China Looks to Silicon Carbide to Reduce AI’s Electricity Bill

The race in artificial intelligence is no longer measured solely in GPUs, memory HBM, or ever-larger data centers. It is also measured in electricity. Every new training or inference cluster raises one less glamorous but crucial question: how to deliver more power with fewer losses within infrastructure that’s starting to strain electrical grids, substations, and cooling systems.

Silicon carbide, better known as SiC, is gaining ground in this area. This semiconductor material, used for years in electric vehicles, renewable energy, inverters, and industrial power electronics, is now appearing in a different conversation: data centers for artificial intelligence. It doesn’t replace NVIDIA, AMD, or the leading accelerators. Instead, it works at a less visible layer—power conversion and management—but every efficiency point counts there.

Chinese semiconductor company Basic Semiconductor, based in Shenzhen, has recently come into focus after advancing its IPO process in Hong Kong. Founded in 2016 by alumni of Tsinghua University and Cambridge University, the company positions itself as one of the few Chinese players with an integrated silicon carbide model, covering chip design, wafer fabrication, module packaging, and control solutions.

The electricity problem in AI opens a new market

The International Energy Agency estimates that the global electricity consumption of data centers could more than double, reaching around 945 TWh by 2030. AI is not the only driver but the main accelerator of this increase, alongside other digital services. In China, the expansion of AI data centers has become an energy planning challenge too, with official targets to increase renewable energy’s share in this sector and reduce grid pressure.

Power electronics fit right into this scenario. A data center doesn’t just draw electricity and pass it to GPUs. The energy undergoes several conversion stages: the grid, transformers, uninterruptible power supplies (UPS), rectifiers, internal distribution, server power supplies, and finally chips operating at very low voltages. Each conversion results in some heat loss.

SiC promises to reduce some of these losses because it can handle higher voltages, temperatures, and switching frequencies than traditional silicon in certain applications. This enables more efficient power supplies, more compact designs, and less heat to dissipate. It doesn’t solve the entire energy challenge of AI on its own but can help each megawatt produce more output.

Basic Semiconductor aims to enter the data center market

Basic Semiconductor wasn’t founded as an AI company. Its business has been more focused on power modules, discrete SiC devices, gate drivers, electric vehicles, renewables, and industrial applications. However, its documentation for Hong Kong already highlights data centers and servers as areas of growth.

The company states that its SiC products are used in power supplies for data centers and servers to improve efficiency, reduce losses, operate at higher frequencies, and withstand more demanding thermal conditions. It also indicates that mass production of its SiC power devices designed for AI data centers began in Q1 2026.

The company isn’t starting from zero. Its forecast includes revenues of 220.6 million yuan in 2023, 299.0 million in 2024, and 311.2 million in 2025. It also claims to have shipped over 140,000 units to new energy vehicles by the end of 2025. This demonstrates a solid industrial footing, though still small compared to major international power electronics suppliers.

Why SiC interests data centers

AI data centers are increasing rack density. Each system generation accelerates power consumption: more GPUs, more memory, faster networks, and better cooling. In high-power architectures, electrical conversion is no longer just an engineering detail but part of operational costs.

Silicon carbide competes because it allows operation at higher voltages with lower losses during power conversion. Practically, this can help design more efficient, compact power supplies. In high-voltage DC distribution scenarios—gaining interest for AI racks—this advantage can be even more evident.

The market opportunity remains small in absolute terms. Basic’s own report, citing Frost & Sullivan, estimates the global market for SiC power devices for data centers will grow from $6 million in 2024 to $86 million by 2030. It’s rapid growth but from a very small base. The main business of SiC still lies in electric vehicles, renewables, and industrial sectors.

China’s quest for chip energy sovereignty

The story has a geopolitical layer. China isn’t just seeking domestic AI chips to reduce dependence on NVIDIA or exports controlled by the U.S. government. It also needs to build a local supply chain: memory, packaging, interconnection, cooling, power supplies, and power electronics.

In this context, SiC fits into the strategy of technological substitution. It’s less glamorous than a GPU but can become an essential component to manufacture more efficient, less import-dependent data centers. Basic Semiconductor presents itself as an integrated vertically oriented supplier—a trait that holds industrial and political value in China.

Investors’ interest in Chinese AI and semiconductor companies also supports this. Hong Kong and mainland markets have seen renewed activity related to chips, components, and advanced manufacturing. For companies like Basic, entering the market could finance capacity, R&D, and expansion—especially at a time when AI narratives attract capital.

Not all SiC will succeed

It’s important to avoid overly enthusiastic assumptions. Just because SiC makes technical sense doesn’t mean all SiC companies will capture this growth. The industry is demanding, validation cycles are long, and automotive, renewable, or data center clients don’t change critical components without extensive testing, certifications, and reliability assurances.

Moreover, China’s SiC market is competitive. Lower costs of substrates and wafers may boost adoption but also pressure prices and margins. Basic’s documentation acknowledges raw material volatility in substrates and epitaxial wafers for SiC. In power electronics technology, good manufacturing, consistent quality, and scaling without failures are often more challenging than market presentations suggest.

Also, the role of data centers should be nuanced. Although AI increases electricity demand, power electronics are only part of the solution. Better grids, supply agreements, liquid cooling, high-voltage distribution, load management software, and territorial planning will also be necessary. SiC can improve efficiency but doesn’t alone prevent the need for enormous power capacities in data centers.

Efficiency is becoming critical infrastructure

What makes Basic Semiconductor’s move interesting is how it shows that the AI frenzy is permeating down the entire industrial chain. First GPUs and memory; then optical networks, energy systems, cooling, transformers, batteries, and materials. Now, focus shifts to power electronic components previously considered distant from core tech headlines.

AI has made electrical efficiency a strategic issue. If data center demand continues growing as expected, reducing losses at every step will be as important as adding more chips. In high-density racks, less energy lost not only saves money on the bill but also reduces heat, simplifies cooling, and allows more compute per unit of space.

Basic Semiconductor aims to position itself at this intersection of Chinese technological sovereignty, capital market opportunities, and AI’s energy needs. An IPO won’t automatically turn SiC into the definitive solution for data center energy problems, but it confirms a trend: the next phase of AI development will also be driven by components consumers never see.

For years, discussions of semiconductors focused on nodes, transistors, and processors. The AI era is broadening that conversation—now also how chips are powered, how much energy is lost en route, and who makes the devices that keep data centers from turning into giant resistors hooked directly to the grid.

Frequently Asked Questions

What is silicon carbide or SiC?
A semiconductor material used in power electronics. It withstands high voltage, high temperatures, and fast switching, which can enhance efficiency in demanding applications.

Why is SiC relevant to AI data centers?
Because data centers are consuming more electricity and need more efficient, compact, and reliable power supplies for high-density racks.

Who is Basic Semiconductor?
A Chinese company founded in 2016 and based in Shenzhen, specializing in silicon carbide power devices. It operates with an integrated model covering design, manufacturing, packaging, and drivers.

Can SiC solve AI’s energy problem?
Not alone. It can reduce losses and improve efficiency in power conversion, but AI’s energy demands also require investments in grids, power generation, cooling, and planning.