Q: What are power semiconductors?
Power semiconductors are essential chips making electricity usable for technology ranging from phones to EVs, and have a key role in the AI infrastructure buildout. Every watt consumed by an AI processor must first be converted, regulated and delivered by power semiconductors. Think of them as the traffic controllers for electricity. They control and convert electricity—ensuring the right voltage and current reach processors, servers, and cooling systems—safely, reliably and efficiently.
Q: How do they actually work?
Power semiconductors act as ultra-fast electrical switches. They turn power on and off thousands—or even millions—of times per second to convert power between different voltages and control how electricity flows throughout a system. Every watt that powers an AI workload is shaped, stepped down, and stabilized multiple times by power semiconductors before it ever reaches a GPU.
They perform three essential functions:
- Convert electricity into usable forms
(e.g., high-voltage AC from the grid → low-voltage DC required by chips)
- Regulate voltage and current
(ensuring AI processors get stable, precise power)
- Switch power on and off extremely quickly
(to minimize energy loss and match real-time compute demand)
Every time electricity moves from the power grid to a data center, from a power supply to a server, or from a server board to an AI processor, power semiconductors are working behind the scenes to ensure the right amount of power reaches the right place at the right time.
Q: How is a power chip different from a CPU or GPU?
CPUs and GPUs are designed to process data by running algorithms, training models, and performing AI computations. They are the “brains” of a device.
Power semiconductors, by contrast, ensure that those processors receive electricity in the right form. They are the circulatory system of a device delivery the energy it needs to function.
Q: Where are power semiconductors found inside AI infrastructure?
Power semiconductors are found throughout the AI infrastructure stack, everywhere from the electrical grid to the processor itself. They are used in power generation and distribution systems, data center power supplies, battery backup systems, cooling equipment, server power units, networking equipment, and voltage regulation modules that deliver power directly to GPU/CPUs and AI accelerators.
They are embedded at every layer of the power delivery chain:
- Energy Grid: Solid-state transformers and grid-connected power conversion systems
- Energy Storage Systems (ESS): Battery storage and energy management systems that store electricity, provide backup power, and help balance fluctuating AI power demands
- Data Center Infrastructure: Power supplies, power distribution units (PDUs), battery backup systems, and cooling equipment
- Server-Level Power: Rack power shelves, voltage regulators, and AI server power architectures
- Chip-Level Power Delivery: Voltage regulation modules that deliver precise power directly to GPUs and AI accelerators
Q: What role do wide bandgap materials like silicon carbide (SiC) and gallium nitride (GaN) play?
One of the biggest areas of innovation in power chips is in the materials used to make them. Silicon carbide (SiC) and gallium nitride (GaN) are next-generation semiconductor materials that deliver higher efficiency, faster switching speeds, and greater power density than traditional silicon. These advantages help reduce energy losses, manage heat, and enable more compact power systems.
As AI data centers demand more electricity and higher-performance power delivery, SiC and GaN technologies can help operators deliver more computing power while using less energy. In many applications, they are enabling the next generation of efficient power conversion systems that support the growing scale and energy demands of AI infrastructure.
Q: Why are silicon carbide and gallium nitride better materials than traditional silicon in power semiconductors?
Compared with silicon, SiC and GaN deliver lower conduction and switching losses while supporting higher-voltage operation and greater power density. These characteristics are increasingly important as AI servers and data centers move to higher power levels and more complex power delivery architectures. The result is improved energy efficiency, reduced heat generation, and a more scalable path to supporting growing AI compute demands.
Q: Why are power semiconductors so important for the AI infrastructure buildout?
AI workloads demand unprecedented amounts of power, and without highly efficient power chips, a significant portion of that energy would be lost as heat during conversion and distribution. By minimizing these losses, stabilizing power delivery, and enabling higher-voltage, higher-density system designs, power semiconductors allow data centers to run more compute on the same power footprint.
In an environment where power availability, cost, and cooling are becoming the primary constraints on AI growth, power semiconductors are the enabling technology that makes scaling AI both technically feasible and economically viable.
Q: What is the shift to 800V DC and why does it matter for the AI infrastructure buildout?
The shift to 800V DC is a move toward higher-voltage power architectures that improve the efficiency of delivering power from the grid all the way to AI processors. Higher voltage means lower current for the same power level, reducing conduction losses and enabling more compact power delivery systems. As AI racks move beyond 100 kW and continue to scale, 800V DC helps improve power density, reduce cooling demands, and support the next generation of AI infrastructure. High-performance power semiconductors are the key enablers that make these high-voltage systems practical and efficient.
Q: What role do power semiconductors play in the shift to 800V DC?
Power semiconductors are the core enablers of the shift to 800V DC because they efficiently convert, control, and protect power throughout the data center. They perform the critical voltage conversion steps from grid power to intermediate bus voltages and ultimately down to the low voltages required by processors and memory. At 800V, power must be switched at high speed and managed with greater precision, requiring advanced devices that can handle higher voltages without sacrificing efficiency. The result is lower power loss, reduced heat generation, improved power density, and more efficient delivery of energy to AI compute workloads.
Q: What is onsemi’s role in the AI infrastructure buildout?
onsemi plays a critical role in the AI infrastructure buildout with its intelligent power solutions that ensure energy from the grid is transformed into stable, usable power for high-performance compute—while minimizing loss and heat at every stage.
Through its technology leadership in advanced materials like silicon carbide and gallium nitride, onsemi helps hyperscalers and infrastructure providers run more compute within the same power footprint, reduce cooling requirements, and improve overall system efficiency.
In a landscape where power availability and cost are becoming the primary constraints on AI growth, onsemi’s role is to turn electricity into usable compute as efficiently as possible—making large-scale AI both technically and economically viable.
Q: Are most AI data centers using advanced SiC and GaN power semiconductors yet?
Data centers traditionally didn’t need smaller, more efficient power systems. However, as AI racks move from tens of kilowatts toward hundreds of kilowatts and megawatt-class racks, the industry is shifting towards SiC and GaN chips because they switch faster, waste less energy as heat, and allow smaller, denser power systems. Nvidia’s push for an 800V architecture in data centers, expected in 2027, will be a major forcing function for AI industry driving even more demand for power chips than they have over the past two years. Bank of America expects this TAM to triple by 2030.
Q: What is the impact of more data centers adopting higher efficiency power semiconductors?
At a basic level, adopting recent generations of SiC power chips in data centers globally could reduce power energy 1%, potentially saving 10 TWh annually, which could power a million homes for a year or a high speed train network for five years.
However, innovations by onsemi and others are constantly increasing the potential energy efficiencies and savings. onsemi’s recent breakthrough vertical GaN chips can reduce loses by up to 50%, drastically scaling the potential energy savings and reducing the cooling and energy needs of an AI data center.
Sources:
https://www.power-sonic.com/400v-vs-800v-charging/
https://www.onsemi.com/company/news-media/press-announcements/en/onsemi-unveils-vertical-gan-semiconductors-a-breakthrough-for-ai-and-electrification
https://www.onsemi.com/company/news-media/press-announcements/en/how-onsemi-is-powering-the-next-generation-of-ai-factories
https://tech.yahoo.com/ai/articles/onsemi-aims-improve-ai-power-151904047.html
