continuously.

This is why the industry is revisiting distribution voltage, with NVIDIA even outlining 800 VDC architectures for future AI factories, arguing that fewer AC-to-DC conversion stages and lower distribution current can reduce copper use, cable bulk, and conversion losses ^[4].

For engineers, the core assumptions of industrial DC thinking are still there, but they need to be redefined under hyperscale constraints demanding new architectures and high-performance technologies like silicon carbide (SiC) and gallium nitride (GaN).

Agriculture: Energy, Duty Cycle, and Environment

Electrified agriculture exposes the same challenge through a different constraint set – not hyperscale density, but energy availability, duty cycle, and operating environment. The tempting assumption is that tractors will follow road vehicles by replacing the engine and tank with a motor, inverter, and battery.

For some applications, especially lower-power, repetitive, or high-value crop work, that is realistic. Fendt’s e107 Vario, for example, is specified at 50 kW rated power, 66 kW peak power and a 100 kWh battery, with 22 kW AC and 80 kW DC charging ^[5]. That is a suitable operating envelope for livestock, orchard, vineyard, and controlled-environment tasks, but heavy tillage is a different challenge.

Spot Power (GBL)

Technology

Power: designing solutions with power at the core

We provide you with the right insights and expertise when you need it most, so you can make the right decisions for your product, and your business.

Power - Power components lighting up on board

SEE POWER OVERVIEW

Main body and conclusion (LC)

Ploughing and other drawbar-intensive operations demand sustained high power output for long operating periods, placing pressure on vehicle mass, charging logistics, thermal management, and critically, battery size. A sustained 200 kW field load over eight hours represents a 1.6 MWh onboard energy requirement before safety reserves and thermal margins are included. At a representative battery pack energy density of roughly 175 Wh/kg, storing that much energy would require around 9.1 tonnes of batteries.

That shifts the engineering decision from “electric or not electric” to “which task, which vehicle, which charging model?” Smaller autonomous or semi-autonomous machines can attack spraying, mowing, weeding, seeding support, and intra-farm transport without carrying a battery sized for peak annual tillage. However, a fleet of lower-mass machines also changes the power electronics requirements. Instead of one large traction inverter, there may be several smaller drives, distributed chargers, ruggedised DC/DC stages, and local energy management systems exposed to dust, mud, wash-down, vibration, long idle periods, and seasonal utilisation.

Recent modelling of full electric farming with on-field energy replenishment supports this more nuanced view. One Applied Energy study found that battery-electric agricultural systems could achieve fieldwork process time only 5% higher than a diesel baseline while consuming 37.5% of the energy, but concluded that “batteries need alternative use during periods of low farming activity to be economically competitive ^[6].” This points towards battery exchange, field charging, farm microgrids, and even vehicle-to-grid (V2G) operation as part of the design space, not afterthoughts.

Alongside the power density challenge, agriculture also brings an inherent ruggedisation requirement. Operating environments, duty cycles, and equipment lifetimes can differ significantly from road vehicle assumptions, so enclosure design, connector selection, sealing, vibration tolerance, serviceability, and long-term component availability have to be considered from day one.

The Assumptions Matter

The common lesson is that power design assumptions age quickly when the context changes. AI infrastructure has its roots in data centres, but stresses density, transient load, and conversion loss to new levels. Agriculture depends on vehicles, but stresses duty cycle, durability, replenishment, and utilisation far more than passenger cars. In both cases, the winning architecture is unlikely to be completely new. It will be a familiar power architecture adapted carefully enough that its original assumptions no longer limit the application.

Sources

Working on a project (LC)

Working on a power project?

Our experts bring insights that extend beyond the datasheet, availability and price. The combined experience contained within our network covers thousands of projects across different customers, markets, regions and technologies. We will pull together the right team from our collective expertise to focus on your application, providing valuable ideas and recommendations to improve your product and accelerate its journey from the initial concept out into the world.

WE'D LOVE TO HEAR FROM YOU!