SpaceX (SPCX): mainly an AI computing company
SpaceX just completed a successful launch, reaching low-Earth orbit and placing several satellites. Looking at revenue, whether the company's own target of $100 billion a year or an open source model estimate of $56 billion in 2026, about a third goes to AI computing. That is the actual compute, separate from Cursor and Grok. Starlink takes about a quarter of revenue, and launches are just under 20%. So today SpaceX is primarily an AI computing business.
The core resource SpaceX sells is engineering talent. They hold the "Navy SEALs" of engineering and can point that talent at any task - AI compute, missile launches, or communications.
Less than a year ago there was heavy talk about whether the xAI acquisition was a rescue from bankruptcy. Today, paired with the strong impact of Cursor, that deal could become one of the best acquisitions in tech history, next to Instagram and WhatsApp.
A new company type: AI and space conglomerates
A new class of company is forming inside the G7 - conglomerates that combine artificial intelligence and space. SpaceX shows the model: its computing and communications business funds space exploration and other research and development. Next this grows into orbital computing, linked through Starlink, delivering superintelligence to any point on Earth.
SpaceX is the only company that owns the full stack of this new superintelligence-and-space conglomerate: rockets, communications, and computing infrastructure. Other smart firms are moving in. Google (GOOGL) just announced solutions built on SpaceX. Amazon (AMZN) has ground-based computing power, a space program, and the ability to join them. These firms see where things are going: data centers move off the planet and connect to the world through satellites.
How fast orbital computing matters
SpaceX has now placed half a gigawatt of power and several terabytes per second of bandwidth into orbit. Even so, since SpaceX is a bit late in making the impossible real, orbital computing is not expected to add much to profit before 2029.
Ground-based computing stays the workhorse for the next few years. SpaceX plans to build about 20 gigawatts of capacity over the next 24 months. Jensen Huang said today that each gigawatt of terrestrial computing can create up to $50 billion in value per year.
Approaching the 2030s, the "not in my backyard" problem for data centers, plus the advantages of solar energy and its high use rate in space, make it clear that computing power is moving off the planet. That shift will make the 2030s exciting, along with the superintelligence that compute will serve.
Anthropic vs OpenAI: diverging paths to market
The last 48 hours brought news from both companies. A leak of the Anthropic prospectus caused heavy market discussion. Sam Altman of OpenAI said the industry needs to slow down and should not release models until security is proper.
The two companies are on very different paths. Anthropic filed its S-1 and plans to go public this year, at a $2 trillion estimate backed by current losses of $42 billion. Buying at that level needs a very price-insensitive investor. OpenAI looks likely to wait until next year, probably because it has not had the strong quarters Anthropic posted in its last two.
It was surprising that Anthropic did not follow the leak of some 2025 financial figures with stronger 2026 figures. The main point: both companies seem to be rushing to market as fast as possible. The likely reason is margin erosion.
The shift to open source AI
Thirteen weeks ago, computing power split 80% toward frontier models and 20% toward open source, according to David Friedberg. Over those 13 weeks the split flipped to 20/80. That is a huge market move toward local, open source solutions.
This shows up in real company behavior, including a move to local open source models. Many managers and colleagues report that their spending on Claude in the spring is 5 to 8 times higher than what it will be in the fall. If demand is shifting that sharply from frontier models, the erosion is real, which is what pushes Anthropic to rush to market.
Why frontier models can still pay off
Frontier models, sometimes called "the sport of kings," are the greatest prize in the history of capitalism. Current gross profit margins for advanced models are expected between 70% and 80%, so there is still much room to grow. A range of highly specialized use cases will always want cutting-edge technology and pay a large premium for it.
The investor base here is not sensitive to interest rates and does not focus much on price. They are betting on the payoff of giving the world superintelligence in the future.
