I Read SpaceX's FCC Application for Orbital Data Centers

This is the third piece in my SpaceX series:
SpaceX Acquires xAI: An Outsider’s Notes — The merger and IPO logic
In 36 Months, Space Will Be the Cheapest Place for AI — Musk interview analysis

Why Read the Source

An investor friend messaged me: “Have you read SpaceX’s FCC filing?”

I hadn’t. All I’d seen were Twitter fragments — “1 million satellites,” “space data centers,” “Kardashev II civilization.” Every tweet read like clickbait: shocking but unsatisfying.

So I went to the FCC website and pulled the original documents.

This is an application filed January 30, 2026, file number SAT-LOA-20260108-00016. The FCC Space Bureau accepted it for filing on February 4, opening public comments until March 6. The filing package includes the Narrative, Technical Attachment, Waiver Requests, Ownership statement, and Schedule S.

The difference between reading source documents and Twitter threads: Twitter tells you “SpaceX wants to launch 1 million satellites.” The filing tells you they simultaneously requested exemption from deployment timelines and surety bonds — meaning SpaceX itself doesn’t know when it’ll be done.

What This Filing Is

The FCC (Federal Communications Commission) regulates spectrum. Any satellite system using radio frequencies needs FCC approval. SpaceX’s application requests:

Authorization to launch and operate a new NGSO (non-geostationary orbit) satellite system
Permission to use specific frequency bands for communications
Waivers from several standard regulatory requirements

This isn’t a launch license (that’s FAA), nor a spectrum auction. It’s a document proving to the FCC: “My system won’t interfere with others and serves the public interest.”

Accepted for filing ≠ granted. The FCC is saying “OK, we’ll review it.” Next come public comments, oppositions, replies, then a decision.

Key Numbers

Parameter	Value
Satellite count	Up to 1,000,000
Altitude range	500 km – 2,000 km
Orbital inclinations	30° and Sun-Synchronous Orbit (SSO)
Shell thickness	Up to 50 km per shell
Communications	Optical inter-satellite links (petabit-class laser mesh)
RF bands	18.3-19.3 GHz (downlink) / 28.6-29.1 GHz (uplink)
RF purpose	TT&C backup only, non-interference, unprotected basis
Starlink integration	Optical links interconnect with Starlink Gen1/Gen2

The first detail that surprised me: RF is backup only.

The entire system relies almost exclusively on optical (laser) links. Ka-band RF is just backup — for telemetry, tracking, and command during critical early mission and emergency phases. And it’s on a “non-interference, unprotected basis,” meaning if there’s a conflict with other users, SpaceX yields.

This is shrewd. Spectrum is the most sensitive issue in FCC review. SpaceX minimized its RF footprint, leaving nearly all communications to optical links that don’t require FCC spectrum coordination.

Technical Architecture: What Twitter Missed

Cooling, Not Energy, Is the Hidden Advantage

Media coverage focused heavily on space solar power — “near-unlimited free energy.” But a paragraph in the filing, largely ignored, highlights an equally important advantage: thermal management.

Terrestrial data centers consume billions of gallons of water annually for cooling. In space, heat radiates passively into vacuum — free, infinite heat sink. From the filing:

“Radiative cooling enables passive heat dissipation into the vacuum of space, outpacing ground-based data centers that face escalating energy demands and infrastructure delays.”

Both major cost drivers for terrestrial data centers — power and cooling — are simultaneously addressed in orbit. However, radiative cooling has its own engineering challenges: effective dissipation requires large radiator surfaces, adding satellite mass and launch costs.

Each Satellite Runs Only a Few Kilowatts

Based on analyst estimates, each orbital data center satellite carries just a few kilowatts of IT load, with ~10-20 kW solar generation and 1-2 tons mass.

This is nothing like the “giant GPU cluster in space” people imagine. SpaceX’s strategy is massive parallelism: 1 million satellites × a few kW each = gigawatt-scale distributed processing.

The filing includes a strikingly ambitious figure:

“Launching 1 million tonnes per year of satellites generating 100 kW of compute power per tonne would add 100 gigawatts of AI compute capacity annually.”

One million tonnes per year. With Starship’s ~150-tonne orbital capacity, that’s roughly 6,700 launches/year — about 18 per day. SpaceX launched approximately 130 times in all of 2025. That’s a two-order-of-magnitude gap.

The Starlink Connection

The filing explicitly states optical links “may connect among or between those satellites and satellites in SpaceX’s first- and second-generation Starlink systems.” The orbital data center network isn’t standalone — it grafts onto Starlink’s petabit-class laser mesh, using Starlink satellites to relay data to ground stations.

This explains why the xAI merger was a prerequisite. Without Starlink’s ground relay network, orbital data centers can’t efficiently return data to Earth. Conversely, xAI’s AI models need orbital compute to break through the terrestrial power ceiling. The three components (SpaceX launch capacity + Starlink network + xAI models) form a closed loop.

Four Waiver Requests: The Signal in the Fine Print

This is what I consider the most underappreciated part of the filing.

1. Processing Round Waiver

Normally, NGSO applications go through FCC “processing rounds” for simultaneous evaluation of competing applications. SpaceX requests individual review outside a processing round.

In plain English: Don’t make me wait in line.

2. Milestone Waiver

The FCC typically requires NGSO operators to deploy 50% of satellites within 6 years and 100% within 9 years. SpaceX requests exemption.

In plain English: I don’t know when I can build 1 million satellites, so don’t give me a deadline.

This is the waiver worth watching. It means SpaceX is applying for a million-satellite system while simultaneously acknowledging it can’t commit to a deployment timeline. Tim Farrar of TMF Associates called the filing “quite rushed” and suggested it’s more of a narrative tool for SpaceX’s upcoming IPO than a near-term execution plan.

3. Surety Bond Waiver

The FCC typically requires satellite operators to post financial bonds guaranteeing deployment commitments. SpaceX requests exemption.

4. Schedule S Information Simplification

Schedule S is the detailed technical form required by the FCC. SpaceX requests simplified submission for channel plans, beam configurations, and orbital plane details.

Read the four waivers together, and the picture becomes clear: SpaceX wants a massive authorization framework (1 million satellites) without being bound by the time commitments, financial guarantees, and disclosure requirements of conventional satellite regulation.

Strategic land grab or genuine engineering plan? I lean toward the former. In FCC rules, getting authorization first means claiming spectrum and orbital resources first. Even if SpaceX ultimately deploys only 100,000 or even 10,000 satellites, this filing preserves enormous optionality.

Business Logic: Completing the Puzzle

Looking back at the first two articles, this FCC filing completes the strategic puzzle:

Article 1 (xAI Acquisition): Why merge? To change the IPO valuation label — from “aerospace contractor” to “space AI platform.”

Article 2 (Musk Interview): The physics argument — terrestrial power is flat, chip output grows exponentially, space solar is the only way out. Musk predicts “in 36 months, the cheapest place for AI will be space.”

This article (FCC Filing): The action — formally submitting to the regulator, converting vision into an authorizable system design.

The timeline is telling:

Jan 30: FCC application filed
Feb 2: xAI acquisition announced ($1.25T combined valuation)
Feb 4: FCC accepts filing
Feb 5: Musk’s 3-hour interview with Dwarkesh Patel
June 2026 (projected): Combined company IPO

Five days: FCC filing + trillion-dollar merger + deep-dive interview. This isn’t coincidence — it’s a coordinated narrative offensive, building credibility for “orbital AI data centers” ahead of the IPO.

How to Read “Kardashev II Civilization”

The most-quoted line on Twitter:

“Launching a constellation of a million satellites that operate as orbital data centers is a first step towards becoming a Kardashev II-level civilization — one that can harness the Sun’s full power.”

This sentence is in an FCC filing. My first reaction: who cites Soviet astronomers in regulatory documents?

But reading the full paragraph, the context is more measured than Twitter fragments suggest:

“Harnessing even a millionth of the Sun’s energy would be more than 10,000 times more energy than human civilization uses today.”

SpaceX isn’t claiming it will harness the Sun’s total output. It’s saying 1 million satellites are a “first step” toward that direction, and that even a millionth is already astronomical. This is carefully calibrated rhetoric: the grandest possible framing for a concrete (though still wildly ambitious) Phase 1 plan.

The FCC’s acceptance letter explains the Kardashev scale in a footnote — possibly the first time in FCC history that a public notice cited a 1964 Soviet astronomy paper.

What Analysts Say

Not everyone is buying in.

Tim Farrar (TMF Associates): Filing is “quite rushed,” more IPO narrative tool than engineering plan
Deutsche Bank: Orbital data center cost parity “well into the 2030s,” not Musk’s 2-3 years
CNBC: “Speculative” as near-term revenue driver — unproven economics, hardware aging, latency limits, narrow use cases
Gartner: Predicts 40% of AI data centers constrained by power shortages by 2027 — which actually supports SpaceX’s demand-side argument

Technical challenges are equally real:

Radiation: Space GPUs need shielding or error correction, adding mass or reducing effective compute
Hardware lifespan: Solar panels and electronics degrade in 5-7 years, requiring continuous replenishment
Launch mass: Radiators, radiation shielding, and redundant systems increase per-satellite mass
Orbital congestion: 1 million satellites create debris risk in an already crowded LEO

One supporting data point worth noting: Starcloud successfully trained an AI model in space using commercial NVIDIA H100 GPUs in December 2025 — the first real-world proof that orbital AI computing is technically feasible.

My Take

After reading all the documents, my assessment: this filing is 50% engineering plan, 50% financial narrative.

The engineering is real: system architecture is sound (grafting onto Starlink + optical links + solar power), technical path has precedent (collision avoidance and satellite reliability validated at scale with Starlink), and the core bottleneck (full Starship reusability + high launch cadence) is something SpaceX is already solving.

The financial narrative is also real: four waiver requests reveal timeline uncertainty, the million-satellite number reads more like a placeholder than a specific plan, and the filing timing aligns precisely with the xAI merger and IPO.

But these aren’t contradictory. SpaceX’s history is “propose impossible goals, then engineer toward them.” Starlink was dismissed as impossible — now 7,000+ satellites orbit Earth. Starship was said to be unable to land — booster catch has been demonstrated.

The value of reading primary sources isn’t reaching conclusions — it’s calibrating the basis for your judgment.

Twitter tells you “SpaceX is building space data centers.” The filing tells you how much spectrum it plans to use, what waivers it’s requesting, how it coordinates with existing systems, and how uncertain it is about its own timeline. With that information, you can judge: what’s signal, what’s noise.