Samsung Electronics has posted its worst quarterly performance since 2023, with Q2 2025 operating profits plummeting 56% year-over-year to ₩4.6 trillion ($3.36 billion), far below analysts' estimates of ₩6.2 trillion. The disappointing results highlight Samsung's ongoing struggles in the competitive AI chip market.
The world's largest memory chipmaker cited multiple factors for the profit miss, including inventory value adjustments and the impact of U.S. restrictions on advanced AI chip exports to China. However, industry analysts point to a more fundamental issue: Samsung's delayed certification for its 12-layer HBM3E memory chips from Nvidia, the dominant AI chip designer.
High-bandwidth memory (HBM) has become critical infrastructure for AI computing, with the global market projected to reach $21 billion in 2025, growing at 70% annually. While Samsung once dominated the memory chip sector, it now trails SK Hynix, which controls approximately 60% of Nvidia's HBM supply chain. Samsung's certification process for its advanced HBM3E chips has reportedly been pushed to September 2025, creating an 18-24 month lag behind competitors.
Despite these setbacks, Samsung is pursuing alternative strategies, including supplying HBM3E chips to AMD for its MI350X AI accelerators since June 2024. The company has also promised that its HBM3E memory will pass certification in the second half of the year, with full shipments to major customers beginning thereafter.
The semiconductor industry as a whole faces potential volatility, with some analysts predicting a moderation in AI investments as hyperscale cloud providers temporarily pause expansion efforts. This aligns with broader concerns about a possible AI chip bubble, where massive sales in 2024-2025 could be followed by reduced demand if enterprise AI use cases fail to materialize at anticipated scale.
For Samsung, the path forward involves diversifying beyond Nvidia, accelerating development of next-generation HBM4 chips targeted for mass production in late 2025, and improving manufacturing yields to regain competitive advantage in the rapidly evolving AI hardware landscape.