First off, referring to it as a “limit” is incorrect. In engineering you have what’s called an S-N diagram, which plots out the average time to failure based on average cyclic stress. Basically, a lower avaerage stress results in a higher average life. Also, this plot uses a logarithmic scale for both axis, because then all of the plots are straight lines.

For steel, the S-N diagram has what’s called the “knee”, which is where you have two distinct lines in the S-N curve: one horizontal and one at an angle, with the two intersecting at 1 million cycles. Referring to the knee as a limit (like in the article) is wrong because it’s not a limit; it’s the threshold where if you design a part to last beyond that (aka less cyclic stress than would get 1 million cycles) then it practically lasts forever.

In reality, the part won’t actually last forever, since the S-N curve beyond 1 million cycles isn’t perfectly horizontal. It’s just that reducing your cyclic stress quickly increases your predicted life into billions or even trillions of cycles. This is known as ultra-high cycle fatigue, and it’s generally impractical to do all the testing required to model because each sample would take months to test on the low end. Plus, there’s little demand for such models in the industry, though there are a handful of PhD students and post-docs working on it