Tech • AI • Robotics
OK, I Admit It: I Underestimated SpaceXAI (by a LOT)
9.4/10TL;DR
Gavin Baker argues that xAI tools such as Grokbot could rapidly gain share in AI agents, while SpaceX may also be positioned to reshape compute economics through Starship, orbital data centers, and the expansion of Starlink.
KEY POINTS
Grokbot seen as a breakthrough in AI agents
Baker described Grokbot as a major step change in usability, saying tasks that previously took hours with coding assistants could now be done in 7 to 12 seconds. Examples included podcast, Substack, and X summarizers, along with sentiment trackers for topics and stocks. The core claim is that the product lowers the technical barrier for automation while delivering stronger results at much higher speed.
xAI’s progress has shifted investor expectations
The discussion framed xAI as a late entrant that has moved quickly from catching up in large language models to competing aggressively in AI agents. Baker’s view is that many observers underestimated what the company could do at the model layer, and that products built on top of Grok may end up taking a meaningful slice of the agent market if current momentum continues.
Automation is already close to end-to-end workflows
The capabilities described go beyond simple summaries. AI agents were portrayed as increasingly able to gather news, draft scripts, clone voice output, edit basic media, upload files, generate titles, and schedule publication with little or no human input. That suggests direct pressure on digital workflows and content operations that until recently still required manual coordination.
Orbital compute depends on Starship reusability
Baker argued that the critical question for space-based data centers is not demand for compute but whether Starship can be reused rapidly and reliably. He cited a rough cost of $50 billion to bring 1 gigawatt of AI compute online on Earth, with about $15 billion tied to power, cooling, labor, and related infrastructure that could be reduced in orbit through solar power and radiators.
Space economics could flip if launch costs fall below $1 billion
On Baker’s math, putting 1 gigawatt of compute in orbit could cost around $35 billion before launch, versus $50 billion on Earth. If fully reusable Starship cuts launch costs to under $1 billion, orbital compute becomes economically compelling, especially as terrestrial costs rise with shortages in labor, power equipment, and materials such as copper. He said training workloads would still favor Earth in many cases because latency and chip proximity matter, but an increasing share of global compute could move to orbit.
A launch window for orbital AI racks is approaching
Baker pointed to plans discussed by Elon Musk and Jensen Huang around a Rubin rack intended for launch in the fourth quarter of 2027, while allowing that timelines could slip into 2028. Even on that delayed schedule, he suggested the industry is not paying enough attention to how quickly orbital compute could become a live commercial category.
Starlink broadens the addressable market
Beyond AI, Baker emphasized Starlink Mobile as another large opportunity. He put wireless revenue potential at roughly $800 billion to $900 billion, and suggested that, combined with broadband, the total addressable market approaches $2 trillion. That would add another major business line even without assuming success in orbital compute.
Bundling across AI, connectivity and advertising is a strategic option
The conversation also highlighted the possibility of combining Starlink, Grokbot, and X advertising into bundled offerings. That model would echo how other tech groups used one product line to reinforce another, potentially using advertising demand and AI services to accelerate uptake across the broader ecosystem.
Infrastructure may matter even if first-party AI applications stumble
A key investment argument was that SpaceX and xAI may still benefit even if they do not dominate end-user AI applications. If the company builds excess compute for its own inference and training needs, it could lease scarce capacity to other firms. In that scenario, the infrastructure business itself becomes the backstop, especially if rivals lack access to reusable heavy-lift launch.
Launch cadence remains central to the long-term thesis
Baker said the real advantage is not merely rocket reusability but operational reusability at scale: catch, restack, refuel, and relaunch quickly. He pointed to infrastructure being built for potentially thousands of launches a year, with conservative assumptions around two launches per day per pad already implying a dramatic increase in capacity. That kind of cadence would be difficult for competitors to replicate and is central to the broader vision for communications and compute in orbit.
CONCLUSION
The debate is no longer only about rockets or satellites. It is increasingly about whether SpaceX and xAI can turn launch scale, AI software, and communications infrastructure into a tightly linked platform with influence across several multitrillion-dollar markets.
Explain this