MacBook Air M5 16GB vs 24GB

Before choosing a number, it helps to understand what you're actually buying. On a traditional PC, "RAM" is a separate set of DIMMs that talks to the CPU over a memory bus, while the GPU has its own VRAM. On Apple Silicon, there is one pool of fast memory physically packaged inside the M5 chip, and the CPU, GPU, Neural Engine, and media engines all read and write from it directly with no copying. That has two real consequences. First, Apple Silicon is more efficient with memory than a comparable PC — 16GB on an M5 Air feels closer to 20–24GB on a Windows laptop because there's no duplication between system RAM and VRAM. Second, that memory is part of the chip package itself — it cannot be removed, replaced, or added to. Whatever you buy on day one is what you live with for the next four to five years. This is the single most important fact about the 16GB vs 24GB decision: it is permanent.

Email, Safari with 20–30 tabs, Slack, Zoom, Apple Music, 4K video playback, Office documents, Notes, Calendar, Messages, Photos, occasional photo edits in the Photos app, light Final Cut iMovie work — 16GB handles all of this comfortably on Apple Silicon. The unified memory architecture and SSD swap are fast enough that the rare time you do exceed 16GB, you barely feel it. If your week looks like browsing, writing, video calls, and content consumption with the occasional creative side project, 16GB is the right answer and the $200 is genuinely better spent on more SSD storage or AppleCare+.

If "a few apps open" actually means Safari with 60+ tabs, Chrome alongside it for work accounts, Slack, two or three Zoom or Teams windows during the day, Notion, Figma in the background, Spotify, Apple Mail, Messages, and a couple of Electron apps like VS Code or Linear — that's where 16GB starts to feel cramped. The system won't fall over, but you'll notice swap activity: a small pause when you switch back to a tab you haven't touched in an hour, beach balls when you alt-tab between heavy apps, fans staying quiet only because the Air doesn't have any. With 24GB, you stay in memory across the entire workday and the machine feels noticeably more relaxed. For dedicated power users, the upgrade pays for itself in subtle smoothness over thousands of context switches.

This is where the $200 stops being a question. Photoshop with three or four large layered files open eats memory fast — each smart-object-heavy document can be 1–3GB resident, and when you stack a browser plus reference images plus the Photoshop scratch buffer on 16GB, the system starts swapping aggressively and you feel it as filter previews stuttering and brush lag. Lightroom with a RAW catalog of more than a few thousand images wants memory for previews and develop-module edits. Logic Pro sessions with 30+ tracks, software instruments, and convolution reverbs benefit clearly from the extra room. Final Cut and DaVinci Resolve on 4K timelines feel smoother on 24GB, especially when you have a browser, reference apps, and proxies running alongside. If you make money from a creative app or use one daily, get the 24GB and forget about it.

Modern development stacks are RAM-hungry in a way that catches new Mac buyers off guard. Xcode itself is fine on 16GB, but the moment you add an iOS Simulator (1–2GB resident), VS Code with multiple workspace folders and language servers (1–2GB), Docker Desktop with a few containers (2–4GB depending on what you're running), a couple of browser windows for testing and Stack Overflow, and Slack — you've blown through 16GB and the system is leaning hard on swap. Every developer we know who bought 16GB and runs a real stack regrets it within six months. 24GB is the genuinely correct pick for any Mac that's used for serious development. The $200 is a rounding error on what your time is worth waiting for builds and simulator boots.

Running language models locally with tools like llama.cpp, Ollama, LM Studio, or MLX is one of the most memory-sensitive workloads on a modern Mac, and it's a use case that's grown rapidly for developers and researchers. As a rough guide: an 8-billion-parameter model in 4-bit quantization needs about 5–6GB of memory just for weights, plus context buffer, plus everything else running on the machine. On 16GB you can run an 8B model but you have very little room left for the OS, your editor, and a browser, and you'll see thermal throttling on long-running prompts. A 13B model in 4-bit quantization is closer to 8–9GB resident and barely fits at all on 16GB once the OS overhead is counted. With 24GB, an 8B model is comfortable with the full development stack alongside it, and a 13B model fits with real headroom. If local LLM inference is part of your workflow — even occasionally — 24GB is the practical floor.

Apple's base memory tier has steadily climbed over the last decade — 8GB, then 16GB starting with the M4 generation, and there's a reasonable chance future Air generations will start at 24GB. Buying a 16GB M5 today means buying the floor of the floor, which is the configuration that ages worst and resells worst. Used MacBook listings reliably show that the upgraded-RAM model holds its value better, especially three to four years out when the next-generation Air models have moved on and 16GB looks dated. Across a typical four-year ownership window, the resale gap on a 16GB vs 24GB Air on the secondary market frequently meets or exceeds the original $200 spread. The upgrade often costs less than nothing in real terms.

Spread the $200 across the realistic life of a MacBook Air — four to five years for most people, longer for many — and the upgrade works out to roughly $40–$50 per year. That's less than a single streaming subscription. In exchange you get smoother multitasking every day, better creative app behavior, room for local LLMs, less swap wear on the SSD, and a meaningfully higher resale value at the end. The only people for whom 16GB is the obviously correct call are tight-budget light users who genuinely will not push the machine. Everyone else should take the upgrade. The mistake people regret on a Mac with soldered memory is almost always picking too little, not too much.