Yes. This is what I came to say. Big-O assumes a model of computation. Typically people assume a random access machine with certain constant time operations like arithmetic. In some models arithmetic might not be O(1) (e.g., in a pure lambda calculus it's O(n)). A random access machine model makes a bunch of simplifying assumptions that don't actually hold in practice. As the article suggests, if you're writing code in industry it's not a bad idea to understand how your specific hardware differs from the model.