Superintelligence sits one rung beyond Artificial General Intelligence: where AGI matches human ability broadly, a superintelligence would substantially outperform the best humans at essentially everything. The usual argument for why it might follow AGI quickly is recursive: a system capable of doing AI research could improve itself or its successors, compressing progress that would otherwise take a field years into a much shorter span.
The concept is central to debates about long-run AI risk because a system this capable would be correspondingly hard to oversee and correct, which is why discussion of superintelligence is usually paired with discussion of control and alignment. It remains speculative; reasonable experts disagree sharply about whether, when, and how abruptly it would arrive.