Whenever you put a question in your article title, you’re trying to lead someone to a conclusion before they even read the article.

We just got the scheduled video link, the press releases, and all the affiliate links from creators are swarming. Samsung Unpacked is coming! The Galaxy S24 is coming! A NEW era for Mobile AI is coming!!!

But is it though? Really?

Rumors and leaks are already detailing what we might see from Samsung in 2024, and it’s not particularly encouraging.

We have to take these leaks with a boulder of salt, and a lot of this could turn out to be bunk, but right now S24 looks like it will be another iterative year for Samsung with a number of consumer concerns.

The S23 series turned out to be a boon for consumers worldwide, as everyone got to benefit from the better chips produced by Qualcomm and TSMC. The days of the Samsung produced Snapdragon and the Exynos were fraught with poor battery life and thermal throttled performance. Often there’s a split, where Korean and North American Samsung phones get the better Qualcomm chips, and everyone else gets the Samsung chips. In 2023, everyone benefited from an improved performance per watt and significantly better battery life.

Moving into 2024, it seems that we might be back to some kind of division in products, chips, and regions. Rumors pointing to a new Exynos chip arriving in some variants of the S24 should be concerning. Samsung has still not demonstrated that they can match the performance of a Qualcomm SOC (System On Chip), especially in terms of heat management and power draw. It does us little good if the Exynos can get the same synthetic benchmark scores, if it drains your battery faster and runs hot in your pocket.

More concerning however, are these claims and promises of “New Era” AI. That’s the big headline on Samsung’s scheduled Unpacked livestream.

The entire tech sector is going crazy for anything labeled AI. It feels a LITTLE like the kind of fervor we witnessed near the peak of the NFT craze. Generative AI at least demonstrates SOME capabilities for enhancing workflow (where NFTs were largely just a rug pull scam), but getting results from an AI requires a substantial increase in compute resources.

Both MediaTek and Qualcomm were aggressively showing off the new AI components of their Dimensity and Snapdragon chips respectively, but that’s not the only part of the chain that improves AI. Yes, we need these little neural cores to process the data, but to get the data THROUGH those cores we need a LOT more RAM.

While attending the MediaTek Executive conference, we got some loose numbers on what kind of system specs a phone would need to use some of the more aggressive on-device AI capabilities of a new chip. AI needs access to a LOT of data, libraries, algorithms, and operations.

Looking at Stable Diffusion work loads, we’re all learning about new terms and metrics to compare performance. An “average” spec premium phone able to parse 7 billion parameters, moving to 16GB of RAM almost doubles the number of parameters at a similar tokens-per-second rate. This is of course based on MediaTek chips and designs (it’s almost exactly the configuration found in the new Vivo X100 Pro), but I would imagine these workloads are similar for Qualcomm silicon too.

In my interview with MediaTek executives, it was alluded that current smartphone AI performance is at the bleeding edge of file compression systems. Fully unpacking these libraries would require 20-30GB of RAM or more.

Using this new current generation of chips (Dimensity 9300 and SD 8 Gen 3) we can further drive on-device AI performance with more RAM at a reduced tokens/second. If we believe on-device generative AI is the future of mobile, the best way to future proof a phone purchase in 2024 is to buy well above your current RAM needs for today.

There is no daily smartphone task that will benefit from 20+ gigs of RAM. High level compute work, high resolution gaming, even increased multi-tasking in a desktop mode, we’re unlikely to reach those limits in traditional compute tasks. On-device AI demands MUCH larger pools of data and the fastest possible access.

What About the Galaxy?

Well, it seems Samsung is placing a bet against on-device AI, or at least predicting minimal user engagement over the next several years.

The concerning rumors, pointing to premium Samsung phones shipping with 8-12GB of RAM depending on what outlet or leaker is reporting on a rumor. While this is certainly adequate for high level traditional compute tasks, it’s at the lean end of the spectrum for any potential AI use.

Even the Pixel 8 Pro, with 12GB of RAM, often defaults to server side processing for its most exciting new AI features. This has lead to some criticisms of Google’s claims, and personally, using features like Video Boost becomes more of an afterthought, than something I consider doing out in the field.

Samsung seems like they’re likely to make a similar play, prompting consumers to upload data to a server for the AI magic to be applied remotely.

Is Samsung Wrong?

If I’m being honest, I’m worried that Gen AI could be another rug pull.

AI is an exciting new buzzword that prompts a TON of investment, but the actual practical future will be FAR more pedestrian than the promises from investors and developers. I think there’s potential in this market if we can reign in some of the insane power and performance costs. Mobile is likely the first market to make that happen. AI could be a LOT more exciting when it can co-exist with traditional phone use on a phone battery.

Samsung might not believe those use cases will exist in the next year or three, but that’s concerning for folks who are increasingly trying to hold on to their phones longer. If we find killer AI apps in 2026, and they require more than base level RAM to work well, S24 owners will face a pain point to upgrade sooner than they otherwise would have.

This is not a condemnation of Samsung in a vacuum. The competition Samsung faces in this arena is already available and operating at an incredible price to performance. The Vivo X100 Pro arrives with 16GB of RAM and incredible camera sensors, at a price near (or undercutting) an S23+. The Chinese OnePlus 12 comes in options up to 24GB of RAM, again near the price we would expect a Plus tier Galaxy. These phones well undercutting the prices we expect to see on the next Galaxy Note/Ultra.

Even if AI turns out to be the bust I’m worried it might be, having those components on phones could help in other areas, bringing more desktop grade experiences to mobile, improving Android’s ability to multi-task. One of the top manufacturer’s in this space though won’t be including those capabilities on the phones we expect to sell the most in PREMIUM Android Land.

We’ll have to see if Samsung does stay lean with RAM, and how that impacts their promises of a “New Era” for AI. It’s hard to see how this won’t have a chilling experience on the experiences other developers are looking to build for mobile AI applications.

If their apps don’t run well on an expensive Samsung, will developers want to field the angry support messages and reviews?