---
title: "A Bash Script Turns Sam Altman’s RAM Joke Into an Automated Complaint Letter"
description: "Ram-Altman reads /proc/meminfo every five minutes and mails you a joke email about your idle memory, once a day, at most"
author: "Luis Goa"
published: 2026-11-09T20:49:09Z
modified: 2026-11-10T00:14:10Z
url: https://rews.cc/a/a-bash-script-turns-sam-altman-s-ram-joke-into-an-automated--ee495f
language: en
tags: ["ai", "openai", "linux", "humor", "tech"]
publisher: "Rews (https://rews.cc)"
---

# A Bash Script Turns Sam Altman’s RAM Joke Into an Automated Complaint Letter

*Ram-Altman reads /proc/meminfo every five minutes and mails you a joke email about your idle memory, once a day, at most*

By Luis Goa · November 9, 2026 · https://rews.cc/a/a-bash-script-turns-sam-altman-s-ram-joke-into-an-automated--ee495f

## In brief

- Ram-Altman script emails users’ daily unused-RAM stats, parodying Sam Altman’s DRAM comments
- Script reads kernel’s MemAvailable field via /proc/meminfo, sends at most one email per day past a 10% threshold
- Altman told the Cheeky Pint podcast idle PC DRAM is a poor use of a supply-constrained resource used ‘5% of the time’
- J.P. Morgan estimates DRAM prices up over 400% from 2024 to end-2026; a 32GB DDR5 kit hit $449.97 in October
- Altman renamed his X account ‘Ram Altman’ and said ‘I want you to have so much ram’ after backlash

A Linux shell script with 1,400 GitHub stars now exists for one purpose: telling Sam Altman, by email, exactly how much RAM your computer is wasting. It reads /proc/meminfo every five minutes, waits until your available memory crosses a 10 percent threshold, and sends at most one email a day, built from a rotating pool of 8 subject lines, 10 message bodies and 6 sign-offs. The project, called ram-altman and written by a developer who goes by Iain Kay, ships without an email address in it; the README tells you to put in your own and to stop if anyone asks you to.

The number that triggers the joke is MemAvailable, a field the Linux kernel has reported since kernel 3.14, defined as the amount of memory a program could claim right now without forcing anything into swap. That is not the same as MemFree, which only counts pages nobody is using at all; MemAvailable adds back disk cache and other memory the kernel could reclaim on demand. On kernels older than 3.14, which lack that field, Kay’s script falls back to estimating availability from free memory plus reclaimable cache, which is roughly what MemAvailable does internally anyway. Ars Technica ran the script’s --dry-run flag on a Fedora 44 machine and got a concrete reading: 14.9 GiB of usable RAM out of 16 GiB of physical memory (the rest claimed by the kernel and firmware), with 10 GiB in use and 4.9 GiB, or 33.1 percent, available, as [Ars Technica reported](https://arstechnica.com/gadgets/2026/11/a-sam-altman-parody-offers-an-amusing-lesson-in-memory-management/). That clears the script’s default 10 percent trigger by a wide margin, which is the point: most idle desktops will clear it most of the time.

**How ram-altman turns idle RAM into email** The script's reporting pipeline, per the project's README

| From | To | How |
| --- | --- | --- |
| cron, every 5 minutes | /proc/meminfo |  |
| /proc/meminfo | used = MemTotal − MemAvailable |  |
| used = MemTotal − MemAvailable | available % > 10 (default) |  |
| available % > 10 (default) | already sent today? | yes |
| already sent today? | sendmail / msmtp / mail | no email sent yet |
| sendmail / msmtp / mail | your own address only |  |

Based on [Iain Kay, ram-altman README](https://github.com/IainKay/ram-altman)

None of this would exist without a specific sentence Sam Altman said on John Collison’s podcast *Cheeky Pint* on October 29, 2026. Pushed on whether local inference on people’s own machines could reduce the need for data centers, Altman argued the opposite case for memory chips: “it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM” sitting in personal computers, used (by his estimate) around “5% of the time,” when “you want maximum utilization in the cloud if you’re trying to deal with this very supply constrained environment,” as reported by [PCGamer](https://www.pcgamer.com/software/ai/sam-altman-argues-that-its-bad-you-want-dram-in-your-computer-and-not-a-data-center-actually/) and [TechRadar](https://www.techradar.com/ai-platforms-assistants/he-can-pry-it-from-my-cold-dead-hands-sam-altman-says-its-quite-terrible-that-youre-using-ram-in-your-pc-when-theres-not-enough-for-ai-data-centers).

The remark landed in the middle of an actual DRAM shortage, not a hypothetical one. J.P. Morgan’s research desk estimates [DRAM prices will have risen more than 400 percent](https://www.jpmorgan.com/insights/global-research/artificial-intelligence/dram-memory-shortage-from-ai) from the start of 2024 to the end of 2026, and ties a 23 percent rise in the software-and-accessories CPI since late 2024, plus a 37 percent jump in the import price index for computer parts, directly to the squeeze; the bank reckons every 10 percent rise in hardware costs adds roughly 0.1 percentage point to core inflation, putting memory’s own contribution at 0.2 to 0.4 points. On the ground, [Tom’s Hardware tracked](https://www.tomshardware.com/pc-components/ddr5/32gb-of-ddr5-6000-ram-now-costs-usd450-ai-induced-shortage-reaches-worst-milestone-yet) the cheapest 32GB DDR5-6000 kit at $449.97 as of October 14, 2026, up from roughly $400 in August and more than 500 percent higher than a year earlier. The mechanism is a wafer-allocation choice, not a mined-out resource: Samsung, SK Hynix and Micron can turn the same silicon into either high-margin HBM stacks for AI accelerators or low-margin DDR5 for PCs, and in 2026 they are overwhelmingly choosing HBM, with SK Hynix telling investors on its October earnings call that its DRAM, NAND and HBM capacity is essentially sold out for the year. Micron retired its 29-year-old Crucial consumer brand in December 2025 to redirect that capacity.

So Altman’s claim that idle PC RAM is a bad allocation outcome arrives from the one company whose own GPU clusters are a direct beneficiary of the reallocation he is describing. That is the detail the backlash seized on, not the arithmetic about utilization rates, which nobody seriously disputed; by November 1 the five most-viewed posts about the remark on X had passed 950,000 views between them. Altman changed his X display name to “Ram Altman,” replied to a user asking “why u want my ram then” with “i want you to have so much ram lets build more,” and [posted a longer clarification](https://x.com/sama/status/2117795243068133756): “I actually do want the world to have tons of memory and for people to put as much as they want in their computers. It is fucking annoying that this was taken out of context. I have argued for a long time that we should produce way more memory globally.”

That clarification reframes the dispute as a production shortfall rather than a claim about where existing chips should sit, but it does not change the fact that OpenAI’s own infrastructure buildout and a gamer’s DDR5 kit are drawing on the same fabs, the same wafer starts and the same finite 2026 capacity. Building more fabs is a multi-year project; TrendForce’s own tracking, cited by multiple outlets, puts server DRAM contract prices up a cumulative 64 percent in the second half of 2025 with roughly 270 percent more expected across 2026, and Gartner has floated a scenario where the broader memory cost surge runs to around 130 percent before it eases, stretching into 2027. None of the manufacturing-side numbers move faster because a CEO renamed his Twitter account.

## What the script actually proves

Kay’s project is careful to concede the asymmetry in Altman’s original claim: Altman was talking about time-utilization, the fraction of hours RAM sits idle, while MemAvailable measures capacity free at a single instant. A machine can show 33 percent available right now and still have been near 100 percent utilized an hour ago during a build or a game session; the script has no way to distinguish a desktop that is idle because nobody is using it from one that is idle because its owner just finished a workload. The README says as much, which blunts the joke on purpose: a satire tool that overstates its own case isn’t making the point it thinks it’s making. The guardrails matter more than the humor. The repository ships with 14 automated tests using simulated memory and a fake sendmail binary, logs every sampled reading to a local CSV rather than transmitting it anywhere, and caps itself at one email per address per day specifically so the gag can’t become, at scale, what it is structurally capable of being: a distributed denial-of-service attack built out of a few hundred people’s cron jobs pointed at one inbox.

The project has also spawned a companion site, ram-altman.lol, that rotates through the script’s sample output for anyone who wants the joke without installing anything. The repository carries 41 forks and sits open to pull requests, a fate shared by most successful jokes-as-software: eventually someone adds a feature nobody asked for, and then it’s a mailing list.

## See also

- [ram-altman GitHub repository](https://github.com/IainKay/ram-altman) — github.com · Source code, README and the script's safety guardrails
- [ram-altman.lol](https://ram-altman.lol/) — ram-altman.lol · Companion site showing rotating sample output without installing anything
- [Cheeky Pint episode 39 transcript](https://cheekypint.transistor.fm/39/transcript) — cheekypint.transistor.fm · Full transcript of the podcast appearance that started the controversy