Is Optical Scale-Up Finally Approaching?
THE SO WHAT
If optical starts to move inside the box — not just between racks — AI system design shifts from PCIe/NVLink constraints to photonics-aware topologies. Infra teams speccing next-gen clusters should treat optical interconnects as a 3–5 year planning question, not a science project.
READ THE SOURCE
MORE FROM THE WIRE
Deep & Emerging TechSuspected Chinese spies are raiding university mailboxes via a Roundcube flaw
Nation-state actors are now systematically harvesting credentials from physics, engineering, and national security researchers via commodity webmail flaws — your academic and R&D partners are part of your attack surface whether you like it or not. If you depend on universities for joint work, treat their email and identity systems as untrusted and segment access accordingly.
Deep & Emerging TechMicron Raises Planned Spending to Meet Memory Chip Demand
Micron upping US investment by up to $3B to bolster memory supply is another reminder that AI demand is constrained as much by DRAM and HBM as by GPUs. If you’re planning large-scale training or inference, treat memory availability and pricing as a first-order capacity risk, not an afterthought.
Deep & Emerging TechMicron Boosts US Spending to $250 Billion Amid Memory Demand
A $250B US capex plan from Micron ties AI’s growth directly to long-horizon memory manufacturing — this is industrial policy territory, not just chip SKU planning. For infra buyers, it means the HBM/DRAM supply crunch will ease over years, not quarters, and long-term contracts may beat spot-market scrambling.
Deep & Emerging TechNordVPN expands scam text protection to all Android users worldwide
Scam filtering moving into consumer VPN apps shows how generative fraud is pushing security down into every comms surface. If your product touches messaging or identity, expect users to assume AI-grade scam detection is table stakes within 12–18 months.