On a podcast, OpenAI's Sam Altman put a question to the technology world that many found hard to shake: why do we keep so much working memory in our personal computers when, by his estimate, we use only a sliver of it? His remark, made amid talk of local AI, open-source models and a scarce global supply of computing power, reframes a familiar machine as a kind of idle reservoir. It also revives an old tension between holding resources close at hand and sharing them across the network, a choice whose costs become clearest when the connection itself falters.
Altman Questions Idle RAM in Personal Computers, Raising Questions About Cloud Dependence
On a podcast, OpenAI's Sam Altman put a question to the technology world that many found hard to shake: why do we keep so much working memory in our personal computers when, by his estimate, we use only a sliver of it? His remark, made amid talk of local AI, open-source models and a scarce global supply of computing power, reframes a familiar machine as a kind of idle reservoir. It also revives an old tension between holding resources close at hand and sharing them across the network, a choice whose costs become clearest when the connection itself falters.
On the podcast Cheeky Pint, OpenAI chief executive Sam Altman asked why people keep so much memory in their personal computers when they seem to use so little of it. The talk ranged across local AI inference, open-source models and the global shortage of computing resources, but the sharpest moment was his claim that a laptop's working memory, known as DRAM, is in use only about 5% of the time.
From a global standpoint, he argued, that is inefficient. Cloud data centres, where many users draw on shared machines, could allocate the same memory far more effectively. The contrast is simple: a laptop's own processor and RAM do the work locally, while in cloud computing the work happens in remote facilities and users connect over the internet.
The Economic Times piece weighs what that model costs the user. Cloud services depend on a steady connection, and when it breaks, access to applications, online storage and AI tools can disappear, leaving files unreachable if they were never downloaded. For people who travel, work remotely or live where signals are weak, that dependability cannot be assumed.
The article also makes the case for keeping work on the device. Local processing means data need not leave the machine, and for some tasks it can be quicker because nothing has to travel out and back. The source does not explain how Altman reached his 5% figure, nor does it describe any plan to change how personal computers are built or sold. What remains to be seen is whether this debate begins to shape hardware choices and data centre demand while computing stays in short supply.
On a podcast called Cheeky Pint, OpenAI chief executive Sam Altman raised a question that spread quickly online: why do people keep so much memory in their personal computers when they use so little of it? The conversation was about local AI inference, open-source AI models and the global shortage of computing resources.
Altman's specific point concerned DRAM, the working memory in a laptop or desktop. By his account, that memory is in use only about 5% of the time. Looked at as a matter of global resource allocation, he argued, it would be more efficient to put that memory in cloud data centres, where many users can draw on the same shared computing resources.
The distinction he was pressing on is simple. When someone opens a programme on a laptop, the device's own processor and RAM do the work. In cloud computing, the processing happens in remote data centres run by companies, and users reach those resources over the internet.
The Economic Times piece uses the remarks to examine what the cloud model costs the user. Cloud services depend on a working network connection between a person's device and remote servers. If the internet is disrupted, a user can lose access to cloud-hosted applications, online storage, remote processing and AI services. Files kept online can also become unreachable, particularly if they were never downloaded for offline use.
That is why the piece argues local computing still matters. Not everyone has a stable connection all day. Many people need to work while travelling, work remotely, or live in places where connectivity is poor, and for them a dependable link is not a given.
The article adds a second argument for keeping the work on the device. Running applications locally means users do not have to send every request or piece of information to an external server. For certain tasks, local processing can also be faster, because the device does not have to send data out and then wait for a reply.
The remarks arrive in the context of a global shortage of computing resources, which is the backdrop to Altman's efficiency argument. The source does not say how he arrived at the 5% figure, and it does not describe any specific proposal to change how personal computers are built or sold.
What to watch is whether the debate between local and cloud AI inference begins to shape hardware choices and data centre demand while computing remains scarce.
Envie de l'histoire complète ?
Lire l'original sur The Economic Times ↗
Technology