Samsung shares rise 6.4% on HBM4 production report

Samsung Electronics’ shares rose up to 6.4% following a report that it will begin mass production of HBM4 memory chips, which are used in AI infrastructure.

The company plans to ship the chips to Nvidia as early as the third week of February, according to Yonhap News, citing industry sources.

The HBM4 chips are expected to support Nvidia’s upcoming Vera Rubin AI accelerators.

Samsung has been catching up to SK Hynix, which previously led as Nvidia’s main memory chip supplier.

The stock has gained over 30% this year amid rising memory chip prices driven by demand from hyperscalers like Amazon and Alphabet.

These companies are expected to increase their data center investments significantly in 2026, with some US stocks tied to AI and data center buildout also seeing gains.

The overall market environment has contributed to Samsung’s stock performance this year.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Samsung is betting on its own tech to catch up in the AI memory race

Samsung plans HBM4 shipments to Nvidia using its own 4-nanometer chipmaking (foundry) process for the base die (logic die), plus its 1c DRAM process for the DRAM cell die 1.
That setup keeps more of the supply chain inside Samsung. SK hynix uses TSMC’s 12-nanometer foundry process for the base die and its own 1b DRAM process for HBM4 1.
Reports put Samsung’s 1c DRAM yields for HBM4 near 50–70%, with some figures at 50–60%, which squeezes margins and slows large-scale output 2.
This push supports a rebound after Samsung fell behind in HBM3E. It now aims to narrow the gap with SK hynix in high-bandwidth memory (HBM) used in AI accelerators, specialized chips designed to train and run AI models 2.

The scramble for HBM4 memory is tightening the entire chip supply chain

Samsung’s move into HBM4 adds pressure to an already tight market, with commentary saying HBM is sold out for 2026 3.
HBM4 also consumes capacity, since one analysis says it needs about three times more silicon wafer space than standard DRAM for the same capacity 3.
That shift pushes memory makers to retool production lines, which can cut supply and raise prices for conventional memory used in PCs and standard servers.
Advanced chip packaging capacity, where chips and memory get assembled into high-performance modules, is also described as heavily constrained. The combined limits slow how fast new AI infrastructure can ship 3.