Samsung Shifts Half of HBM Production to HBM4 as Memory War With SK Hynix Heats Up

Samsung Bets Big on HBM4

Samsung Electronics has dedicated roughly half of its high-bandwidth memory (HBM) production capacity to sixth-generation HBM4 chips, a decisive move to capitalise on surging AI memory demand and close the gap with rival SK Hynix. South Korean industry sources confirmed on 24 June that Samsung allocated about 75,000 of its 150,000 monthly HBM DRAM wafer inputs to HBM4, with the remainder assigned to 12-layer HBM3E. Production of the slower 8-layer HBM3E, now facing weaker demand, has been temporarily halted and its resources redirected to HBM4.

A $1 Billion Milestone in Four Months

The shift coincides with a remarkable financial sprint. Samsung's HBM4 revenue reportedly surpassed $1 billion just four months after mass shipments began in February 2026, a first for the industry. Revenue is expected to exceed $1.2 billion by the end of June, with insiders projecting full-year HBM4 sales could top $10 billion. Samsung says it anticipates total HBM sales will more than triple in 2026 compared with the prior year.

Racing to Reclaim Market Share

The reallocation reflects Samsung's urgency after losing ground to SK Hynix, which surpassed Samsung's market capitalisation in late June to become South Korea's most valuable listed company for the first time in 26 years. SK Hynix held an estimated 57 to 62% of the global HBM market in late 2025, versus Samsung's roughly 22%, though some analysts now place Samsung's HBM share in the 35 to 40% range as HBM4 scales. Notably, SK Hynix has chosen to slow its own HBM4 ramp to prioritise margins and the tightening commodity DRAM market, handing Samsung a window.

Technical Edge and the Next Generation

Samsung's HBM4 launched with a processing speed of 11.7 gigabits per second, roughly 46% above the industry standard, helped by a 4nm base die. The company has already extended its lead by shipping the industry's first HBM4E samples in late May, including to Nvidia, with performance gains exceeding 20% and bandwidth up to 3.6TB/s per stack.

Why It Matters

HBM is the plumbing behind the AI boom, feeding data-hungry chips inside the data centres that power tools like ChatGPT and Gemini. The intense rivalry is driving faster hardware but also a supply crunch: HBM4 commands a premium of about $500 per stack versus $300 for HBM3E, and the broader memory shortage is pushing up prices on everyday PCs, phones and consoles.