SK Hynix will unveil the model of its “HBM4 12-layer,” a sixth-generation High-Bandwidth Memory (HBM) scheduled for mass production in the latter half of this year, at NVIDIA’s annual GTC conference in California.
SK Hynix announced on Wednesday that it will showcase its next-generation semiconductor products at GTC 2025 from Monday to Friday in San Jose, California. The booth will feature the theme “Memory, Powering AI and Tomorrow.”
The main attraction is the HBM4 12-layer. Currently, the most advanced HBM available in the global market is the fifth-generation HBM3E 12-layer. SK Hynix exclusively supplies HBM for NVIDIA’s graphics processing units (GPUs).
SK Hynix aims to commence mass production of HBM4 in the second half of this year. The company recently posted a video on social media showing a semiconductor chip’s engraved number changing from 3 to 4, accompanied by the message, “What comes after the number 3? The next chapter of AI memory unfolds.” This has led to speculation that the mass production timeline for HBM4 might be accelerated.
In addition to HBM4, SK Hynix will showcase its memory product lineup, including SOCAMM, which is gaining attention as a standard for AI server memory. The company will also present memory solutions for AI data centers, on-device, and the automotive sector. SOCAMM is a low-power DRAM-based memory module specifically designed for AI servers.
Key executives from SK Hynix, including CEO Noh Jung Kwak, CMO Joo Sun Kim, and Vice President Sang Rak Lee, will attend this year’s GTC to strengthen partnerships with global AI industry leaders.
Joo Sun Kim expressed pride in showcasing leading products for the AI era at GTC, stating, “Through our differentiated AI memory competitiveness, we aim to accelerate our future as a “full-stack AI memory provider.”