my personal opinion on this matter rant: spacexai got a shit ton of compute because elon wanted to compete on ai with open ai. and running amazing physical supply chains has always been elon's moat. no one can build out data centers as fast as he did with colossus. tesla was in this market anyways since they had to train fsd for their cars. xai is currently miserably failing to compete on the frontier of ai. despite having the compute their models failed to reach a sota level on a time frame that made sense to stay competitive, they brought cursor to fix this problem which seems to work well. countless of other great people joined spacexai to work on this, and new great people are coming in though. its the same issue meta had for a while with their ai department and still has, training ai models is hard and not a compute issue. especially at the frontier. if its not your main goal its hard to get it right. hence why anthropic is winning at this. its the only company that is hyper focused on ai out of all that compete. if you have compute and can't turn it into a profit with ai usage you have to rent it out to others. they rented it out mostly to anthropic, hence why anthropic rate limits haven't been a thing in a while and grok bot is currently using opus 5.5 as well. getting financing for gpu clusters is only going to get more expensive and risky because: a) the new line of upcoming gpu clusters will be more expensive then previous ones. with more expensive materials and research costs. ( constant innovation in this space has physical limitations ) b) due to shortage's across the entire supply chain and rising demands the timelines of building out chips, factories and grids will take longer, which means those who finance's it all take way more risk over long time periods. ( what his post came down to ) which they don't want to risk so they converge for companies that can finance a down payment with limited risks. ( space xai is one of the few that can ) i've always said it but its obvious space xai will become the aws cloud inference of the ai timeline. elon is good at running physical supply chain driven companies. the moat isn't going to be in the models/ software anyways in a few years, and he always knew that. little to no people will need sota ai model intelligence from one singular provider in a few years anyways. since 90% of people don't actually need super intelligence for their day to day lives / don't know what to do with it. people need ai to be affordable and accessible though, which is solely an inference, infra and compute problem once ai models are no longer the primarily moat ( and we’re converging toward commoditization right now ) neoclouds will most likely die, but their compute will not. it will just be bought up cheaply by companies like spacexai. and then we're back to the beginning of the end where every small player is being outcompeted out of every market we know in tech. very similar patterns where played out during internet software era we saw in the last 2 decades. look at the companies that survived that and what they do and sell right now. google makes around 45% of their revenue renting out tpu's, despite by most being known as an internet company.