"CCNA said switch can have simultaneously bandwidth for all access ports and router can only have sharing of bandwidth"
Hopefully, there's was additional context to that CCNA's pronouncement, because it's certainly not always true.
Before we even get into ACI, keep in mind "switch" and "router" are very generic, usually also implying "switch" is doing L2, while a "router" is doing L3, but for some time now, even some "L2 switches" often provide some very basic L3 routing capability (above and beyond, for years, smart/enhanced L2 switches might do things like support L3 ACLs), while router's can often bridge L2 traffic. L3 switches, muddy the waters.
This is especially true with the old 6500 "switch" vs. the old 7600 "router" which might be populated with EXACTLY the same line cards and supervisor module(s) and, I recall (?), at one time would run EXACTLY the same IOS binary! Of course, though, their chassis were "different".
ACI approach is much akin to how a chassis switch will operate between its supervisor module(s) and its intelligent edge/line cards.
In brief, I find it difficult to make clear, in some ways, there's nothing really new about ACI, other than it being all wrapped up in one package.
Forget ACI and/or leaf/spline architecture, and consider the classical 3 layer architecture, core(L3)/distro(L3)/access(L2). Do we really/truly need the L3 core? What if, we pull the core device and full mesh the distro (L3)? What if, we replace the L3 core with a pure/dumb L2 core and have all the distro have one IP'ed interface same network, same L2 broadcast domain? What are all the implications, good and bad?
Consider, your network is "small enough" (4032 or less edge ports) that it could be handled by VSS 6800 pair using IAs. Is that better or worse than a traditional 3 layer network? Which, in theory, is easier to manage?
Back to ACI, leaf/spline, it's sort of the "external" architecture of a stackable (spline) switches to which you can attach edge (leaf) switches. In theory, one unified logical device which you can increase the capacity of by adding stackable (spline) switches and/or edge (leaf) switches. Can you see how such an architecture might deal with bandwidth issues and/or processing issues, with (in theory) less complexity?
Again, I don't know whether the above will help in understanding ACI leaf/spline architectures, but, also again, conceptionally, you may have seen the same similar architecture on a much smaller scale than for a campus, where we all, from training and experience think of the classical 3 layer network topology.
The only other thing I'll add, keep in mind, the classical 3 layer network topology dates from only having L2 hubs and/or (very slow) L3 routers. Hardware technology has much advanced, which allows us to scale up other network topologies/architectures.
