Speaking on the Cheeky Pint podcast, OpenAI CEO Sam Altman argued that the memory shortage is not correcting itself, saying “the invisible hand is failing right now.” He noted that DRAM makers who once spent $20 billion to earn $19 billion are about to generate roughly $100 billion of free cash flow a year each, yet remain too scarred by past busts to build aggressively. Altman estimates data-center costs at about $70 billion per gigawatt and says $20 billion would be a far healthier number. He also outlined how memory utilization of just 5% makes local AI inference a poor use of scarce supply, why OpenAI is vertically integrating chips, data centers and robots, and why he prefers “a trillion AIs of slightly superhuman intelligence” to one trillion-IQ model. He disclosed a ~$10 million-per-proof cost for OpenAI’s Navier-Stokes math work, the widening productivity gap between elite AI users and everyone else, and his view that personal agents are “a small number of ideas away” from a mainstream product. For investors, he said memory capex, energy and data-center margins are the bottlenecks to watch.