OpenAI halved inference costs — so why aren't GPUs getting cheaper
openai cut inference costs for guest chatgpt users by over 50%, reducing gpu needs to a few hundred. this efficiency could boost margins or scale without easing chip demand.
openai engineers informed colleagues that they cut inference costs by more than half for non-logged-in chatgpt users, significantly reducing gpu requirements to a few hundred. this move targets the product's most limited tier, leaving questions about scalability to the full platform. the focus on inference optimization mirrors industry trends like deepseek's new open-source method, which accelerates requests by up to 85 percent. these efficiency gains allow labs to serve more users and improve response times without immediate hardware expansion, sustaining high demand for chips despite constrained data center growth.