Sam Altman’s RAM Argument Has Me Thinking About Who Controls My PC
My laptop does not need to be busy every minute to be useful to me. It needs to be available when I want to work. That is the thought I kept coming back to after Sam Altman’s comments about memory in personal computers and the efficiency of running AI in the cloud.

I can understand the infrastructure argument. Sharing expensive equipment among many users can keep it working for more of the day. But I also care about what I gain by having a capable computer of my own.
The argument was about scarce resources
In the October 29 Cheeky Pint interview, Altman discussed local inference in the context of constrained memory supply. He argued that allocating scarce DRAM to individual computers used only a small fraction of the time is inefficient compared with high utilisation in the cloud.
His reference to 5% utilisation was part of that argument. I do not treat it as a measured description of my laptop, every home computer or every local AI setup. Nor does the interview establish that a hardware shortage was deliberately created to stop people running models locally.
Efficiency is one benefit; control is another
A local model can be useful when I want to process files on my own machine, work without a connection or avoid a per-request charge. Those benefits depend on the actual setup. A local interface that calls a remote model is still using the cloud, and a locally running model can send information out through connected tools.
Ollama’s documentation is a useful practical reference: it explains how to see whether a model is loaded into GPU memory, system memory or both, and links to its hardware compatibility guidance. “Runs locally” describes where computation happens; it does not promise that every model will be fast on every computer.
There is also an important complication in the company-versus-local narrative: OpenAI released the open-weight gpt-oss models under Apache 2.0. A preference for cloud utilisation does not mean OpenAI has never supplied models for local use.
I start with the computer I already own
My ThinkPad E15 Gen 1 still does my everyday work well under Ubuntu. I have already had its battery replaced and its thermal paste serviced. That history makes me interested in extending useful life, rather than treating every new AI feature as a reason to buy another machine.
I have not benchmarked that laptop for this article. Before considering an upgrade, my checklist is straightforward: choose a specific task, check the model’s licence and download size, allow memory for the runtime and context, and see whether the result is useful at an acceptable speed.
That is the approach behind my earlier post on smaller local models and existing devices. A modest task may not justify an expensive graphics card.
The spending decision belongs to me
For occasional demanding work, paying for cloud access may be cheaper than buying hardware. For frequent private or offline work, local capability can have value that a utilisation percentage misses. Electricity, setup time and maintenance count on both sides of that comparison.
I want the freedom to choose that balance. Hardware sitting quietly on my desk is still doing something useful: keeping an option available when I need it.
Comments
Post a Comment