OpenAI's strategy chief says compute is not the real edge for AI companies — 3 other things are the key

In a recent episode of the "Auren Hoffman" podcast, Jason Kwon, OpenAI's chief strategy officer, shared insights on what truly drives success within the AI sector. While acknowledging the significance of computational power, Kwon argued that it is not the defining factor that distinguishes top-performing labs from their competitors. Kwon emphasized that equating compute solely with financial resources oversimplifies the complexities of the AI landscape. He stated, "If you just reduce compute to capital, it’s just money; then you buy the physical infrastructure. We don’t necessarily assume that being the most capitalized organization guarantees victory." According to Kwon, three critical elements play a more pivotal role: scarcity, bet selection, and organizational structure. He pointed out that scarcity can often spur innovation, compelling teams to make thoughtful decisions about how to allocate limited resources effectively. Additionally, the ability to choose the right research directions, know when to invest further, and when to pivot is essential for gaining a competitive edge. Kwon also highlighted the necessity of having an appropriate organizational framework to support these decisions. The right capacity and criteria for making selections are vital for sustaining impactful initiatives. At the macro level, Kwon acknowledged that computational resources remain crucial for the breadth and diversity of research experiments. OpenAI has consistently expressed its growing demand for computing resources. CEO Sam Altman recently mentioned on X that the company is exploring new features by utilizing substantial compute resources. Similarly, Kevin Weil, OpenAI's chief product officer, noted that every additional GPU they acquire is immediately put to use. In fact, Altman projected that by year-end, the company would integrate over 1 million GPUs into its operations. In comparison, Elon Musk's xAI revealed that it employs a massive supercluster of over 200,000 GPUs named Colossus to enhance its training processes. Other tech leaders, including Mark Zuckerberg, have also openly discussed their strategies to leverage compute power as a competitive advantage, with Meta reportedly outspending its competitors on GPU acquisition and the specialized infrastructure required to support them.