I think it's possible that with the move to TSMC 3nm, and attendant power-consumption/heat reduction from the process advancement, we may be offered an M3 Pro option on the 15" MacBook Air.
Very unlikely, for the reasons outlined by @
senttoschool. But I can imagine that upcoming Mx series could be capable of higher performance and might have few extra spare watts in the bigger enclosure to flex their muscle.
I suspect they will make a tradeoff of lower energy usage and increased speed, because that translates to better battery life under load.
Battery life under load is not really an interesting metric, at least not with today's technology. Let's say you are running some moderately demanding CPU-driven workload on a M2 Pro. Your system power consumption will be around 30 watts, 20 of those coming from the CPU itself and the rest being RAM, I/O devices, as well as supporting circuitry. With the 70 watt-hour battery of the 14" Pro you are looking at 2 hours and 30 minutes of battery life. Let's say you have reduced your CPU power consumption by whopping 25%. Now it draws 15 watts instead of 20. Your resulting battery life is 2 hours and 50 minutes. You only gained 20 minutes (or roughly 10%), which is not going to make any difference.
Under load, it makes much more sense to optimise for energy instead of power, at least for laptops. If you want good battery runtimes, the key is minimising the baseline power consumption. The reason that Apple Silicon has such an excellent battery runtime is not because the CPU/GPU are more efficient. The M2 Pro will draw the same amount of power as a similar class x86 mobile CPU running a long multithreaded workload. The battery life is so good because Apple Silicon consumes almost no power while idling, which is achieved by aggressively powering down circuitry and busses when not needed, using custom RAM with extremely low baseline power consumption (lower than standard LPDDR5!), and caching the data using the power-efficient on-chip SRAM. Apple even goes as far as maintaining an expensive on-chip framebuffer cache for the display controller in order to save power transferring that data from RAM.
They will want to try and recapture the single-core speed lead that they held briefly with M1, and at the same time also bring down energy and heat.
If anything, they are more likely to increase the power consumption of the chips in order to improve performance. Doing so will not compromise the battery life (because baseline will stay very low).
The places where they still need to improve are mostly GPU speed, they still lack significantly compared to Nvidia’s desktop solutions which is not good for a lot of rendering, development and acceleration workloads.
Yeah, that's the tricky one. Nvidia is ahead both in the number of compute units and clock speeds. AD104 (the chip used for 4080 mobile) has 60 SMs compared to M2 Maxes 38 GPU cores at the same node size, and that with M2 Max die being 70% larger! Even if Apple were to clock the Max GPU at 2Ghz it would still not be enough to catch up with the 4080 mobile in raw TFLOPS. They would either need to produce much larger chip to accommodate additional GPU cores (which would be extremely expensive) or use multi-die tech (which will likely consume more power). This is where the disadvantage of the SoC model comes into play. For desktop they can of course always go multi-die, but the cost of these products will likely remain relatively high.
