The production timeline comes as cloud service providers and AI laboratories have shifted away from 4-high and 6-high configurations for low-end inference, establishing 8-high as the industry minimum. Other 2027 products are moving toward even higher capacities, such as AMD's MI450, which is expected to use 12 HBM stacks to provide 288GB of HBM4, and OpenAI's planned adoption of 12-high HBM4.
