Samsung and SK Hynix are reportedly preparing to start mass production of their sixth-generation high-bandwidth memory (HBM) chips, dubbed HBM4, in 2026. Designed specifically for AI workloads, HBM4 promises major performance improvements and better customisability. According to reports, Samsung will kick off production in February, while SK Hynix will follow in September. Most of these chips are set to power Nvidia’s Vera Rubin AI accelerator system.

ALSO READ: Xiaomi Buds 6 Launched With 35 Hours Of Battery Life And Harman-Tuned Audio

Priority Access for Nvidia

HBM4 is unlikely to reach the consumer market next year. Both Samsung and SK Hynix have deals with Nvidia to supply the chips for its AI accelerators. Samsung recently cleared Nvidia’s quality tests, making it eligible to supply the memory needed for the Vera Rubin system. This exclusivity gives the two semiconductor giants a head start over competitors like Micron, who aren’t expected to produce HBM4 in 2026.

Technology and Performance Gains

The HBM4 chips bring several technical upgrades. Samsung is using a 10nm logic process for the base die, while SK Hynix is collaborating with TSMC to adopt a 12nm process. These new chips are expected to deliver twice the bandwidth of previous generations, with power efficiency gains of nearly 40 percent. HBM4 also offers enhanced customisability for AI and non-AI workloads, enabling better integration with various chipsets.

Production Volumes and Market Impact

Samsung is reportedly set to produce 1,70,000 HBM4 units per month, slightly ahead of SK Hynix’s 1,60,000 units. Similarly, Samsung’s monthly DRAM output of 6,50,000 units surpasses SK Hynix’s 5,50,000 units. Despite the high production, the consumer market won’t benefit as the next year’s HBM4 capacities are fully allocated to AI companies. This means the ongoing RAM shortage is expected to continue through 2026.

ALSO READ: A Month-By-Month Look At Google’s Biggest Announcements In 2025: Google AI Stand Quick Recap 2025

Conclusion

With HBM4, Samsung and SK Hynix are strengthening their position in the AI hardware market. These chips bring significant performance, efficiency, and customisability gains, but consumers will not see them directly. Nvidia’s Vera Rubin accelerators will be the primary beneficiaries, while global RAM scarcity will likely persist.