[Edaily Reporter Choi Oh-hyun ] As global demand for artificial intelligence (AI) semiconductors continues to surge, expectations are growing that SamsungElectronics’ foundry (semiconductor contract manufacturing) business will return to profitability. With its 4-nanometer (nm, one-billionth of a meter) process effectively operating at full capacity, SamsungElectronics has also begun full-scale mass production of NVIDIA’s next-generation inference chips, which it is manufacturing.

According to industry sources on the 25th, NVIDIA announced that “Groq 3 LPX,” an inference acceleration system specialized for running Agent AI agents, has secured a customer and entered full-scale mass production. Nebius, an AI cloud company, has decided to adopt Groq 3 LPX for its inference platform, “Nebius Token Factory.” The Groq 3 LPX is a system comprising 256 language processing units (LPUs) in a single rack. It is integrated into NVIDIA’s next-generation AI platform, “Vera Rubin,” and serves to boost AI inference performance and speed.

The Grok-3 LPU, the core component of the LPX, is manufactured by SamsungElectronics’ Foundry Business Division using a 4-nanometer process. As SamsungElectronics moves to expand mass production of 4-nanometer products—with a focus on AI and high-performance computing (HPC)—in the second half of this year, a product from a major client has now entered full-scale mass production.

With a surge in orders for AI semiconductors, the 4-nanometer production lines are already operating at virtually full capacity. Furthermore, unlike in the past when prices were lowered to secure customers amid oversupply, the company is regaining its pricing power based on rising demand. It is reported that prices for new orders on certain foundry processes—such as 4-nanometer, 5-nanometer, and 8-nanometer—have recently increased by up to 10–15%. The 5-nanometer and 8-nanometer processes are also used in AI, HPC, and data centers.

Changes are also evident in the company’s financial performance. SamsungElectronics announced that its foundry revenue improved significantly in the second quarter of this year, driven by demand for High Bandwidth Memory (HBM) base dies and increased orders from U.S. clients. In the second half of the year, the company plans to expand mass production of 4-nanometer-based LPUs and sales of HBM base dies, while also accelerating efforts to secure clients for its cutting-edge 2-nanometer process. The Foundry Business Division plans to increase orders from major customers, focusing on AI, HPC, and new mobile products.

Collaboration with NVIDIA has taken shape for the 4-nanometer and 8-nanometer processes. For the 2-nanometer process, the company secured Tesla and Apple as customers last year and plans to begin mass production of the Exynos 2700 mobile application processor—based on the second-generation 2-nanometer process—in the second half of the year. SamsungElectronics is targeting double-digit growth in foundry revenue for the second half of the year, as demand increases across all process nodes, particularly from U.S. and Chinese customers.

Samsung Foundry, which has been unable to escape the red since 2022, has long been considered a “problem child.” However, the outlook is changing as utilization rates for the 4-nanometer process have risen—driven primarily by AI semiconductors—while pricing power for the 5- and 8-nanometer processes has improved and the company has secured customers for the 2-nanometer process. Some observers predict that the company could not only reduce its losses but also return to profitability as early as this year.

An industry official stated, “It is true that positive signs continue to emerge,” adding, “Since there is a time lag between securing orders and mass production in the foundry business, and the burden of fixed costs resulting from large-scale capital investments is significant, we need to take a longer-term view.”