By Roman Aglaganov
Memory chips are in short supply, not least because AI data centres need so many of them. So when Sam Altman was asked recently about people running AI models on their own computers, he made an efficiency argument. “From a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer and being used 5% of the time,” he said in an interview with the Cheeky Pint YouTube show. In the cloud, he said, you get “maximum utilization”.
Where he is right
On its own terms, the argument holds. A server shared by thousands of users is busier than a laptop that sits idle overnight. If the only goal is to get the most computation out of every chip, centralising wins.
Where it falls short
But nobody buys a computer to maximise global chip utilisation. They buy it to have something that works for them, when they want, without asking permission. Local AI keeps your text on your machine; Opera’s new WebAI feature, for example, runs models on your own computer so websites can offer AI tools without sending what you type to anyone else’s servers. A local model can keep working without a connection. It does not change its terms when a company changes its pricing. And it cannot be switched off by someone else.
There is also a conflict of interest worth naming. As Tom’s Hardware points out, OpenAI’s main business is selling access to models that run in its data centres. An argument that memory belongs in data centres rather than in homes is also an argument for renting AI from OpenAI rather than owning it.
The better question
The real question is not whether memory sits in homes or in data centres, but whether people keep the choice. A world where every useful AI tool must run on someone else’s computer is more efficient in one narrow sense and more fragile in several others: for privacy, for competition, and for anyone whose connection, budget or country makes the cloud unreliable. The 5% of the time your computer’s memory is busy running your own model may be the most valuable 5% you have.
