Earlier this week, Google partially U-turned its decision to make the on-device Gemini AI exclusive to the Pixel 8 Pro. Despite initially citing "hardware limitations" as the reason why the base model of the Pixel 8 would not be included, Google has now announced that the Gemini Nano (a lighter version of the software) will be included on both devices after all.

Given the very similar specs of both versions of the Pixel 8, that seemed a bit odd; what exactly were the hardware limitations that would make it difficult to use the Gemini Nano on a lower-end smartphone?

Courtesy of Seán Chau, Google's VP of Device and Software Services, we now have the answer. It is in RAM.

"With 12GB of RAM, the Pixel 8 Pro was the perfect place to put [Gemini Nano] on a device and see what it could do," Chau explained on the Made by Google podcast. "When we looked at the Pixel 8 as an example, the Pixel 8 has 4GB less memory, and it wasn't that easy to say, 'Okay, let's enable it on the Pixel 8.'"

It doesn't look like it can be implemented without cost, and as a result, Google will have to go through some hoops to try it out.

"We will be making Gemini Nano available on the Pixel 8 behind the developer options so that developers who know exactly how it will impact the user experience can use this feature," Chow explained. Of course, users who understand the potential impact on the user experience will also be able to use this feature."

In other words, the extra 4GB RAM that the Pixel 8 Pro has seems to make all the difference in making the phone's usual non-AI features run smoothly. And if generative AI models like the Gemini Nano are not just a passing fad, it means the smartphone may suddenly need a big RAM boost.

Currently, Android devices typically have 8 or 12 GB of RAM. Some budget devices have only 3GB, while top-end devices boast an excessive 24GB, and there are outliers on both sides, but the majority are in the 8-12GB zone.

At the moment, that's fine; Android Authority has done extensive testing on this, putting phones with varying amounts of RAM through their paces on various tasks, and their conclusion is simple: "8GB to 12GB of RAM is ideal, 16GB seems to be merely boastful.

However, if generative AI for consumers is not a mere blip, these numbers may soon be revised upward.