No One Realizes What Elon Musk Just Created: Why a SpaceX Reverse Merger Unlocks Orbital AI

The most powerful company in human history is about to be born. A few days ago this was casual speculation on a podcast. Now Bloomberg has confirmed it, the legal entities exist, and the banks are already in the room. Most people are still reading this as a money grab. That is the wrong frame. This is about who owns the infrastructure layer of the AI age.

On the All-In podcast, Chamath Palihapitiya floated a contrarian take: SpaceX does not IPO. It reverse-merges into Tesla. Elon uses the moment to put his two seminal assets on one cap table. Jason Calacanis jumped in — you could put Neuralink and The Boring Company in there too. Chamath stuck to the core: no clean SpaceX IPO. A reverse merger. SpaceX, last marked around $800 billion on a December tender near $420 a share, and Tesla, roughly $1.4 to $1.5 trillion, becoming one public company. That would be the biggest blockbuster merger in the history of capitalism.

Then the rumor stopped being a rumor. Bloomberg reported SpaceX is formally considering combining with Tesla, xAI, or both ahead of what could be a $1.5 trillion listing. Nevada entities were created on January 21. One lists SpaceX and CFO Bret as managing members. Bank of America, Goldman Sachs, JPMorgan, and Morgan Stanley are in the mix. This is moving faster than the surface narrative admits.

I have tracked Elon and these companies since 2012. The headline — Musk companies might merge — misses the point. The shallow read is the richest person on Earth consolidating toys. The real read is control of the stack that trains and runs AI at scale. Whoever owns that stack owns the decade.

Start with the numbers. SpaceX at about $800 billion. xAI just closed a $20 billion Series E at a $230 billion valuation — up from $50 billion a year earlier, roughly 4x in twelve months. Tesla near $1.5 trillion. Add those up and you are already flirting with $3 trillion. The merged entity would be worth more than the sum because of what the pieces do together.

SpaceX built reusable Falcon 9 and cut the cost of getting to orbit by as much as 80% in some cases. Starlink is the cash cow — over 9,400 satellites, billions a year in revenue, funding Starship, which is the cargo capacity jump that makes the rest of this story possible. Tesla builds cars and batteries and is pivoting hard into Optimus humanoid robots. Full Self-Driving already does coast-to-coast, parking-spot-to-parking-spot drives with cameras and an onboard AI computer — rain, snow, night, ice. That same stack is the brain path for Optimus. xAI builds Grok, burns through compute at a brutal clip, and is standing up the supercomputers that would feed models into X, Tesla vehicles, and robots.

Each company is impressive alone. They share one bottleneck: compute. Inference chips — the chips that answer your question after a model is trained — are about to be the scarce resource for the next ten to twenty years. Training clusters matter. Serving the public at scale matters more. Some estimates say we need on the order of a thousand times more inference silicon in the next five years than we have today. Without chips there is no intelligence. Without intelligence there is no AI. Without AI, the robots, the robotaxis, the assistants — the whole timeline — collapses.

The race for AGI is not mainly about who has the cleverest algorithm. Models are starting to commoditize. Open-source is catching up. Synthetic data narrows data moats. Papers diffuse. What does not commoditize is physical infrastructure: rockets, satellites, power, manufacturing at millions of units a year, global distribution. Whoever has the most compute at the lowest cost with the best energy economics wins. That is the actual competition.

At Davos, Elon said something that got buried under tariff noise: the lowest-cost place to put AI will be in space, and that will be true within two to three years. Most people heard hyperbole. He meant physics.

A terrestrial data center needs power, cooling, land, cheap electricity, and a regulator who will let you build. Every watt of compute becomes heat. Removing that heat costs more energy. The biggest training clusters — the 100,000-GPU monsters — need their own power plants. Microsoft is reviving Three Mile Island because the grid cannot feed the load. In space the economics flip.

Starship is coming online with roughly 10 to 100 times Falcon 9's cargo to orbit. The plan is not metaphor: put AI computers in orbit, collect solar power, serve Earth. Sun-synchronous orbit can keep one face on the sun and the other radiating into cold space. About 1,400 watts per square meter, 24/7, no clouds, no weather. Heat dumps into a 3 Kelvin background with passive radiators — no chillers, no water, no cooling energy bill. No NIMBY fights. No utility throttling your draw. No endless permitting. Free energy and free cooling for the life of the satellite.

Starlink V3 birds are targeted for the first half of this year with about 1 terabit per second capacity — roughly 10x the current generation. They are being designed with edge compute, modular GPU housings, better thermal control, and larger solar arrays. SpaceX is already testing separation hardware on Starship that can loft about 60 V3 satellites per mission. At a cadence of 100 Starship launches a year, that is 6,000 compute-capable satellites a year. In a few years you can have tens of thousands of AI-enabled satellites as a distributed orbital computer.

Connect the dots. SpaceX has the only reusable, reliable, affordable heavy launch system at this scale. SpaceX has the constellation. xAI has the models that need the compute. Tesla has the manufacturing machine — about 2 million cars a year — plus batteries and robots. Put them under one roof and you do not just compete in the AI race. You own the track the race runs on. Photons on solar arrays in orbit all the way down to the model in your car or home robot.

Which deal lands first? Polymarket recently had SpaceX–xAI around 48% and SpaceX–Tesla around 18%. The xAI path looks cleaner on paper: both private, no public shareholder vote theater, less SEC choreography, and no Tesla Shanghai exposure. Tesla builds roughly a quarter of its cars in China — on the order of 400,000 to 500,000 a year. Pair that with defense-critical SpaceX assets and CFIUS shows up. That risk is real.

I still think people are underweighting Tesla. Chamath called a SpaceX–Tesla combination the Berkshire Hathaway of this century. Buffett used insurance float to fund operating businesses under one capital allocator. Here the pieces are different: rockets and satellites for capability and cash, cars and robots for cash flow and manufacturing, AI for the intelligence layer that makes everything else more valuable, X for distribution and data. One decision maker allocating across the set. The whole compounds in ways the parts never could alone.

The reverse-merger mechanism matters. Instead of a classic SpaceX IPO — roadshow, book building, underwriting fees, lockups — SpaceX merges into already-public Tesla. Instant liquidity. Tesla shareholders wake up owning SpaceX too. The "Elon is distracted across too many companies" analyst line dies overnight: one company, one cap table, one mission.

The cross-investments already point this way. SpaceX put $2 billion into xAI. Tesla put $2 billion into xAI in early 2026 after the board approved it despite a prior shareholder vote that fell short on abstentions. In parallel, Model S and Model X production ends after Q2 — Musk called it an honorable discharge on the Q4 2025 earnings call — and those Fremont lines convert toward at least a million Optimus robots a year. Capex is doubling past $20 billion in 2026. The old car company is dying so the AI-and-robotics company can live.

Bull case, stated clean: own launch, own the constellation, own the models, own the factory, own the energy storage, own the fleet that collects real-world data every day. Amazon has pieces. Google has pieces. China has pieces. Nobody has all of it. Replicating this means twenty years of rockets, a five-year Starlink head start at minimum, an AI lab burning billions, fifteen years of Tesla-scale manufacturing, and a global distribution network. That moat is not a slogan.

Bear case, also real: orbital data centers are unproven at commercial scale. Radiation kills electronics. Debris is a collision risk. Servicing is hard. Elon time of two to three years often means four to six. SEC, CFIUS, and antitrust will all have opinions. Putting this much vertical power in one company — really one person — makes people uncomfortable for good reasons. What if something happens to Elon? What if Beijing squeezes Shanghai? Do not dismiss those.

I keep landing on the same place. The AI race is an infrastructure race. Models will diffuse. Physical stack will not. June 2026 IPO talk, reverse merger, or some hybrid — the exact path matters less than the destination. One entity from orbit to your driveway. Launch your own satellites, train and run models in space, push that intelligence into millions of vehicles and robots, and feed the data back to improve the models. A flywheel that feeds itself.

Reusable rockets landing on drone ships sounded absurd ten years ago. So did EVs at scale and unsupervised robotaxis. This track record is always later than promised and more complete than the bears expect. Nobody realizes what just got assembled. Now you do.

Check the video here.