Running AI on your PC is a waste of memory, according to Sam Altman, CEO of OpenAI. He talked about it in an interview on the Cheeky Pint podcast, as reported by PCMag on Friday, the 30th. The criticism from Sam Altman on local AI targets a trend that keeps gaining ground. More people are using artificial intelligence models directly on their computers, without going through the servers of big tech companies.
The comment reignites a debate that goes well beyond Silicon Valley. Choosing between cloud or local AI is already a real decision for many Brazilian companies that rely on these models day to day. And each path weighs differently once privacy, cost and scale come into the picture.
What Altman said
His argument centers on DRAM, the type of memory used both in regular computers and in the servers that run AI. According to PCMag, Altman argues that the supply of this component is limited. So it makes more sense to concentrate its use in data centers, like the ones behind the OpenAI cloud, where memory runs close to full capacity. He admits that many people have the opposite intuition.
The outlet also reports that Altman questions the idea that open source models running on personal computers solve the problem and make data centers unnecessary. In his view, spreading scarce memory across a bunch of machines that sit underused much of the time would be a poor way to distribute that resource around the world.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
It is an efficiency argument. It makes sense on paper. But it does not come from a neutral source.
Not even OpenAI partners agree
The OpenAI ecosystem itself is split. PCMag points out that partner companies such as Nvidia and Microsoft have been promoting the ability to run AI on your PC with new machines packed with memory. In other words, the OpenAI CEO turns his nose up at the idea while hardware makers and software developers bet on it as a selling point.
The outlet also lays out the advantages of local AI. The main one is privacy, since data never has to leave the machine. Another is that people and companies can use models without paying for cloud services. As PCMag sees it, that second point is a direct threat to the business model of OpenAI and other major providers.
Put simply, the report suggests there is commercial interest behind what Altman thinks. To be clear, that interpretation comes from the American outlet, not from a statement by the executive. Still, it is useful context that helps explain why the topic is far from settled.
Cloud or local AI: what matters for Brazilian companies
For anyone thinking about AI for Brazilian companies, the discussion leaves theory behind pretty fast. There is no single answer. There are criteria, and each business will weigh them differently.
- Privacy and sensitive data: running AI on a PC keeps information inside the company. That matters a lot for anyone who must comply with the LGPD, the Brazilian data protection law, and handles customer data, contracts or health information.
- Cost: cloud services usually charge per use, often in US dollars, which leaves the bill exposed to exchange rate swings. Local AI, on the other hand, requires an upfront investment in hardware with plenty of memory, plus someone to handle the setup.
- Scale and maintenance: providers like OpenAI usually offer bigger, more robust models, so the company does not have to worry about servers, updates or equipment failures.
Picture, just as an example, a small firm that reviews legal documents. It deals with confidential material and has a predictable workload, so keeping everything on its own machine might be the better call. Now think of an online store that serves lots of customers through chat, with spikes during big sales events. In that case, the flexibility of the cloud tends to make more sense. Capacity follows demand, and the store does not have to buy equipment.
Anyone following the OpenAI bet on bringing AI into businesses can see that the company wants to own exactly that second space. But relying on third-party infrastructure has its own risks, as the CoreWeave crisis and its impact on Brazilian startups shows.
What about marketing teams?
Marketing teams use AI to write copy, analyze campaigns and organize customer data. Most of these tasks work well in the cloud, with ready-made tools that are easy to use. Caution kicks in when the work involves contact lists or personal information. At that point, knowing where the data goes stops being a technical detail and becomes a matter of responsibility.
A path many teams take is the middle ground. Creative and generic tasks go to the cloud, and sensitive work stays in a controlled environment. It is not a rule, but it is usually a sensible starting point.
What to take from this debate
The comment from Altman reflects the position of someone who sells cloud capacity, and that needs to factor into how you read it. It does not mean the memory argument is false. It means it is one part of the story, told by someone with a direct stake in the outcome.
For entrepreneurs and tech professionals in Brazil, the choice still depends on the type of data, the budget and how much you use it. If you want to see how other businesses are organizing this routine, check out our piece on AI-native companies and their workflows.
