OpenAI president Greg Brockman explains how the company allocates GPUs internally: 'pain and suffering'

In a revealing discussion on the "Matthew Berman" podcast, OpenAI's president, Greg Brockman, shared the challenges of distributing graphic processing units (GPUs) within the organization, describing the process as one filled with "pain and suffering." Managing these essential resources has proven to be both emotionally taxing and complex, as Brockman explained how teams pitch innovative projects in hopes of securing GPU allocations. At OpenAI, the division of computing power is carefully structured between research and practical applications. While the company's chief scientist oversees the allocation for research projects, senior leadership, including CEO Sam Altman and applications leader Fidji Simo, determines the overall resource distribution. A dedicated internal team, including Kevin Park, plays a crucial role in reallocating GPUs as projects conclude, responding to urgent requests for additional hardware. Brockman emphasized the critical nature of GPUs in driving team productivity, noting the palpable tension surrounding resource availability. "People really care," he remarked, indicating that the emotional stakes involved in securing compute resources cannot be understated. The growing demand for GPUs has been a consistent theme at OpenAI. Chief product officer Kevin Weil stated on the "Moonshot" podcast that every new GPU addition is quickly utilized, underscoring the insatiable appetite for computational power in AI development. He likened the surge in GPU usage to the explosion of video content made possible through increased bandwidth. Recently, Altman announced the introduction of new compute-intensive offerings, with some features initially available only to Pro subscribers due to associated costs. This initiative aims to push the boundaries of AI infrastructure, exploring what is achievable with significant computational resources directed at innovative concepts. OpenAI is not alone in its quest for GPU dominance. Tech leaders like Meta's Mark Zuckerberg have acknowledged the competitive edge that high compute availability provides, with Meta reportedly outspending competitors on GPU resources and the infrastructure required to support them.