Malyun Isman
351 posts

Malyun Isman
@Bellisimo99
I am an enthusiastic and goal-driven individual who thrives in challenging environments. With a focus on continuous learning and improvement, I approach each op




Warning: The entire AI industry will likely be bottlenecked by two companies: 1. $AXTI ($700M) 2. $SMTOY ($31.7B) Which both control 60–70%+ of the world's InP substrates. Future $NVDA, $GOOGL TPU v7 pods, $META, $MSFT, $AMZN hyperscaler clusters require InP-based lasers and receivers. $AVGO, $LITE, $COHR use for EMLs for 800G/1.6T transceivers, DFB lasers, and other optical infra. Without InP substrates, the supply chain falters. After looking at TPU BOM to Maia BOM, it looks like future ASICs + GPUs + hyperscaler deployments are heavily reliant on photonics. And two vendors could freeze the global InP substrate market covering nearly all of: - Hyperscaler optics (TPU pods, etc) - Optical transceivers (5g, data) - LiDAR (robotaxis, drones, military) -Optical Modules (interconnect clusters) - Silicon photonics laser dies (Nvidia’s future co-packaged optics and Intel/Broadcom SiPh engines use InP CW laser arrays.) Since these companies make up majority of the market supply: -AXTI (est. ~30–35%) -Sumitomo (est.~30%) - JX Nippon (est. 10-15%) That’s it. (eg. 2021 industry note from Yole states that "Sumitomo Electric + AXT together had “more than 75%” of the InP substrate market") Hyperscalers/AI are moving toward photonics but the entire AI industry is fragile. If either $AXTI or $SMTOY stop supplying materials, the entire future AI buidlout gets crippled. It's even crazier that a $700m company could become the the center of it all. InP substrate will likely one of the biggest bottlenecks alongside HMB as the AI industry shifts to photonics.


























