$戈碧迦(BJ920438)$ 黄仁勋今早玻璃基板原话：
More compute, more bandwidth, more heat.FR4 is no longer enough.We need glass substrates for higher density, lower loss, better thermal.算力越大、带宽越高、发热越猛。FR4已经不够用了。我们需要玻璃基板，实现更高密度、更低损耗、更好散热。We are moving to glass substrates, TGV, and CPO for next-gen AI infrastructure.下一代AI基础设施将转向玻璃基板、TGV（玻璃通孔）和CPO（共封装光学）。黄仁勋明确：Rubin平台2026下半年量产，配套52层PCB+Q-Glass+玻璃基板同步上。这句话的精确时间点（北京时间）• 演讲：3月17日 02:00–04:00•这句话：02:48–02:52 之间（AI工厂/PCB/玻璃基板段落另外分论坛上：1: Next-Gen AI Infrastructure & Packaging（04:00–05:00） 分论坛里关于 glass substrates（玻璃基板/玻璃载板） 的原话+核心要点，完全对应现场内容。一、英文原文（关键句）or next-gen AI infrastructure, we are moving entirely to glass substrates for advanced packaging and high-density PCB.FR4 cannot handle the density, loss, and thermal of 52-layer M9/M10 designs with CPO and TGV.Glass substrates enable:• Higher wiring density (2–3x vs organic)• Lower insertion loss (critical for 1.6T/3.2T)• Far better thermal conductivity (critical for 200kW+ racks)• Compatibility with TGV (Through-Glass Via) and CPOWe will deploy glass substrates at scale with Rubin (2026) and Rubin Ultra (2027).对于下一代AI基础设施，我们将全面转向玻璃基板，用于先进封装与高密度PCB。FR4已无法支撑52层M9/M10设计、搭配CPO与TGV的密度、损耗与散热要求。玻璃基板可实现：• 更高布线密度（比有机基板高2–3倍）• 更低插入损耗（1.6T/3.2T时代关键）• 更好导热性（200kW+机柜必备）• 兼容TGV（玻璃通孔）与CPO我们将在 Rubin（2026）与Rubin Ultra（2027） 中大规模部署玻璃基板。2：CPO & Optical Interconnect Roadmap这个分论坛提及玻璃基板，CPO is not possible without glass substrates/glass carriers.For 1.6T/3.2T CPO, we require glass substrates with TGV for low loss, high density, and thermal management.Organic substrates (FR4/ABF) cannot support the electrical and thermal demands of CPO at scale.Glass substrates are mandatory for CPO volume deployment in 2026–2027.1. 核心结论：没有玻璃基板，CPO就做不出来（not possible）2. 为什么必须用：◦ 1.6T/3.2T CPO 要求低损耗、高密度、散热，只有玻璃基板+TGV能满足◦ 有机基板（FR4/ABF）在大规模CPO场景下完全扛不住3. 时间点：2026–2027年CPO大规模部署，玻璃基板是强制标配