by LU Keyan

Surging global demand for AI computing hardware is driving shortages of microcontroller units, or MCUs, used in AI server power systems and high-speed optical modules, industry sources told Jiemian News, with overseas customers increasingly turning to Chinese suppliers.

Industry sources told Jiemian News that foreign makers of AI power supplies and optical communication equipment have begun large-scale procurement of Chinese MCU chips to cope with surging demand tied to AI computing.

As GPUs and NPUs consume larger bursts of electricity and require higher power density, MCUs have become a critical control component in AI power systems, managing dynamic loads and power distribution.

Monolithic Power Systems (MPS), a major supplier of power-management chips used by customers including Nvidia, AMD, Intel and Supermicro, reported first-quarter net profit of US$193.2 million, up 43.1% year on year, with gross margin at 55.3%. Its shares have risen more than 73% this year, giving it a market value of about US$77.3 billion.

The tightness is now spreading upstream to MCU suppliers. Industry sources said Chinese MCU maker CMSemicon raised prices for some MCU products by 15% to 50% in January, while Nationz Technologies later notified customers of price increases of 15% to 20% effective April.

Jiemian News learned that Nationz has recently begun volume shipments of AI server power MCUs to a top global power-management company. Its power-monitoring chips are now in stable mass production, with unit prices ranging from US$1.5 to US$2.

"The boom in AI server deployments is sharply increasing demand for power-control chips, while demand for 800G and 1.6T optical modules is also exploding," one industry executive told Jiemian News. "The rise in server and optical-communication power requirements is the core reason for the MCU shortage."

The executive estimated that Nvidia's GB-series server racks use more than 500 MCUs per cabinet for power boards alone. Each rack contains 18 computing units, each requiring more than 10 power modules, with three MCUs inside every module.

Based on deployments of 100,000 Nvidia racks alone, the AI power MCU market could reach about 450 million yuan (about US$66 million), the source estimated. Including customers such as Google and Amazon, the broader market could exceed 1 billion yuan.

Demand from optical modules is also rising rapidly as AI clusters scale up. Each optical module typically requires one or two MCUs to manage lasers and signal transmission.

Several Chinese MCU suppliers issued price-hike notices in April, with products used in optical communications rising 15% to 20%, and some specialized products increasing by more than 50%. Industry estimates suggest overall MCU prices in China's optical-communications sector have climbed roughly 40% this year.

Goldman Sachs has forecast global shipments of 800G optical modules will double in 2026 from 2025 levels. Separate industry forecasts project shipments of 1.6T modules could jump from around one million units to more than 20 million this year.

Among Chinese suppliers, GigaDevice Semiconductor was one of the earliest entrants into the optical-module MCU market. Its GD32E501 series, launched in 2020, has been widely adopted in data centers, cloud servers and telecom applications.

Nationz, meanwhile, is focusing on higher-end 1.6T optical-module controllers. Before the Lunar New Year, the company introduced MCU products supporting 400G, 800G and 1.6T modules. Industry sources said the chips are now being tested by multiple domestic optical-module and equipment makers.

Industry executives said supply constraints are worsening the shortage as foundries shift mature-node capacity toward higher-margin memory chips amid the AI investment boom.

At the same time, Chinese optical-module makers are increasingly seeking domestic alternatives to foreign suppliers such as Texas Instruments over supply-chain security concerns, creating another tailwind for local MCU producers.