DegenQuant

“Don't be afraid to give up the good to go for the great.” — John D. Rockefeller

英伟达给了 2500 万只 NPO 需求指引？ 旭创1000万只指引，约160亿美元=1140亿人民币。旭创2025年全年营收382.4亿，光NPO这一块相当于现有体量的3倍。打五折也有570亿，相当于再造一个半旭创。 这就是牢美开了新易盛和中际旭创二倍做多的理由了？ 以及机构上调了旭创目标价到 1650 的核心原因也是因为 NPO 要放大量。

new: Meta is doing a 180, trying to be vanguard of token-minimizing. 2 months ago Meta epitomized tokenmaxxing, on track to spend billions a year on claude etc.

LONG $AVGO $MRVL Key Points >CPO co-packaged optics technology is actively advancing >NVIDIA Spectrum-6 Ethernet version experiences optical loss over 3.5dB higher than previous generation >Active alignment requires sub-micron precision with over one hundred operations per module >Quantum-X modular design has lower yield pressure while Spectrum-X direct co-packaging causes yield to deteriorate exponentially >NPO near-packaged optics serves as short-term transitional solution Conclusions >CPO is inevitable in the long term but the timeline carries significant uncertainty >Regardless of whether CPO or NPO ultimately adopted the optical components supply chain will continue to benefit Discussion This article thoroughly dissects the controversy triggered by the SemiAnalysis report The focus is not on denying the future of CPO but on highlighting the precision alignment difficulties during mass production Active alignment involves six degrees of freedom making it extremely easy for tiny deviations to cause high optical loss Spectrum-X has high port density and cannot easily replace faulty optical engines like Quantum-X This leads to a situation where with 95% yield per engine the probability of all 32 engines being good drops sharply to about 19% To achieve acceptable overall yield each attachment step needs to approach 99.5% yield threshold This presents a major challenge under current manufacturing processes NPO only needs to handle alignment for a single optical engine greatly reducing complexity It is suitable as a bridging technology before CPO matures Overall the article maintains a balanced perspective It acknowledges the reasonableness of the technical obstacles while pointing out that AI networking demand remains strong Investors and industry practitioners should closely monitor subsequent verification and solution progress