The stage for artificial intelligence (AI) is expanding from the cloud to real-world devices such as robots, factories, home appliances, and cars. In cloud AI, shortages of compute and memory, along with power supply issues, have emerged as bottlenecks, and physical AI faces the challenge of implementing AI functions with low power. ST is focusing its capabilities on easing the power bottleneck in AI data centers, and in physical AI, it is combining sensors, microcontrollers (MCUs), and power semiconductors to enhance devices' perception, inference, and control functions.
Hiroshi Noguchi, senior vice president of sales and marketing for Asia-Pacific, who visited Korea on the 15th to attend STMicroelectronics' (ST) annual tech conference (K-TEC 2026), said this in an interview with ChosunBiz.
He sees the AI Semiconductor market evolving around solving power and energy issues, Edge AI (technology that runs AI functions directly on devices where data is generated), and smart mobility. Senior Vice President Hiroshi Noguchi said, "With the spread of Generative AI, power demand at data centers is surging, making efficient and scalable power delivery not a choice but a must," and added, "There is also a strengthening trend to process information closer to the device rather than in AI data centers, and as cars become electrified and digitalized, the amount and role of semiconductors in each vehicle are growing."
ST is an integrated device manufacturer (IDM) that supplies products to a wide range of electronic applications, including automotive, industrial equipment, data centers, and home appliances. It has a broad portfolio of power semiconductors, MCUs, sensors, optics, and radio frequency (RF) technologies, and supplies products and solutions to more than 200,000 customers worldwide. Last year's revenue was $11.8 billion (about 17.7 trillion won), and operating profit was $175 million (about 260 billion won). It is regarded as a company that is aggressive in technology development, reinvesting 14.5% of total revenue in research and development (R&D).
◇ AI data centers: Solving the power bottleneck is a growth pillar
Senior Vice President Hiroshi Noguchi believes that changes in the AI Semiconductor market ultimately come down to how efficiently electricity is used to deliver performance. He said, "While electric vehicles, robots, and data centers are different markets, the integration of AI has created a common challenge of processing data efficiently with less power."
With the spread of Generative AI, data center infrastructure investment is focused on graphics processing units (GPUs) and high-bandwidth memory (HBM). The more computation GPUs and HBM handle, the more data center operators are challenged to build structures that deliver stable power to these semiconductors. He said, "AI, which demands massive power, is difficult to support with the existing data center power architecture alone," adding, "Ultimately, the process of delivering electricity to data centers must be overhauled."
High-voltage alternating current (AC) entering a data center is stepped down multiple times and converted to direct current (DC) as it passes through the building, racks, and inside servers. In the final stage, it must be converted down to the low voltages around 1V used by GPUs. The more stages of power conversion, the greater the losses and heat, and the more space is required for power supplies and cabling.
ST believes that switching data center power delivery to an 800V DC-based approach can resolve much of this issue. In the existing architecture, electricity is converted multiple times before being sent into the server, but the 800V DC architecture sends higher-voltage DC power closer to the server rack and then steps it down to the voltage used by GPUs. ST's approach is to reduce losses and heat by cutting power conversion stages and to ease the burden of power supplies and cabling within the rack.
ST has built a power conversion portfolio that spans from 800V DC to 50V, 12V, 6V, and down to the GPU power stage. This portfolio was developed to match Nvidia's 800V DC design. ST has also entered into a multi-year strategic collaboration worth several billion dollars with Amazon Web Services (AWS) in the field of high-performance computing infrastructure for cloud and AI data centers.
Senior Vice President Hiroshi Noguchi cited capabilities in silicon carbide (SiC), gallium nitride (GaN), and digital power as the foundation for ST to build this portfolio. He said, "SiC-based solutions reduce power consumption, and GaN-based solutions help shrink the area of the power stage." SiC is used to efficiently step down high voltages such as 800V DC, and GaN enables power supplies to operate smaller and faster so they can be placed closer to GPUs. Digital power precisely controls the power conversion process to reduce waste as electricity passes through multiple stages.
He said, "AI data center power efficiency solutions will become a new growth pillar," adding, "We will deliver results by providing solutions needed across the entire 800V DC power architecture, not just by supplying power semiconductor components." ST has said its data center-related revenue will far exceed $500 million (about 750 billion won) this year and could top $1 billion in 2027.
◇ "Responding to the physical AI market with a broad product lineup"
Senior Vice President Hiroshi Noguchi said that while ST is targeting the AI data center market with power semiconductors, it is also accelerating expansion into physical AI-related businesses. He said, "The next wave of AI is not all in the cloud," and added, "ST is already mass-producing silicon for Edge AI."
ST's physical AI strategy focuses on supplying, in one go, the semiconductors needed for devices to see, decide, and move in the field. He explained, "When people hear physical AI, they think first of Humanoid Robots, but the real market is much broader, including factory automation, collaborative and service robots, and even arm, hand, and vision modules."
To expand its physical AI business, ST recently launched a new MCU (STM32N6). It also unveiled micro-electro-mechanical systems (MEMS) sensors with AI functions and time-of-flight (ToF) and vision sensors. The STM32N6 is equipped with ST's in-house Neural Processing Unit (NPU), delivering up to 600 GOPS (60 billion operations per second). MEMS sensors detect device motion, while ToF and vision sensors perceive distance and space. Noguchi said, "By combining these products, we help customers' robots and industrial equipment perceive their surroundings, make decisions, and control movements."
He received a bachelor's degree in electrical engineering from Northwestern University and a master's in electrical engineering from Stanford University. After serving as ST's Japan country manager, since 2022 he has led sales and marketing across the APeC region excluding China. The following is a Q&A.
- In February this year, you acquired NXP's MEMS sensor business.
"With the acquisition of NXP's MEMS sensor business, ST's automotive portfolio has expanded considerably. Korea has a strong automotive industry and finished car makers, so we believe we can deliver more value to Korean customers.
This acquisition is not just about adding NXP's MEMS products to ST. By combining them with ST's existing portfolio of MCUs, sensors, actuators, and automotive semiconductors, we can approach customers from a broader system perspective. It is difficult to mention specific projects currently underway, but our Korea sales and marketing organization is in close discussions with customers based on the new products."
- Tariffs and the war in the Middle East are raising the importance of semiconductor supply chain issues.
"ST owns and operates its own fabs that manufacture silicon. Because we have manufacturing bases in multiple regions, we can respond to customer demand worldwide. Supply chain and logistics challenges are not unique to the semiconductor industry but a critical issue that every industry around the world must address. ST also operates logistics teams and has various measures in place to ensure customers receive products without disruption."