> I don't think it holds up.
Could you clarify where it doesn't hold up? The biggest issue I can think of with the loom as computer analogy is reusing previous work. Perhaps what you are describing about the boundary between the cards and the data (at least for a loom) is that you can't write data out the cards and then read it back in again. But even here, I'm not exactly sure that's theoretically watertight.
Suppose you had a seed set of cards that formed a chain. Based on that you weaved fabric which had strategically placed holes such that they could function like a card. Given that you can now write output in the same format you read input, does this mean you could use a Jacquard Loom as a Turing Machine (never mind whether someone would do such a harebrained idea)? And of course, just because you maybe could doesn't mean that anyone bothered, in large part for the reasons that the inventor of the tabulator machine chose to use punch cards (rather than fabric) in the first place!
That machine you reference does seem to be the first machine that can both write and read to its input medium, the punch card, and from there, I suppose, it is history. Still, I'm convinced that there's something very interesting and maybe core to computation embedded in humanity's usage of the loom and in fabric itself. We can earnestly consider it a predecessor to the printing press (and tabulating machines) in how it provided technology to store and preserve bytes of data (representing bitmapped images) in fabric. This is another grasp in the dark, but it almost reminds me a little bit of the FFT.