Samsung and SK Hynix Establish 2026 High Bandwidth Memory Production Targets to Meet Increasing Artificial Intelligence Demand from Nvidia and Broadcom
Samsung together with SK Hynix has established their production targets for high bandwidth memory to be achieved in 2026. The semiconductor market worldwide experiences substantial growth according to actual shipment data which shows that Samsung Electronics and SK Hynix will reach 30 billion gigabits of combined capacity by the end of this year. The two companies will keep their planned shipment dates because they have resolved their recent clashes over the assessment of their technology capabilities. The memory production operations throughout South Korea experience a surge because major artificial intelligence companies like Nvidia and Broadcom maintain their increased demand for memory products in this period.
Samsung Electronics has set a production target which requires them to produce triple their output from the previous year. At GTC 2026 conference Vice President Hwang Sang jun revealed that the company is expanding its facilities to achieve a projected output of 11 billion gigabits. The development of HBM4 which uses 10nm class DRAM technology from sixth generation 10nm class DRAM and a specialized manufacturing method for base die logic functions as the foundation for this expansion. The technological combination enabled Samsung to become the first supplier to deliver the 11.7 gigabits per second performance level which Nvidia requires for its newest hardware.
Samsung is finalizing supply agreements with AMD and Broadcom while maintaining its existing partnership with Nvidia. Broadcom serves as an important customer because it enables the development of specialized AI processors which Google uses for its AI needs. The search giant integrates high bandwidth memory into its v7p Tensor Processing Unit which uses HBM3E technology for its latest version. Current estimates suggest Samsung will dedicate nearly all of its 130000 monthly units of 1c DRAM production to these high priority memory products through the end of the year.
SK Hynix expects its total shipments to reach approximately 20 billion gigabits which represents a sixty percent increase over its prior annual performance. Despite the company resolving power delivery issues in its 12 layer HBM4 samples the leadership team confirmed shipment plans would not change. SK Hynix depends on fifth generation 10nm class DRAM to manufacture its main elements while TSMC provides support for base die logic manufacturing. Nvidia depends on Samsung as its main supplier although Samsung uses a different manufacturing method from its foundry approach which Nvidia currently utilizes for 66 percent of its chip manufacturing needs.
Nvidia currently executes its procurement strategy through multiple levels to achieve successful large scale production of its Vera Rubin architecture. This involves utilizing both peak performance HBM4 and standard tier 10 gigabits per second products. SK Hynix President Kwak No jung recently informed shareholders that while minor adjustments to the product mix are occurring in response to client feedback the total volume of shipments will remain robust. The market observers believe both Samsung and SK Hynix can sell their entire inventories because the total market supply remains below worldwide demand while their 2026 plans require no major changes.
