Rohan Paul (@rohanpaul_ai)T2·78 pts
- Micron plans to introduce 'vertical HBM' by 2028, stacking memory directly on GPUs to boost AI output per watt.
- AI agents writing chip logic and firmware reduced SSD controller design time by 50% and customer readiness by 60%.
- Data reveals a single HBM4 stack moves 2.8 TB/s, while AI context length is growing at 30x per year.