You have two jobs when you're running an enterprise foundation model company. One, build amazing models, and two, buy enough compute to meet all the demand you get. Both jobs are hard, but hard in very different ways. Last quarter each of the top two FM companies got one job right. OpenAI nailed compute but until 5.5 they whiffed on the model. Anthropic nailed the model but came up light on compute. It’s pretty obvious why building a model is hard. It’s worth taking a minute to look at how hard capex planning must be. Three factors come together. First, these companies are capex-intensive, every dollar of revenue needs 3x plus in capex. Second, capex has at least a one year time lag, and third, the companies are growing at a 10x-plus rate. Put all that together, and it means that a company doing 1Bn in ARR now, looking out one year, has to be thinking about 10x growth times 3x capex, which means committing to capex that is thirty times the current revenue run rate. And you have to do that every year. It makes running an airline look easy.