IBM wants to build a 100k-qubit quantum computer

Correct me if I am wrong but it is my understanding that one only gets speedups based on qubits being mutually entangled. I don't see how it would be possible to entangle 100k qubits. The time it would take to mutually entangle the bits would likely exceed the coherence time.

You could see speedups if you encoded the same problem in parallel on the same machine so you could compare statistics immediately rather than using multiple runs.