The power requirements of artificial intelligence (AI) have become one of the defining challenges of the technology’s rapid expansion. As the International Energy Agency (IEA) said in its 2025 report, Energy and AI, “Training and deploying AI models takes place in large and power-hungry data centers. A typical AI-focused data center consumes as much electricity as 100,000 households, but the largest ones under construction today will consume 20 times as much.”
As early as 2024, these data centers already used 415TWh, accounting for about 1.5 percent of global electricity consumption and growing by about 12 percent annually since 2017, according to IEA. By 2030, this is forecast to more than double and reach about 945TWh.
A white paper from Infineon Technologies in late 2025 said that the load current for a single processor would reach 10,000A in the next decade. This is a ten-fold increase over current levels.
In this scenario, traditional 12V systems will prove inefficient in handling the power conversion for specialized AI accelerators such as GPUs, TPUs, FPGAs and ASICs. This prompted the Open Compute Project (OCP) in late 2024 to underscore “48V architecture as the standard for meeting the electrical, mechanical and thermal requirements of evolving data centers,” according to Vito Savino in an April 2025 EETimes article.
48V advantages
Traditional 12V systems lose significant energy through conversion and distribution, particularly as power demands escalate. This high requirement translates to enormous currents, requiring thick copper busbars and generating substantial heat – resulting in considerable resistive losses as well as posing a fire risk.
Moving to 48V reduces current for the same power delivery, creating better efficiency and safety. In addition, 48V systems use much thinner and smaller cabling, which frees up space for more computing and cooling components while reducing copper costs. As such, they offer a scalability and futureproofing option.
Technologies for power solutions
GaN and SiC are considered critical technologies for power solutions designed for AI applications, replacing silicon-based systems. They enable higher switching frequencies, lowering energy loss, boosting efficiency while also allowing more compact and lightweight designs.
Navitas has released an 8.5kW GaN/SiC PSU with an efficiency of up to 98 percent. The company claims this to be the first of its kind for next-generation AI data centers. Infineon, ON Semiconductor, EPC, Texas Instruments, GaNext, Corenergy and Danxitech are also in this power device field, according to a November 2024 TrendForce News report.
It is also widely recognized among industry analysts and players that SiC MOSFET+800V HVDC will bring about a revolutionary upgrade in AI data centers. Supporting Nvidia’s bid for new 800V HVDC architecture data centers is ROHM and other power semiconductor companies.
AI market
Artificial intelligence has rapidly extended its reach into the medical and healthcare, merchandising, automotive, manufacturing, financial and agricultural fields. Its global market scale reached $638.23 billion in 2024, according to Precedence Research. If the estimate for 2025 correlates with $638.23 billion, then the forecast $3.68 trillion by 2034 means an astounding CAGR of 19.25 percent during the period. The Asia-Pacific region is expected to post a remarkable 19.8 percent CAGR, making it the fastest-growing market, while North America remains the largest.
In China, AI yielded $128.5 billion in 2024 and is poised to almost double to $247 billion in 2035, according to CCTV News, the English language news channel of China Central Television.
In the AI power solutions sphere, China is projected to beat the global CAGR of 45 percent from $2.846 billion in 2024 to $60.81 billion in 2031, according to QY Research. The country will achieve 58 percent during the period and will have a 25 percent share of the market worldwide by 2031.
China’s power bid
Sungrow Power, a major player in China’s renewable energy sector, has entered the AI power category with the setting up of an AI data center division in May 2025. Its first scheduled product release is for the high-end market within 2026.
Other key suppliers include Shenzhen Honor, Hangzhou Zhonhen and Shenzhen Megmeet. Shenzhen Megmeet, a provider of AI server power supplies to Nvidia, has been focusing on high power density power modules. It allocates over 11 percent of total sales for R&D every year and has acquired more than 1,990 patents, so far. The company’s new projects in Changsha in Hunan province and in Thailand reached full-scale production in the fourth quarter of 2025. From this expansion, Shenzhen Megmeet targets a total revenue of $5.71 billion by 2027.
AI-related suppliers
There are now more than 6,000 AI-related enterprises in China, triple the number of five years ago, according to CCTV News. Many are found in coastal electronics hubs Shenzhen, Dongguan, Zhongshan, Suzhou and Hangzhou.
They include manufacturers of AI server power supplies, such as Aohai Technology Co. Ltd, Boluo Xurong Electronics Co. Ltd and Xiamen Xunheng Electronics Tech Co. Ltd.
Aohai offers the AH-CRPS2000A1A, with one version having 100 to 127VAC, 12A input, 12.2V, 82A and 12V, 2.1A, maximum 1,000W output and the other 220 to 240VAC, 10A input and 12.2V, 164A and 12V, 2.1A, maximum 2,000W output. Both have an efficiency rating of 94 percent and meet CCC, CE, UL, cuL and CB requirements.
Boluo Xurong’s model AI 3000W has an input voltage of 200 to 240VAC and a rated power of 3,000W, with conversion efficiency certified to 80 Plus Platinum. It conforms to CE, CB and FCC standards.
Xiamen Xunheng’s power supply, the XH300-500WG-5813, has 1,300W rated power and 95 percent efficiency. It complies with UL, CE, CB, EMC, UKCA, LVD, FCC, ETL, BIS, CCC and RoHS.
