Excuse me for pointing out something stupid however Nexus 7000 are modular chassis as explained above, while Nexus 9300 are 1U or 2U devices with uplinks.
In the Nexus 7000 case, the interconnection between all line cards is done through fabric, however in the Nexus 9300 case the interconnection is done by means of physical cabling between all those switches. So in this case you need to ensure extra cabling and modules for an extra backbone between all of the 9300s. All that topology needs to be thoroughly planed as it will take over part of your switching volumes currently done between different line cards in-fabric. If you don't plan the traffic volumes precisely enough you risk ending up with congested backbone between switch A and switch B, and I would like to point once again, that backbone doesn't exist right now due to in-fabric switching between different line cards and modules.
In that case the migration from 7000 to 9300 may involve extra equipment including cabling and optical modules and I'm not sure if you already done proper design how to handle that in technical and administrative aspects with your datacenters. For example some datacenters deny MPO optic which higher the price between one 9300 and the other 9300 by 400% if not more.
In 2017 I started to deploy N3K-C3164Q in our core so that I can replace existing 3064s which were used for that purpose. The challenge was that I was lacking experience of all those different kinds of optics and cabling and I was learning about it on field. I still think I didn't make any errors in my cabling selection for the backbone in 90% of the cases however I could have bought 10 more 3164 switches with the price spent on optics and cables from that time on.
Nexus 7000 is what we are deploying now in order to serve end consumer access layer in our datacenter. I don't think anybody who was never engaged in this would ever understand how expensive delivery and management of backbone QSFPs and optics can be. I'm urging you to build proper plan including your new topology on paper and all costs set on the table so that you can consider all aspects of this migration. I suspect you are trying to lower down electricity costs by removing high-power 7000s and replacing them with low power 9000s; if so I can only recommend you not to forget my hundreds of thousands Euros spent on cablings and QSFPs for 4 years. None of this does exist now in your case due to 7000s being high density machines which perform your switching in-fabric.
Last but not least, this is a very extensive migration and you never mentioned your exact reasons till now. I mean, you need to explain what's currently not done properly with 7000s and what do you want to achieve with 9300s. Are you missing a feature or what? How big is your network? Why not do some diagram with rough estimations for your racks, how distant are they from one another, how many ports are you using, what's the failover concept currently used (bonding+LACP+VPC or STP etc). I think you can get more precise recommendations regarding your migration only if your exact use case is made aware of to the community out there.
Looking forward to your answer. Thank you.
