For the last decade, the commercial space industry has operated on a simple assumption: access to orbit would become cheaper, more frequent, and eventually abundant. SpaceX made that assumption credible. Falcon 9 became the default transportation layer for a generation of space companies. Now that model may be changing. SpaceX has reportedly stopped accepting new commercial Falcon 9 reservations. Existing commercial commitments and government missions will continue, but new customers are increasingly being told there is no availability. The obvious explanation is the transition to Starship. But something more important may be happening. SpaceX may simply have better things to do with its rockets than sell launches. That is already true with Starlink. Orbital compute could push that logic much further. SpaceX has said it may prioritize its own payloads over additional government or commercial customers to pursue orbital compute. If a launch can either generate a one time fee or deploy infrastructure that creates far more value through connectivity or compute, the customer is no longer just competing with another customer for capacity. They are competing with SpaceX itself. This matters for AI too. The bottleneck is becoming physical: chips, power, cooling, grid access, and data center capacity. If orbital compute becomes viable, SpaceX could gain access to a new source of compute while also owning the launch system needed to deploy it and, through xAI, one of the companies consuming it. Then consider Blue Origin. The obvious alternative is New Glenn. But Blue has proposed Project Sunrise, its own orbital compute constellation. Blue may eventually face the same choice: sell launch capacity to outside customers, or use it to build a higher value service itself. That creates a different market for space startups. There is a big difference between launch being expensive and launch being unavailable. If your business requires thousands of tons in orbit, launch is no longer a line item. It is a strategic dependency. This could also create an opportunity for independent providers such as Rocket Lab. For years, every launch startup faced the same question: how do you compete with SpaceX? The better question may become: Who serves the customers SpaceX no longer wants to serve? None of this requires orbital compute to work perfectly. Heat rejection is hard. Radiation degrades hardware. GPUs and networking equipment fail. Hardware refresh cycles are short. For those reasons, I suspect early orbital compute is more likely to focus on inference than frontier model training. But even the credible possibility changes the incentives. For years, we assumed cheaper launch would mainly benefit the companies buying launches. What if the biggest beneficiaries are the companies that own the rockets? And for everyone else: Who is actually going to launch you?