Sam Altman Argues Centralized RAM in Data Centers Beats Distributed Local Memory

OpenAI chief executive Sam Altman recently told a podcast audience that the current scramble for DRAM makes a strong case for keeping memory in large‑scale data centers rather than scattering it across personal computers.

He explained that when every user holds a sizable chunk of the world’s limited RAM but only uses a fraction of it, overall utilization stays low. By pooling memory in the cloud, operators can run those chips near‑continuously, extracting far more value from each wafer.

The buying power of projects such as OpenAI’s Stargate has already locked up multi‑year supply agreements with Samsung and SK Hynix, leaving far less capacity for laptops, phones and workstations. That dynamic is a key driver behind the recent price spikes on consumer hardware.

Why it matters for GPU and AI infrastructure

High‑performance AI workloads depend on massive, low‑latency memory pools that sit next to GPU clusters. Centralizing DRAM in purpose‑built facilities lets cloud providers like AiGpu deliver the bandwidth and capacity that model training and inference demand, while avoiding the inefficiencies of under‑utilized local memory.