So, How Much RAM Do You Actually Need?

OpenAI CEO Sam Altman thinks it’s quite terrible for people to have RAM in their computers rather than in a data centre, as the scarcity of memory caused by the AI boom continues to drive up costs. This has led to plenty of memes, parodies, and humour, including:

Of course, I’m probably one of those that Altman seems to be targeting. After all, both my Mac Studio and Dell Pro Max GB10 have 128GB of RAM. Heck, even my laptop has 64GB. And with the tasks I perform, I often max out the RAM usage on all of those systems (especially the Dell Pro Max GB10, where I’ve had to do a great deal of optimization to ensure my data science workflow runs adequately in 128GB of RAM).

And the AI boom and enormous demand from data centres have definitely driven up RAM and storage costs this past year.

The same Dell Pro Max GB10 I got this time last year have now ships with half the RAM and storage but boasts a 50% higher price tag! Even 32GB of RAM in a desktop or laptop is a premium feature this year with a hefty price increase compared to last year when inexpensive 32GB laptops were everywhere.

Now I’m not the average user when it comes to RAM needs. Most of us can get away with less RAM… but not much less and you shouldn’t feel bad about that for the sake of data centres. We need to run apps locally for many reasons (privacy, cost, control) regardless of what Sam Altman thinks.

So how much RAM do you actually need? (shameless title repeat)

Well, that depends on your use case. Everyday computing tasks like web browsing, email, office applications, and video streaming will probably be fine on a Mac or PC with 16GB of RAM. Even software developers often get away with 16GB of RAM most of the time (32GB if they are testing larger deployments locally). People who often edit video require at least 64GB of RAM for high-res effects and rendering buffers. And for data science and AI development, it depends on the dataset size and how much you’re running locally (vs. on a cloud server). Large datasets, Pandas DataFrames, notebooks, machine learning models, LLMs, feature engineering, and model inference caches could easily require the amount of RAM I use (128GB).

But this also depends heavily on how much stuff you are running. For example, consider a software developer running:

• Visual Studio Code or another IDE like Xcode
• A browser with 20 tabs open and 12 extensions running as background processes
• Docker Desktop with several containers (that don’t have sensible memory limits configured)
• A local database, which requires additional RAM for query caches, indexes, and buffer pools
• A development server in a virtual machine that is provided 8GB of allocated memory

None of those applications is necessarily outrageous in isolation. Collectively, however, they can consume a substantial amount of memory… and that’s before the operating system get their share for reserved system use and caches. And your choice of operating system matters: Linux requires less RAM than macOS, and far less RAM than Windows.

When it comes to data science, there are plenty of times where you may chew up more RAM. For example, a CSV file that’s 5GB on disk doesn’t necessarily require just 5GB of RAM when loaded into Pandas. Depending on the data types, the DataFrame can consume considerably more memory. And guess what happens when you create intermediate DataFrames, perform joins, apply transformations or convert data types? More RAM is used.

Don’t believe me? Try something like this out with a 5GB dataset:

import pandas as pd
df = pd.read_csv("dataset.csv")

# How much memory is the DataFrame consuming?
print(df.memory_usage(deep=True).sum() / 1024**3, "GB")

# Memory consumption by column
print(df.memory_usage(deep=True) / 1024**2)

That said, here’s what I consider a practical RAM guide for 2026 when buying a new computer or upgrading the RAM in an existing one:

How can I get away with less RAM?

If you’re one of those people who keeps everything open indefinitely, it’s time for a change. Close apps you’re not using. If a website is important, you can bookmark it for later. You don’t need to keep 47 tabs open just in case you might need them again!

Also find out what apps are using the most RAM… you’ll likely find many you don’t need running. Stop them and configure them to not start at boot time. To view app RAM usage on Windows, you can open Task Manager (Ctrl+Shift+Esc), highlight the Processes tab, and sort by the Memory column. Or you can right-click Start, choose Terminal (Admin) and run the following in Windows PowerShell:

Get-Process |
  Sort-Object WorkingSet64 -Descending |
  Select-Object -First 15 Name,
    @{N='RAM (MB)';E={
      [math]::Round($_.WorkingSet64 / 1MB, 1)
    }}

On macOS, you can open the Activity Monitor app, click Memory, and sort by the Memory column. Or you can open the Terminal app and run either the top -o mem (press q to quit), vm_stat, or memory_pressure commands. On Linux, just open a shell or Terminal app (any one) and run top -o %MEM or ps aux --sort=-%mem | head -15.

If you are working with datasets, load only the data you need and pay attention to data type storage. For the earlier CSV example, you could:

# Load only the columns you need
df = pd.read_csv(
    "dataset.csv",
    usecols=["age", "income", "category"]
)

# Use more appropriate data types
df["age"] = df["age"].astype("int16")
df["category"] = df["category"].astype("category")

Summary

So, how much RAM do you actually need? Probably less than you think. But probably more than you used to. And if you’re still running 8GB in 2026, perhaps it’s time to start thinking about an upgrade. Just don’t tell Sam Altman. He might complain that your 8GB could be put to better use somewhere else.