Your options are multi-pod and multi-site if you don’t want to replace your spines.
Multi-Pod lets you leverage your existing APICs and if this is basically one large Data Center that shares policy this may be your best bet. It also a good approach for larger data centers just to reduce failure domains. So you will need:
- IPN routers or L3 switches (2 for redundancy)
- New Pod2 Spines (and here for a larger DC you may want to look at moving away from the fixed configuration spines and move to a chassis based spine...will give you some good flexibility in the future)
- New Leafs
You will need spine interfaces for the IPN so you will have to free up some interfaces for that to make this work.
Multi-Site can be done within a Data Center. I have a client looking at just such a design within their data center because they have a large dev/test area and so for them it makes sense to have two separate fabrics with some commonality and managed uniformly even in one location. But this is two separate fabrics.
There is no easy answer...the downside the fixed form factor switches when you get to the scale you are dealing with :D.
The downside of going with Multi-pod/Multi-site deployment in this case is bandwidth constraints for your east-to-west traffic from one pod/site to another. Without adding to much complexity to current network, I would just upgrade the Spines modular 95xxs
There is yet another design option in this case which is ACI Multi-tier, read through the white paper and see if this is a better fit for your network