SNL

借一张图解释华为Tao定律。 Photo credit to 睿而不酷 & Gemini.

JUST IN: Accenture to roll out Microsoft Copilot to 743,000 employees.

The AI data center has a problem nobody is talking about enough. The GPU is no longer the bottleneck. The wire is. Copper cannot move data fast enough to keep up with what Nvidia is building. It consumes too much power. It generates too much heat. As AI clusters get denser and models get larger, the physical limits of copper become a hard ceiling on what AI can do. The solution is light. Optical interconnects transmit data faster, cooler, and at a fraction of the power consumption of copper. Co-packaged optics, where the laser is packaged directly onto the GPU chip itself, reduces power consumption in AI clusters by up to 40%. That market is growing from under $400 million today to nearly $3 billion by 2032. Two companies own this transition. $LITE controls roughly 50-60% of the specialized laser chip market that powers these systems. Q2 revenue just came in at $665 million, up 65% year over year. Q3 guidance is $780-830 million, implying over 85% growth. Their backlog of optical circuit switches is sold out through end of 2027. They are targeting a $2 billion quarterly revenue run rate within two years. The CEO said they are at the starting line. Rosenblatt price target $900. $COHR just reported data center revenue up 36% year over year. A 20-year relationship with Nvidia just got formalized into a full strategic partnership covering next-generation silicon photonics, ultra-high power lasers, and priority capacity rights. Zero sell ratings on the Street. Rosenblatt price target $375. Both just got added to the S&P 500. Every index fund on earth is now a permanent forced buyer. Mizuho named $LITE a top AI pick for 2026 alongside Nvidia and Broadcom. The GPU era gave us Nvidia. The photonics era is giving us $LITE and $COHR.

wait NVL1152??? what? 1152?

AI is going to drain a lot of moats.

$CRDO $AVGO ON WHY HIGHLIGHTING DIRECT-ATTACH COPPER & WHY CPO CAN WAIT: “if you’re building AI data centers and architecting larger and larger domains or clusters, you want to connect experts to experts directly where you can.” “The best way to do that is direct-attach copper. That’s the lowest latency, lowest power, and lowest cost… especially in scale up as long as possible.” “In scale out… we can use optical. That’s fine. But… in a rack, in the cluster domain, you really want to use direct-attach copper as long as you can.” “We can push the envelope from 100G to 200G to even 400G… we have SerDes now running 400G that can drive distance… to run copper.” “All I’m trying to say is you don’t need to go run into some bright, shiny objects called CPO… even as we are the lead in CPO. CPO will come in its time. Not this year. Maybe not next year, but in its time.”