OpenAI CEO Sam Altman is really gunning for villain-of-the-year status with PC gamers in particular, but owners of pretty much any computing device everywhere.<\/p>\n
Altman appeared on Stripe cofounder John Collison\u2019s Cheeky Pint podcast this week, sipping beers and weighing in on the state of AI. Gizmodo covered and analyzed the bulk of Altman\u2019s remarks here<\/a>, though another moment from the podcast is worthy of a shout-out. Altman thinks it\u2019s \u201cquite terrible\u201d that individual device owners and\/or operators of local AIs have access to DRAM, which is already catastrophically expensive due to massive demand from AI data centers like OpenAI\u2019s (and may be nowhere close to peaking<\/a>).<\/p>\n You see, it\u2019s inefficient not to just let OpenAI handle it. \u201cA lot of people want local inference. And somehow, they assume that the problem is solved now because we are going to have open-source models, we\u2019re going to be able to run them on our computers, and so we\u2019re not going to need all the stuff in the data center,\u201d Altman said.<\/p>\n \u201cBut from a global allocation perspective, actually, it\u2019s quite terrible to have everybody have a bunch of the world\u2019s limited supply of DRAM sitting in their own computer and being used 5% of the time,\u201d he added. \u201cYou want maximum utilization in the cloud if you\u2019re trying to deal with this very supply-constrained environment. And somehow many people\u2019s intuition is the opposite of that.\u201d<\/p>\n There is an unobjectionable point buried somewhere in there. Perhaps a future utopian society could provide universal access to centralized compute. But there\u2019s a more obnoxious point that doesn\u2019t need excavating: Altman isn\u2019t describing the optimal distribution of resources so much as the distribution of resources optimal to him. That arrangement would entail DRAM manufacturers devote even more capacity to fueling OpenAI\u2019s incomprehensibly expensive infrastructure rollout<\/a> ($750 billion through 2030!) and consumers delegate ever-increasing control of their work and lives to AI assistants. The cost of that rollout is directly driving up prices of DRAM and other components through the roof, even for products like smartphones<\/a>. Closed-source data centers are OpenAI\u2019s bread and butter, as is platformization and consolidation.<\/p>\n Presumably the dream is that any time someone needs to write a document, call someone, stream media, play a video game, or do pretty much anything else that requires computation, they pay an OpenAI cloud agent based in a data center to do it. But hey, they won\u2019t need to worry about paying for RAM, right?<\/p>\n