The PCI bus has nothing to do with the instruction set. Usually it is just a block a designer can add to a chip, and connect to an internal bus like AXI, give or take a few other adjustments on the chip.
You can have PCIe buses without proper CPUs, even: it's quite common to find them paired with FPGAs.
For instance, Rasberry Pis have had a PCI bus for a few generations now, at first used for USB3. The Pi 5 breaks it out on a dedicated connector, making it easy to plug external devices: https://raspberrytips.com/pcie-raspberry-pi5/ (random link).
Of course, discrete GPUs are less ideal from a power efficiency perspective (duplicated memory controller, buses, and power circuits), so they wouldn't fit the Steam Deck. But write a big enough check, and I'm sure that AMD or Intel would be willing to share their iGPU designs. NVidia also makes Tegras.
For instance, Rasberry Pis have had a PCI bus for a few generations now, at first used for USB3. The Pi 5 breaks it out on a dedicated connector, making it easy to plug external devices: https://raspberrytips.com/pcie-raspberry-pi5/ (random link).
Of course, discrete GPUs are less ideal from a power efficiency perspective (duplicated memory controller, buses, and power circuits), so they wouldn't fit the Steam Deck. But write a big enough check, and I'm sure that AMD or Intel would be willing to share their iGPU designs. NVidia also makes Tegras.