When the Bottleneck Migrates: Interconnect, HBM, and the Repricing of the Entire Storage Empire
- SK hynix and SanDisk introduced the first standard specification for High-Bandwidth Flash (HBF) at FMS 2026.
- NVIDIA is considering downgrading the HBM configuration for Rubin Ultra from 12hi HBM4E to alternatives including HBM4E 8hi.
- Micron, SK hynix, and SanDisk saw pre-market stock increases of 3.68%, 3%, and 4.42%, respectively.
- The report indicates that the interconnect is becoming the primary bottleneck in AI infrastructure, with HBM following closely behind.
- The migration of the bottleneck from inside compute units to the interconnect layer is reshaping the entire storage system.
On August 4, 2026, two significant developments occurred in the storage industry. In the morning, SK hynix and SanDisk unveiled the first standard specification for High-Bandwidth Flash (HBF) at FMS 2026, establishing a new storage tier between HBM and SSDs. In the afternoon, TrendForce reported that NVIDIA is evaluating a downgrade of the HBM configuration for Rubin Ultra, considering alternatives to the originally planned 12hi HBM4E.
The stock prices of Micron, SK hynix, and SanDisk rose in pre-market trading, reflecting positive market sentiment following these announcements. The report emphasizes that the supply bottleneck in HBM is being recognized, and new alternatives are being proposed, indicating a shift in the value distribution within the storage sector.
Crucially, the article suggests that the interconnects are becoming the primary bottleneck in AI infrastructure, with HBM close behind. This shift indicates a structural change in how bottlenecks are perceived, moving from within compute units to the interconnect layer, which is expected to reshape the entire storage ecosystem.
瓶颈迁移：互连、HBM与整个存储帝国的重新定价
2026年8月4日，存储行业发生了两项重要发展。上午，SK海力士和闪迪在FMS 2026上推出了高带宽闪存（HBF）的首个标准规范，建立了一个介于HBM和SSD之间的新存储层。下午，TrendForce报道NVIDIA正在评估将Rubin Ultra的HBM配置从原定的12hi HBM4E降级为其他替代方案。
美光、SK海力士和闪迪的股票在盘前交易中上涨，反映出市场对这些公告的积极情绪。报告强调，HBM的供应瓶颈正在被认可，并且新的替代方案正在被提出，表明存储行业内价值分配的转变。
重要的是，文章指出互连正在成为AI基础设施的主要瓶颈，HBM紧随其后。这一转变表明瓶颈的认知结构正在发生变化，从计算单元内部迁移到互连层，预计将重塑整个存储生态系统。