Altman argues for centralized memory use

Sam Altman, CEO of OpenAI, told John Collison on the Cheeky Pint podcast that RAM is better used in data centers than in personal computers. He claims that personal computers use only about 5 percent of their memory at any given time, suggesting that centralizing memory could improve global utilization.

Critics highlight local computing benefits

Critics of Altman’s view argue that RAM is used for more than just AI, such as video editing and gaming. They note that running AI locally offers benefits like data privacy and offline access, which may reflect real user needs rather than inefficient resource use.

OpenAI’s role in memory demand

Altman’s stance has drawn attention because OpenAI itself contributes to the global demand for memory through its partnerships with manufacturers. This creates a paradox: individuals are affected by this demand, yet Altman suggests that memory would be better used elsewhere.

OpenAI’s ambitions in AI infrastructure

In addition to advocating for centralized computing, Altman outlines OpenAI’s goal to control the full AI industrial chain. This includes developing its own chips and using AI to build infrastructure, from solar panels to data centers, in a self-sustaining industrial loop.

Cost reduction goals for computing power

Altman aims to reduce the cost of a gigawatt of computing power from about 70 billion dollars over five years to 20 billion. This highlights the importance of infrastructure and energy in AI development, alongside new methods to improve computing efficiency.

OpenAI’s investment in computational research

OpenAI is now spending amounts previously considered excessive on computational research, such as 10 million dollars on mathematical work related to Navier-Stokes equations. This reflects the growing scale of on-demand computing and the shift from limited to extensive resource use.