Nvidia made a point to say they used copper to connect their latest GPU's. They said optics was too expensive. That's maybe fine when there's 400g per lane coming off the chip. How about when it double that? How about when it's quadruple that? The cost per bit will be so "worth it" in optics soon especially when Suresh unleashes the road to 3.2T, there's no way copper can conduct it's properties efficiently in that environment, the cooling, the cross talk, the stability just isn't there.

I think these companies are jockeying for a tech to get them out of this pickle, that's why Poet has engaged all the major players (that article about Mitsubishi supplying 50% of lasers for current data centres really hit home....I don't think they want to lose much business......do you?

Another fascinating point I learned is that the pluggable road map will go on for a long time, no way Broadcom comes in with co packaged optics if we can do a pluggable to 6.4T. The evidence is provided by a market researcher in these posts.

https://www.linkedin.com/authwall?trk=qf&original_referer=&sessionRedirect=https%3A%2F%2Fwww.linkedin.com%2Fposts%2Fmark-lutkowitz-b22abb2_thanks-to-mark-lutkowitz-of-fibereality-activity-7243928317324660737-0_C7%3Futm_source%3Dshare%26utm_medium%3Dmember_ios

https://www.linkedin.com/authwall?trk=qf&original_referer=&sessionRedirect=https%3A%2F%2Fwww.linkedin.com%2Fposts%2Fmichael-kanellos-20b803_we-will-be-with-pluggables-for-the-next-ugcPost-7243914989026058240-YUFg%3Futm_source%3Dshare%26utm_medium%3Dmember_ios