NVIDIA Puts Optical Engine Inside Switch Package and Starts Mass Production, but Where the Lasers Go Is Still Unclear

The NVIDIA AI Infrastructure account announced that Spectrum-X Ethernet Photonics has entered full mass production, calling it the world's first co-packaged optics Ethernet switch and the world's most advanced Ethernet switch. The two benefits officially given are straightforward: better signal integrity and higher energy efficiency. "First" is just a label; what is more worth noting is that the package location has changed.

Traditional Ethernet switches plug optical modules into the faceplate. Electrical signals start from the switch chip, travel over the PCB to the faceplate, and then enter the optical module to be converted into light. As speeds increase, this copper link becomes harder to handle, and both loss and power consumption rise. CPO's idea is to move the optical engine next to the switch chip, or even into the same package, so electrical signals travel less. Putting silicon photonics directly into the package is this approach.

Chen Haochong replied in the comments with just an image and no text. This kind of response is common on X; attitude arrives before technical discussion.

Kuldeep Pisda asked more specifically: Are the lasers still in pluggable modules on the faceplate? He added that lasers inside the package failing has always been a major concern for CPO. The question is not harsh, but it hits an engineering reality. Putting the optical engine into the package solves the electrical link, but the laser's heat, lifespan, and replacement method still need to be explained. The official post did not elaborate.

Capital & Codes said that copper has already hit a physical wall at these scales, and moving silicon photonics into the switch package addresses thermal density and interconnect power consumption. This assessment matches the selling points NVIDIA gave. But hitting a wall does not mean copper will immediately exit; short-distance and low-cost scenarios will still use it.

Garrison Simon asked how this compares with Broadcom's similar solution. NVIDIA did not provide a comparison in the post, nor did it lay out mass-production scale, port speeds, or actual power-consumption data. For buyers, these are more useful than the "first" title.

fueloncker directly made a meme, writing NVIDIA as Ratvidia. Stock sentiment and product news mix in one timeline—typical X.

Another account relayed the same news with a vertical video.

One judgment: CPO will first land in high-bandwidth switches, especially hyperscale clusters. Ordinary data centers do not need to rush to change their procurement lists. "Full mass production" is NVIDIA's claim; third-party shipment and field-maintenance data have not yet been seen. The comments asking about lasers first show that people were not convinced by the two words "first." That is actually normal—engineering questions have to wait for engineering data.