Parallel sends 8 bits at once and serial sends 1, so why are fast modern buses serial?

Because at high speed and any real distance, parallel's apparent advantage collapses. The 8 (or more) parallel lines don't all arrive at exactly the same instant, propagation differences cause skew, and as the clock rate climbs, that timing spread between lines becomes a larger fraction of the bit period until you can no longer reliably latch all bits on one edge. Many fast-switching parallel lines also create crosstalk and EMI, and they cost pins, board area, and connector size. A single well-controlled serial line has none of the inter-line skew problem, so it can be clocked far faster, and techniques like differential signaling and embedded clocks push it further. The net result is that one fast serial lane beats a wide skewed parallel bus, which is exactly why USB replaced parallel ports, SATA replaced PATA, and PCIe replaced PCI. Parallel now survives mainly on-chip and very short, like memory buses. So "more wires" doesn't mean "more throughput" once speed and distance enter.