[Edaily Reporter Kwon Oh Seok ] DaishinSecurities stated on the 7th that PROTEC Co., Ltd.(053610)is a stock worth watching, considering the improved performance of its subsidiaries and new equipment orders. However, it did not provide an investment recommendation or a target price.

Park Jang-wook, an analyst at DaishinSecurities, said, “Although the company has posted solid earnings every year, consolidated operating profit has declined somewhat due to losses at its major subsidiary, Protec Mems Technology. In 2025, Protec Mems Technology, a consolidated subsidiary, recorded a loss of 20 billion won.” He added, “Following Protec Mems Technology’s successful qualification for NAND-oriented probe cards for domestic memory semiconductor companies, the company achieved a turnaround with an operating profit of 200 million won in the first quarter. The company is expected to narrow its deficit to reach the annual break-even point (BEP)," he analyzed.
The company projected revenue of 282.6 billion won (up 22.6% year-over-year) and operating profit of 77.1 billion won (up 66.4%) for this year. He noted, “In addition to the reduction in Protec Mems Technology’s losses, we expect earnings growth driven by increased demand for high-performance dispensers in the global semiconductor market resulting from the expansion of HBM (High Bandwidth Memory) and CoWoS (Chip-on-Wafer-on-Substrate) packaging, which is driving increased demand for high-performance dispensers in the global semiconductor market," adding, "Starting in 2025, the company is also generating some revenue from global OSAT customers for its 'LCB' (Laser Compression Bonder) equipment, which combines bonding alignment with laser thermal compression bonding."
He pointed out that concerns over a cut in semiconductor CAPEX (capital expenditures) driven by Meta are excessive. Researcher Park explained, “It is a natural asset utilization cycle to prioritize the use of the latest high-performance chips for internal training and to lease out previous-generation GPUs—introduced two to three years ago—to external parties for inference purposes.” He added, “This is a trend also observed at other companies, such as xAI, where existing GPUs previously used for training are redeployed to meet inference demand.”
He further elaborated, “Rather than interpreting the release of previous-generation GPUs to external markets as a sign of slowing AI inference demand, it is more reasonable to view this as a process in which internal training and inference demands are shifting toward the latest chips, thereby expanding the use of older assets to the external inference market.”