Yes, AWS's Graviton2 is exciting, but no, it's not because they are more power efficient. That ARM processors are more power efficient is a stupid thing widely believed by stupid people. 
Amazon doesn't mention electrical power efficiency in their announcements. Instead, they mention price/performance efficiency. aws.amazon.com/about-aws/what…
In the 1980s, the "instruction set" dictated the "microarchitecture". The RISC microarchitecture was better, so everyone has naturally assumed the RISC instruction set is better. It isn't.
In the 1990s came OoO (out-of-order) microarchitectures that are better than RISC. They also decoupled the instruction set from the microarchitecture. Yes, ARM's instruction set is slightly easier to decode than x86, but this leads to no meaningful differences in efficiency.
Airflow around the mirrors makes a big difference in car aerodynamics, because the mirrors are large relative to the car. But when you scale up to trucks, semis, and diesel locomotives, they become less and less important to efficiency.
You think RISC is a turbo charger, meaningful to both the passenger car and diesel locomotive. It's not, it's rear view mirrors that matter less and less as the rest of the system scales.
ARM's CPUs have never been more "efficient" in terms of power consumed per computation. Yes, they run in power sensitive devices like mobile phones, but they achieve low power consumption by slow computation.
The ARM Cortex A76 is the first design from ARM itself that is scaled to the same speed as Intel's chips, with a 4 issue out-of-order microarchitecture. It consumes roughly the same power per computation as Intel's chips. That's the microarchitecture at the core of AWS Graviton2.
What's exciting here isn't power efficiency but the change in the market. Our understanding of the market is one of domination by box vendors, server vendors like Dell, HP, IBM, and so on. It's been moving away from that for some time. The new market is cloud/data-center vendors.
We should've seen this 20 years ago when Google stopped buying "boxes" from server vendors and started buying raw motherboard and building its own boxes. Then started building their own motherboards. Building your own chips is the next logical evolution.
Only, building your own microarchitecture is difficult. You don't want to do that. Instead, you want to build chips. Just get a microarchitecture for a CPU from somebody else, then combine that with the features you want, like more memory channels, encrypted memory, accelerators.
That's not a market Intel can easily play in. They sell chips themselves. But it is a market ARM can play in. They don't sell chips, they sell designs. They've never made a chip. Lots of other people make chips based on designs licenses from ARM.
The first generation AWS Graviton 1 was based on the same ARM Cortex A72 found in the Raspberry Pi 4. In other words, you could optimize your code on the RPi4 and it will run the same on Graviton1. But at the same time, the actual chips were utterly different.
The RPi4's chip is built on the aging 28nm process so it consumes a huge amount of power. The Graviton1 chip was built on the new 10nm process so is vastly more power efficient. Thus, huge difference even though they are the same microarchitecture.
Amazon's strategy will work because there's so many RPis in the world. SPARC and POWER CPUs aren't viable because developers can't get easy access to them. With ARM, developers can buy cheap RPis, Surface laptops, or AWS instances and develop for them.
This doesn't spell doom for Intel. They still make excellent chips. But it does spell doom for the server vendors of the world. Theirs is a shrinking world dedicated to legacy markets.
Anyway, you can buy Raspberry Pi shaped computers based on Intel's "Atom" processor. A good example is the Upboard. It performs roughly the same speed as an RPi4, but consumes about 30% less electrical power. Test the power efficiency yourself.
