Compare the scrolling behaviour of OPs example table, with the scrolling behaviour of Excel or Numbers.

With OPs version you get immediately lost in the grid because the headers don't float along with the scrolled view, which makes this best case useless, worst case harmful (e.g. confusing important columns).

If you have thousands of columns you're not going to do manual eyeballing of values anyways, you'll lose track because scrolling is not accurate enough, so you might as well represent rows as documents which densely shows the fields/columns and values, and which can then in turn be shown in a vertically scrollable list.

The problem you describe isn't horizontal scrolling, it's scrolling as implemented by OP. It also applies to vertical scrolling, as the headers don't follow around.

Yeah, but on one dimension that's workable. On two dimensions it's useless.

Having the headers float makes 2 dimensional scrolling usable but not great.

Why do we store data as rows and columns? To be able to quickly compare them. Is that possible with 50 columns? No.

If you have both horizontal and vertical scrolling, gesture-scrolling systems (like touchscreen and two-finger trackpad) behave weirdly. It either wobbles back-and-forth, or it locks to a particular direction.

Also, mice with physical scroll wheels can only scroll in one direction.

Most mice these days have wheels that can be tilted to scroll sideways. It's not perfect, but it works.