The single biggest threat to Nvidia, Amazon and Azure right now is a rocket company: SPACEX
SpaceX generated $18.67 billion in revenue in 2025. At $184 per share today, it is valued at $2.41 Trillion, 129 times trailing revenue.
Nvidia trades at 22 times.
Amazon at 4 times. The valuation only makes sense when you understand what SpaceX is actually building.
SpaceX declared a total addressable market of $28.5 trillion in its IPO filing, equal to the entire annual GDP of the United States.
$26.5 trillion of that is AI alone. At just 1% capture of its own stated market, SpaceX generates $285 billion in annual revenue.
Starlink is the proven starting point.
The constellation has crossed 10,000 satellites, holds 60% of the LEO broadband market, and has 10 million monthly active users on Direct-to-Cell. Goldman Sachs projects Starlink reaches $144 billion in revenue by 2030.
At a standard 10x revenue multiple for subscription infrastructure, Starlink as a standalone business is worth $1.4 trillion, more than SpaceX's entire market cap today, before counting anything else.
The bigger bet is orbital AI infrastructure and this is where SpaceX directly threatens Nvidia, AWS, and Azure simultaneously.
AI data centers have a problem that is getting worse. A 1-gigawatt facility now draws more electricity than some mid sized nations. Local governments are rejecting new projects because of grid strain and water use.
AWS and Azure pay $0.05 or more per kilowatt-hour for electricity. That cost is rising as AI demand grows. SpaceX's answer is to move the compute off the grid entirely.
In orbit, solar panels generate more than 5 times the energy of an identical array on Earth because there is no atmosphere blocking sunlight. Capacity factor exceeds 95% versus 24% for terrestrial solar.
No cooling is required, the vacuum of space handles it passively. No land, No grid, No permits, No energy shortages.
An independently verified 10-year cost model for a 40-megawatt cluster: $8.2 million in orbit versus $167 million on the ground.
Effective energy cost in orbit: $0.001 to $0.005 per kilowatt hour, 10 to 50 times cheaper than what AWS and Azure pay today.
SpaceX unveiled the AI1 satellite a few days ago, its first orbital data center prototype with 150-kilowatt peak compute, a 70-meter wingspan, and Nvidia GPUs confirmed by the CFO.
The company has filed with the FCC to deploy up to 1 million AI satellites. Goldman Sachs projects this division alone generates $322 billion in revenue by 2030, a 100-fold increase from $3.2 billion today.
Nvidia generated $115 billion in data center revenue last year at 50 to 80% gross margins. AWS and Azure each generate approximately $120 billion in annualized cloud revenue.
All three are energy buyers locked into grid infrastructure. SpaceX's orbital compute becomes an energy generator with near-zero marginal operating cost. That is not a competitive advantage, it is a different physics regime.
SpaceX is also moving to eliminate Nvidia's chip pricing entirely.
Terafab is a joint semiconductor fabrication venture between SpaceX, Tesla, xAI, and Intel near Austin, Texas.
Total investment: $119 billion.
Target output: 1 terawatt per year of AI compute capacity.
Intel brings its 14A process node, one of only three sub-5nm processes in commercial production globally. The facility will produce D3 chips, radiation hardened and optimized for orbital data centers, alongside AI chips for terrestrial use.
Every dollar Terafab saves on chip costs is a dollar Nvidia loses.
The xAI merger on February 2, 2026 completed the vertical integration. At a combined valuation of $1.25 trillion, SpaceX now owns the AI models (Grok), the training infrastructure (Colossus 1 in Memphis), the orbital deployment system (Starship), the connectivity layer (Starlink), and the chip manufacturing (Terafab).
Anthropic is already paying SpaceX to rent unused Colossus 1 compute, a competitor writing checks to SpaceX for access to its own infrastructure. This is the exact model Amazon used when it built AWS from its own retail logistics and then sold access to the world.
The entire orbital data center thesis depends on Starship. At current launch prices of $900 per kilogram, a 1-gigawatt orbital data center costs $42 billion, three times the ground-based equivalent. At $200 per kilogram, the math inverts completely.
SpaceX has completed two consecutive Super Heavy booster catches. Full ship-side reusability has not yet been demonstrated. Goldman's $474 billion 2030 revenue projection requires growing from $18.7 billion at a rate faster than any company at this scale has ever managed.
SpaceX itself acknowledged in its IPO filing that orbital AI infrastructure "relies on unproven technologies and may not become commercially viable."
But if even half of this works, $2.35 trillion looks cheap before 2030.