---
title: "The Chip Inside the Charger: Why Gallium Nitride Is Moving Into AI Data Centers"
date: 2026-11-22
category: Chips & Compute
site: NeuroAI
canonical: https://neuroai.site/a/na-chips-gan-power-data-center
language: en
---

# The Chip Inside the Charger: Why Gallium Nitride Is Moving Into AI Data Centers

> A material that made phone chargers smaller is now being sold into server racks and robot joints, where saving a little power matters a lot.

A phone charger used to be a heavy little brick that got warm in your bag. The replacement is smaller and cooler, though most users never notice why. The reason is a material called gallium nitride (氮化镓, GaN), and the same property that shrank the charger is now pulling the material into a very different place: the power systems of AI data centers.

Innoscience (英诺赛科), a Hong Kong-listed Chinese company coded 2577, publishes results that show this migration happening in real numbers. Its 2025 annual report describes a business moving from consumer electronics into AI infrastructure, electric vehicles and the joint motors of humanoid robots (人形机器人).

## Key takeaways

- Innoscience reported 2025 revenue of ¥1.2133 billion (≈ US$170M / HK$1.32B), up 46.4 percent.

- Gross margin turned positive at 7.3 percent, from minus 19.5 percent a year earlier, and adjusted EBITDA turned positive.

- The company says it operates an end-to-end 8-inch GaN platform and has expanded applications from consumer electronics into AI data centers, new energy vehicles and robot joint drives.

- GaN's value is efficiency: faster switching and lower losses mean smaller power components and less wasted heat.

- For ordinary readers, this is the hidden layer behind both a cool phone charger and the electricity bill of a large AI service.

## Why a material matters more than a transistor count

Silicon has dominated power electronics for decades. Gallium nitride is a different semiconductor material that handles high voltages and switches current on and off very quickly. Faster switching lets engineers use smaller inductors and capacitors. Less resistance while conducting means less energy lost as heat.

In a phone charger, that translates to a smaller adapter that does not get hot. In a data center, the same physics scales up. Each server rack needs power converted from the building's supply into the voltages the electronics use. Multiply a small efficiency gain by thousands of racks running for years, and the saving becomes substantial — in electricity, in cooling load and in the space available for actual computing.

This is why a power-chip company can be part of an AI story without making a single AI accelerator. When processors become denser, the power delivery around them becomes a constraint. The bottleneck moves from computation to the ability to feed and cool it.

## What the numbers show

According to the company's press release and Hong Kong annual report, 2025 revenue was ¥1.2133 billion (≈ US$170M / HK$1.32B), an increase of 46.4 percent. Gross margin moved from minus 19.5 percent to plus 7.3 percent, a swing of 26.8 percentage points, and adjusted EBITDA turned positive. The net loss narrowed to ¥841 million (≈ US$118M / HK$917M) from about ¥1.046 billion (≈ US$146M / HK$1.14B) a year earlier.

The company attributes the improvement to scale, a better product mix and tighter cost control. That language is standard, but the direction is consistent with a fab that is finally running closer to capacity. An 8-inch GaN platform, which the company describes as end-to-end, is a way to make devices at a size that lowers unit cost compared with smaller wafers.

Financial-media summaries of the annual report give a more granular picture: GaN revenue from AI and data-center applications was about ¥63 million (≈ US$8.8M / HK$69M), up 50.2 percent, while automotive-grade GaN revenue was about ¥58 million (≈ US$8.1M / HK$63M), with shipments roughly doubling. These segments are still small relative to the whole company. They matter as a direction of travel, not as proof that data centers have already been transformed.

## From chargers to joints

An unusual application in the annual report is robot joint drives. A humanoid robot (人形机器人) has many motors packed into a small body, each needing compact, efficient power electronics. Heat and space are brutal constraints: a robot that overheats cannot work for long, and a bulky driver makes the machine awkward. GaN's high power density fits the requirement.

The same logic applies to electric vehicles, where onboard chargers and DC-DC converters benefit from smaller, more efficient components. In both cases, the material is not creating a new capability. It is making an existing machine practical enough to ship.

This is a useful way to think about semiconductor materials generally. The headline advances are usually about logic and memory — faster processors, denser storage. But many real-world products are limited by power conversion, and a modest improvement there can unlock a design that was previously too hot or too heavy.

## The honest scale check

It would be easy to overread these figures. A company with ¥1.2133 billion (≈ US$170M / HK$1.32B) in annual revenue and a remaining net loss is not yet a dominant supplier to the world's data centers. The AI and data-center slice, at roughly ¥63 million (≈ US$8.8M / HK$69M), is a small fraction of total sales. Positive adjusted EBITDA is a milestone, not sustained profitability.

The industry also has to decide which architectures will dominate. Data centers are evaluating higher-voltage direct-current (HVDC, 高压直流) distribution precisely because conventional layouts struggle with dense racks. Whether GaN wins a large share of that transition depends on cost, reliability and qualification work that takes years. The company's own materials speak of expanding application boundaries; they do not guarantee which design will prevail.

For readers, the right conclusion is modest but clear: a material once associated with phone chargers is being tested in much more demanding environments, and one supplier has published numbers showing early commercial traction.

## Honest limitations

This is an industry deep read based on Innoscience's official press release, its 2025 annual report filed with the Hong Kong exchange, and financial-media summaries of that report. Segment-level revenue figures come from those summaries and are not independently audited here. Claims about the 8-inch platform and application areas are company statements. No performance comparison with silicon or silicon carbide devices is made, no claim is advanced about any specific customer relationship, and the article does not discuss trade measures or government programs. Forward demand is not predicted.

## What readers can do now

- When comparing chargers or power banks, look at size and heat as well as wattage; GaN devices are often why a small adapter can deliver high power.

- If you work near data-center infrastructure, treat power conversion efficiency as part of the AI cost story, not a separate facilities issue.

- For investors, read the segment figures carefully: a fast-growing AI line can still be small compared with the mature business that pays the bills.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-chips-gan-power-data-center
Free to quote with attribution and a link to the original.
