NVIDIA spent fifteen years building CUDA before the market it was built for arrived. That head start, not any single chip, is what competitors have failed: an accelerator is only as useful as the software willing to target it.
On the hardware side the cadence has compressed to roughly a year: Hopper to Blackwell to Blackwell Ultra to Vera Rubin, each generation arriving with a rack-scale system rather than a part number.
NVIDIA is also the reason cross-vendor comparison is so difficult. It quotes sparse throughput by default, which roughly doubles the headline figure relative to, and most people repeating the figure do not mention which one it is.