OpenAI CEO Sam Altman has argued that it is “quite terrible” for consumers to keep the world’s scarce DRAM in their personal computers, saying AI workloads should instead run in data centres.
Sam Altman · AI compute policy
Altman calls keeping scarce DRAM in home PCs "quite terrible"
The OpenAI CEO argues AI workloads should run in data centres, saying personal computers use their memory only about 5% of the time. His framing treats DRAM as a constrained resource better allocated centrally.
5%
Share of time a home PC uses its memory, per Altman
95%
Share of time that same memory sits idle
Memory utilization on a home PC (illustrative scale)
Each block = 5% of memory time. 20 blocks = 100%.
Altman's point: memory held on a personal device is mostly unused capacity, which he sees as an inefficient use of a scarce input.
Case for centralized inference
Providers can pool memory and accelerators and keep them busy.
Scarce DRAM is spread across many idle devices instead of shared capacity.
Fits the industry's heavy investment in data centre capacity.
Questions the statement leaves open
The 5% figure is a leadership position; its measurement method is not disclosed.
Utilization ignores privacy, latency and offline use, the main reasons local AI is promoted.
Shifts in where models run would affect DRAM pricing, device design and consumer hardware economics.
The open trade-off
Whether AI moves to centralized inference or keeps pushing capable models onto phones and laptops will depend on utilization economics weighed against privacy, latency and memory pricing.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...