16B transistors at 5nm.
Architecturally it does not seem too exciting. They went wide and slow to get the efficiency where they wanted it. Manufacturing costs of M1 are probably way above what competition is currently doing.
It would be cool to know what the die size of M1 is but there has been no official data as far as I can find and I cannot estimate from the SIP image because no idea what the memory chips and their sizes are.
For some idea how large of a chip we are talking about - this is more than Radeon VII (13.6B) and almost the same size as GA104 powering RTX3070 (17.4B).
Renoir (4800U) is reported to have ~9.8B transistors.
Granted, M1 has additional stuff like Neural Engine, ISP, AES encryption, encoder(s)/decoder(s), Thunderbolt controller. On the other hand, some or most of that also exists for example in Renoir.
There are tradeoffs, amount of USB ports seems to be limited, it does not seem to have PCIe, max 16GB RAM and there is probably a reason it only runs up to 10W (if nothing else then marketing, definitely).
2.6TFLOPS for iGPU is impressive for the power envelope but again I am quite sure they went wide and slow to keep the power usage down.
For comparison, 2.5TFLOPs is something like RX560 can do with 3B transistors. Renoir does 2.1TFLOPs in desktop SKUs (8CU running fast). Both obviously use a hell of a lot more power.