Yields are the main barrier to single pieces of silicon that big. You have a certain chance of getting a defect per square mm of you chip and as they get bigger the chance of a bad defect gets higher and yields go down. Often they'll occur in a place where you can just disable a core or bank of cache and still sell the chip but not always. So yield rates tend to go down as chips get bigger. Also larger chips make less efficient use of the wafer.

Economically, there aren't so many people looking for 1000 cores that it makes sense to put in the NRE to assemble a giant package to put all of that in versus just selling a system that can have multiple sockets. Cooling limits also make spreading out work across multiple sockets a better choice.

Is there a market need for 1000 cores running simultaniously that doesn't already need all the other supporting infra you get from a couple extra "full fat" servers each running 64 cores? Presumably if there is any need, it's currently being done by gpus

Heat, energy usage.

Memory bandwidth. Top end Xeon Phis have 288 threads. They need high bandwidth memory and the Xeon Phis are limited to only a few gigabytes of HMC.