1. Google has only been saying it isn't focusing on outright performance with its Tensors but "AI" as an excuse - and fanboys lap it up. AI performance is NOT all about the TPU - the TPU is an AI accelerator, yes, but AI features require the processing performance of the whole chip. Just take a look at what Qualcomm means by its "AI Engine" - it is the CPU, GPU and NPU all working in unison. The fact of the matter is that to get AI functions happening on device, you absolutely need a high-performance chip.
2. There is no way the Tensor G4 is "based on the Exynos 2400" - it might have the same newer CPU cores, but the configuration and clock speeds are completely different, not to mention the custom AMD-based Xclipse GPU and NPU. Plus, Google kept the same mid-GPU from the G3 and the same TPU - so much for prioritising AI...What the TG4 likely shares with the Exynos 2400 is FOWLP, which has completely transformed the thermal performance and efficiency of the chip. Google claims 20% increase in efficiency for the TG4 and that will be why - plus, the new more efficient modem.
3. In addition to the better thermal performance the TG4 will have, Google has also added a vapor chamber to the Pixel 9 Pro series for the first time.