Samsung Engineers Are Defecting to SK Hynix in Droves — the Talent War Over HBM Chips Heats Up

A talent war of unprecedented scale is unfolding between Korea's semiconductor giants over high-bandwidth memory (HBM) chips, the technology underpinning the AI boom. According to a detailed report by MIT Technology Review, engineers in Samsung Electronics' semiconductor division are being urged by colleagues and even their managers to "move to SK Hynix."

Behind the exodus is a stark pay gap: SK Hynix handed out bonuses of roughly $476,000 (about ¥68 million) per employee, while employees in Samsung's foundry division received about $135,000 (about ¥19 million). SK Hynix leads the global market in HBM chips for Nvidia's AI accelerators, and those profits are being passed on to its employees.

In a survey conducted in June by Samsung's labor union, 81.5% of foundry division employees — and roughly half of the semiconductor division as a whole — said they wanted to change jobs within the next two years. On some teams, nearly all 30 members reportedly applied to SK Hynix. The fight has also moved into the courts: in July, Samsung obtained a preliminary injunction barring two former employees from working at SK Hynix for 18 months.

The brain drain could also affect the development race for the HBM4 generation. Samsung's strength lies in being the only company with both memory and foundry operations in-house, but if engineers continue to leave the foundry division, coordination with the memory business could become difficult.

NYT's Lawsuit Against OpenAI Has Cost More Than $20 Million

According to a WIRED interview, The New York Times has already spent more than $20 million (about ¥2.8 billion) on the copyright infringement lawsuit it filed against OpenAI and Microsoft in 2023. Publisher A.G. Sulzberger remains committed to the fight, and the case is drawing attention as a symbolic courtroom battle over AI companies' unauthorized use of media content.

Selling Pressure Mounts on AI-Related Stocks

As The Guardian reports, selling is accelerating across the AI sector, hitting chip stocks such as SK Hynix and Samsung especially hard. Investors are beginning to question the sustainability of the AI boom, and money is being pulled out of the semiconductor sector. While demand for HBM itself remains strong, overall market sentiment is cooling.

Anthropic CEO Dario Amodei Criticizes Closed-Weight Models

A remark by Anthropic CEO Dario Amodei that made waves on Reddit's LocalLLaMA community is drawing attention: he suggested that models developed in secret with closed weights (non-public parameters) may produce worse results than open-weight models. The sight of the head of a major AI company acknowledging the advantages of an open approach sparked lively debate in the community.

AI-Generated Code Quality Improves as Main Branch Breakage Rates Are Halved

Mergify's "State of Merge Queues 2026" report finds that the frequency of AI-generated code breaking the main branch has dropped by roughly half. The report, which also trended on Hacker News, is drawing attention as data showing that the quality of AI coding tools has improved to a practically useful level.

Kumamoto Earthquake Prompts Warnings About Generative AI Misinformation

In domestic news, following an earthquake registering shindo 7 (the maximum on Japan's seismic intensity scale) in Kumamoto Prefecture, fact-checking initiatives have called on the public to beware of false and misleading information created with generative AI or recycled from past images. The event has once again highlighted how quickly AI-generated content can spread during disasters, underscoring the importance of media literacy.