Nvidia Rubin AI Accelerators to Retain 12-Layer HBM4 Design
Nvidia is set to maintain a 12-layer HBM4 configuration for its Rubin AI chips, while market analysts forecast a 121% surge in HBM prices by 2027 due to technological shifts.

Nvidia is set to maintain a 12-layer HBM4 configuration for its Rubin AI chips, while market analysts forecast a 121% surge in HBM prices by 2027 due to technological shifts.

Samsung is negotiating with AMD to expand their partnership beyond HBM memory supply to include contract chip manufacturing services.

AMD CEO Lisa Su met with Samsung co-CEO Jun Young-hyun in Seoul to discuss strategic cooperation across a broad range of technology sectors.

AMD is scaling up production of its Helios AI racks to meet demand from major hyperscalers, despite persistent supply chain constraints for HBM4 memory.

AMD CEO Lisa Su is visiting South Korea to finalize HBM4 supply agreements, discuss DDR5 memory for EPYC processors, and explore potential foundry partnerships with Samsung.

Samsung is targeting a price of $4 per Gb for its upcoming HBM4 memory, aiming to challenge SK Hynix's market leadership through superior performance metrics.

Samsung's upcoming HBM4 memory may cost three times as much as HBM3E, signaling a major shift in DRAM industry pricing and production priorities.

South Korean startup FuriosaAI has announced its third-generation AI inference accelerator, featuring 32 PFLOPS of performance and 48TB/s of memory bandwidth.

Samsung Foundry is dedicating half of its 4nm production capacity to HBM4 base dies to meet growing demand for custom logic integration.
