Concurrency
The number of operations in progress at once — distinct from parallelism, which is how many are literally executing simultaneously.
Concurrency is a structuring property (many things in flight, interleaved); parallelism is an execution property (many things running at the same instant on different cores). A single-threaded event loop is highly concurrent and not parallel at all, which is exactly right for I/O-bound work and useless for CPU-bound work.
Architecturally the important quantity is the concurrency limit at each tier, because it is usually the real capacity ceiling: thread pools, connection pools, worker counts, and the platform's per-instance request limit. Little's Law sizes them.
Two failure modes to design against. Unbounded concurrency, where the system accepts everything and collapses under queueing rather than shedding load. And a shared limit, where one slow dependency exhausts a pool used by unrelated work — which is what bulkheads exist to prevent.