Real sections are made of parts, and a part's own $I$ is quoted about its own centre. Move it away from the assembly's neutral axis and it becomes far more useful than it was. The parallel axis theorem prices the move: $I = I_c + A d^{2}$ — read aloud I equals I-c plus A d squared.

Name the letters once. $I_c$ is the part's second moment of area about its OWN centroidal axis, in mm⁴ — the subscript c stands for centroid. $A$ is the part's area in mm². $d$ is the distance between the two axes, centroid to centroid, in mm. $I$ is what that part contributes about the NEW axis. And $A d^{2}$ is called the transfer term.

Run one flange plate through it and the point makes itself. A 200 × 20 plate has an $I_c$ of about 0.13 × 10⁶ mm⁴. Set it 260 mm off the axis and its transfer term is 270 × 10⁶ mm⁴ — two thousand times larger. The plate's own stiffness is a rounding error; what matters is how far out you put it. That single fact is the reason the I-beam is shaped the way it is: shove the material to the flanges, where $d$ is big, and leave a thin web to hold them apart.

For a doubly symmetric I-shape you can skip the part-by-part bookkeeping and take the box minus the voids: $I = \dfrac{B H^{3} - (B - t_w)(H - 2t_f)^{3}}{12}$, where $B$ is the flange width, $H$ is the OVERALL depth, $t_w$ is the web thickness and $t_f$ is one flange's thickness, all in mm. The two subtractions each have a trap in them: the void is only $B - t_w$ wide, because the web is solid, and it is $H - 2t_f$ deep, because there is a flange at the top AND one at the bottom.