Samsung Electronics regained second place in the high-bandwidth memory market in the third quarter of 2025, reaching a 22% revenue share against Micron’s 21%, according to Counterpoint Research estimates disclosed on December 19. Samsung overtook Micron for the first time in three quarters, while SK hynix remained the leader with 57%.
Samsung’s share rose seven percentage points from 15% in the second quarter. Micron’s share was unchanged, while SK hynix’s fell from 64%. HBM is used in AI accelerators, including products from NVIDIA and AMD.
Counterpoint’s figures put Samsung’s share at 40% in the fourth quarter of 2024, followed by 13% in the first quarter of 2025 and 15% in the second. The third-quarter increase therefore recovered part of the ground Samsung had lost earlier in the year. Micron held 18% in the first quarter and 21% in the second.
Counterpoint attributed Samsung’s first-half difficulties to export restrictions to China and its third-quarter share increase to stronger HBM3E performance.
HBM3E sales expand
In its results for the quarter ended September 30, Samsung said its memory business achieved record quarterly revenue as it expanded HBM3E sales. The company said HBM3E was in mass production and being sold to all related customers, while HBM4 samples were being shipped to key clients.
AMD said at its Advancing AI 2025 event on June 12 that its Instinct MI350X and MI355X accelerators would use 12-layer HBM3E from Samsung and Micron. Micron separately announced that day that its 36GB, 12-high HBM3E was being integrated into the MI350 series.
NVIDIA also described a supply collaboration with Samsung covering HBM3E and HBM4 in its announcement of a joint semiconductor AI factory. That announcement did not specify memory shipment volumes or a timetable for HBM4 deliveries.
HBM4 production plans extend into 2026
Samsung said its memory business would focus on HBM4 mass production in 2026 and expand its HBM sales base. The company also projected rising HBM4 demand and said it planned to expand capacity for its 1c DRAM process.
Micron announced on June 10 that it had shipped 36GB, 12-high HBM4 samples to key customers. It said it planned to ramp HBM4 production during calendar year 2026, aligned with customers’ next-generation AI platforms.
AMD’s announced roadmap includes HBM4 in its MI400 family. The company said its MI455X accelerator would use 12 HBM4 stacks and power the Helios rack-scale platform, which it targeted for availability in the third quarter of 2026.
