South Korean semiconductor giant Samsung is currently grappling with a significant loss of engineering talent as employees increasingly defect to rival firm SK Hynix. The trend is largely driven by disparities in performance-based compensation, as SK Hynix leverages record profits from the artificial intelligence boom to offer more lucrative bonus packages.

The competition centers on high-bandwidth memory (HBM) chips, which are essential for powering Nvidia’s AI accelerators. While Samsung has historically dominated the sector, SK Hynix gained a strategic advantage by doubling down on HBM technology in 2019. This shift has resulted in SK Hynix leading the global HBM market, with the company’s recent financial success allowing it to offer bonuses reaching $476,000 per employee. In contrast, Samsung’s bonus structure is tied to individual division performance, leaving employees in its foundry division—which has faced recent losses—with significantly smaller payouts.

Advertisement

Internal morale at Samsung appears to be waning. According to a June survey by the company’s labor union, nearly half of the semiconductor division’s staff expressed a desire to move to another firm within the next two years. Union chief Choi Seung-ho reported that over 200 members had already departed for SK Hynix in the four months leading up to April. Some employees noted that even management has encouraged staff to explore opportunities at the rival company.

The rivalry has escalated into legal territory, with Samsung recently securing a court injunction to prevent two former employees from joining SK Hynix for 18 months, citing the need to protect national core technologies. Meanwhile, both companies are aggressively expanding their workforces to meet projected demand, with plans to invest over $2 trillion by 2040. However, industry experts warn that the talent drain could hinder Samsung’s competitive edge, particularly in the development of next-generation HBM4 chips, which require deep integration between memory and foundry expertise.

Source: MIT Technology Review