Samsung Electronics Allocates Half of HBM Capacity to HBM4

Samsung Electronics has been found to have assigned half of its high bandwidth memory (HBM) production capacity to its sixth generation HBM4. After becoming the first in the world to ship HBM4 this past February, the company is now expanding production in earnest and moving to recover market share.
According to the semiconductor industry on the 24th, Samsung Electronics is understood to have recently allocated roughly 75,000 wafers, half of its monthly HBM DRAM wafer input of 150,000, to HBM4. The remaining half has been assigned to 12 layer HBM3E, its fifth generation product. In the case of the relatively weaker demand 8 layer HBM3E product, production has reportedly been temporarily halted and redirected to HBM4. HBM3E goes into Blackwell, NVIDIA's current generation artificial intelligence (AI) accelerator, while HBM4 goes into Rubin, the next generation AI accelerator. Setting aside HBM3E production used to meet existing demand, the company has in effect concentrated all of its capabilities on HBM4 production.
Samsung Electronics plans to ramp up HBM4 volume and thereby restore the HBM market competitiveness in which it had fallen behind. In HBM3E, SK Hynix effectively secured NVIDIA supply ahead of others, while Samsung Electronics, delayed in quality certification, was unable to win much volume. From Samsung Electronics' standpoint, concentrating on HBM4, where it began supply ahead of SK Hynix, is more advantageous than chasing as a latecomer in HBM3E. After its first mass production shipment of HBM4 this past February, Samsung Electronics recorded 1 billion dollars (about 1.54 trillion won) in revenue within four months. Total revenue this year is projected to reach 10 billion dollars.
Another backdrop to Samsung Electronics putting its weight behind HBM4 is the expansion of the application specific integrated circuit (ASIC) market. As large cloud companies such as Google, Amazon, and Microsoft increase development of their own AI accelerators to reduce dependence on NVIDIA GPUs, HBM demand going into these chips is also growing rapidly. Because base die and packaging competitiveness tailored to each customer's design requirements becomes important starting with HBM4, the analysis is that this could be an opportunity for Samsung Electronics, which holds memory, foundry, and advanced packaging together.
By contrast, SK Hynix has relatively little need to rush HBM4 mass production the way Samsung Electronics has. It has already secured a strong customer base and a volume advantage in HBM3E, and it is also projected to supply NVIDIA with the largest volume in HBM4. The point is that SK Hynix has more room than Samsung Electronics to keep taking existing HBM3E demand steadily while gradually shifting its center of gravity to HBM4. A semiconductor industry official said, "With NVIDIA CEO Jensen Huang having already stated during his visit to Korea this month that SK Hynix is its largest supplier, there is no need to rush, and Hynix's HBM4 revenue will also grow in line with the timing of Rubin mass production."
The competitive structure between the two companies surrounding the HBM market is also expected to intensify further. According to market research firm TrendForce and investment bank Bernstein, Samsung Electronics' HBM market share is projected to rise from 27 percent last year to 37 percent this year. SK Hynix's share is expected to fall from 56 percent to 43 percent. There is also analysis suggesting that next year Samsung Electronics will overtake SK Hynix's share.