And maybe through 2027. Why? The RAMpocalypse. The RAMageddon. AI hyperscalers are panic-buying wafers and locking down production, so RAM and SSD prices are ridiculous. Macs used to be a great deal, apparently due to supply chain inertia, but after the recent price hikes, that's not the case anymore.

I usually like to upgrade my personal tech gear (laptop, smartphone, tablet...) every 1-2 years. Can I still afford it? Yes. Do I need it? Not really. One thing I hate is feeling silly. And I would feel very silly if I paid for RAM twice.

The official explanation for the extra cost is "RAM is scarce now". Why? Because it's all going to AI datacenters. My point is: you don't need to pay for both. If you are actually going to run local LLMs, go ahead and buy the hardware. If not, just use the cloud, the ram you can buy is expensive because they're using it all for the cloud. And your existing devices are 100% functional for everything else, most probably, unless they're very old. HODL your devices. The RAM shortage has been "predicted" to last till 2028 by some of the people directly handling it. We'll see. It looks plausible.

That's it, that's the main point of the post.

Here you have an actual photo, totally non-AI-generated (ok, I lied here) of the Great God Pan causing literal "panic" in the RAM market 😂

Yeah I use AI to generate images and a few other things, especially at the heavily subsidized prices of the 20€/month plans, while they last. But I don't depend on it. I wouldn't pay for API usage to generate a silly image, instead I would post this without an image or I'd paint stick figures myself. In any case this isn't stealing a job from a professional illustrator because I wouldn't pay for that either. It's fun. But I don't need it.

Now, if you're still reading this, I'd like to clarify a few more details. The second part of this post is going to be a review of the previous points. Bear with me, it's my thinking process, I'm typing this manually, this text is not LLM generated. I'm not even going to correct it, so enjoy my probable mistakes (as English is my third language) 🤓

Mac vs PC vs common sense

Macs used to be a great deal indeed, especially the Mac mini M4 at launch for 719€ (2024, VAT included, in Spain) which you could actually get for 500-600€ for a while (sales, barely used by impulse buyers, etc.) Now the base model (same storage, same RAM, same CPUs/GPUs/NPUs) is 969€. So you pay 250€ more for the processor that was awesome in 2024. Not such a good deal.

A maxed out MacBook Pro to run (some, not too big) LLMs locally used to cost around 7.300€, but since June 2026 the exact same laptop costs about 10.200€, an extra 2.900€ for exactly nothing, they'll ship you the same box with the same product inside. I'm not saying I was going to buy that monster, but it's a good example. iPhone prices are expected to increase 200-300€ each too, which is a market much bigger than M5 Max laptops.

What about PCs? Well, they're sometimes a better option, it depends on what you do. PC hardware is severely lagging behind Apple if you care about size, temperature or noise, but the OS can make a positive difference, especially if it's Linux. I recommend this whole YouTube video by Theo, for the entire content, but especially around minute 13:30 where a file benchmark takes 35 seconds on a maxed out MacBook Pro, and just 2.5 seconds on a 700€ gaming mini PC (BTW it's the GMKtec NucBox K8 Plus AMD Ryzen™ 7 8845HS with 32GB DDR5, 1TB SSD)

Some of the reasons for this Linux superiority for development workloads according to Theo, and I agree, are syspolicyd (kind of an antivirus / antimalware integrated in macOS which analyzes what processes do) and the APFS (Apple File System). If you're interested I recommend watching the video as it's much better than anything I can explain here.

What to run and where?

In Theo's video, you can see he's using a maxed out MacBook Pro M5 Max with 128GB, and he's actually using about 32GB of it, while probably being one of the most advanced / intensive users ever, and knowing he's got all that memory at hand. I owned a gaming PC with 64GB which I used for several experiments and I also had a really hard time using more than 32GB most of the time running realistic development workloads.

If you're curious about what happened to that PC of mine, I sold it because it was big and noisy, the CPU and GPU were getting obsolete, and I was sick of rebooting Windows a dozen times every weekend I wanted to play.

You only need insanely expensive computers with large amounts of RAM such as 128GB to run LLMs locally. And then you'll find out the best models don't fit anyway, but that's another story. Models that do fit in 128GB (minus a few GB for OS, browser and agents or whatever) are very capable. Arguably, you can run meaningful tests with heavily quantized models with 64GB or 32 GB, or even less (but as you reduce the amount of memory, forget about using it for actual code generation).

At any rate, are you sure you need to spend all that money in hardware that's going to be obsolete in a few years? You can pay for a LOT of cloud usage if you're smart, with that money.

And we're talking personal usage here. You pay for the RAM in full. Then, are you going to use the hardware 24x7? Are you going to find 2 friends or coworkers and run 8 hours shifts on the laptop? Seriously? It's way more efficient just using Ollama Cloud or OpenRouter strictly for the time you need them. Unless you're running insane amounts of inference, which most people don't.

Or simply pay for the heavily subsidized 20€ or 200€ plans from OpenAI or Anthropic, while they're still there. Just make sure you understand you'll be actually using thousands of euros worth of computing and these prices won't last forever.

Oh, and all this is to run AI workloads. For any other workload including development, the computer you're currently using to do it is still fine. You can wait unless it's extremely old (and in that case I hope you're not waiting for my advice, you've already ordered the replacement and thats good 😅)

More, or better?

The topic of keeping your gear for longer reminds me of something related: efficiency. Some people are proud of spending a lot. That's always happened, it was Louis Vuitton bags and caviar before, now it's tokens. So many lines of code. So fast. Wow. Ooops, your fancy hello world needs a 2.000€ computer to run. That's how things are, right? 😅

Ahem. Nope.

A former Microsoft engineer has rebuild Notepad, it's 2.5KB in size, and I bet it could run on any PC built in the last 3 or 4 decades. All because it doesn't have any of the bloat. It doesn't analyze what you do and report back home, it doesn't require a cloud account, it doesn't show ads, it doesn't have assistants or agents. It's not built using layers upon layers of abstraction, libraries and frameworks. And the few features it has, have been built by an actual expert. It's just what it has to be, nothing less, nothing more. Not inferior, just as useful as ever.

If everything was built like this, RAM would be nearly free because we wouldn't need so much of it. Same with electricity, water, and other resources.

Yeah, we're talking about our personal devices in this post. But what if you applied this to enterprise workloads? Who wouldn't love to reduce their cloud bills a lot, maybe even an order of magnitude? Well, a hint: you won't do that by writing more lines of code, faster.

HODL

Stay rational. Hang on there. Reality will catch up.