In This Issue

Who is running the network when the control room is in orbit?

🛰  The pitch: Intel wants to lift a satellite network's control room off the ground, so a data centre at 36,000 km runs the low-orbit fleet beneath it.

⚠️  The catch: The filing says such a system may need no ground-based control at all. Spectrum licences and lawful intercept assume a ground control point.

💰  Follow the money: Starcloud raised US$170M at a US$1.1B valuation, SpaceX asked the FCC for a million compute satellites, and every dollar of it went to low orbit. Nobody is funding the layer Intel patented yet.

📄  The paperwork: This is a continuation filed five weeks before its parent issued, and the claims already moved.

Somewhere Over the Pacific

You are four hours into a flight, on a video call you could not move. The picture holds, stalls for half a second, then holds again, and you blame the airline, because that is what airlines are for.

What happened is that a small antenna on the roof of the aircraft let go of one satellite and caught another. It does that every few minutes for the whole flight. The software follows a routing table calculated on Earth, uploaded during a short pass over a ground station, and already slightly stale by the time it is executed.

What if the satellite that catches your call next is much higher, much slower, and is also the machine telling the rest of the fleet what to do, because the control room is up there with it?

Intel recently published a patent describing that arrangement. A data centre in space, the filing says, may be fully autonomous and need no ground-based control at all.

Intel Corporation, based in Santa Clara, California, wants to move a satellite network's control room into a higher orbit. Three inventors are named, one of them Udayan Mukherjee, an Intel Senior Fellow and chief architect in the Network and Edge Group, the division that sells the silicon telecom networks run on.

A satellite in low Earth orbit flies 160 to 1,000 km up, laps the planet every 90 to 120 minutes, and carries a battery, a solar panel, thrusters and radios. Compute burns power, and power is what these satellites have least of, so today, the thinking happens downstairs. Right now, engineers on Earth calculate the routing tables and antenna schedules, then push them up through TTAC, short for telemetry, tracking and control.

The patent says one LEO satellite covers a 600 km radius for about eight minutes at a time, roughly every twelve hours. That is your slot.

Intel's filing moves the maths upstairs. Medium Earth orbit sits around 6,000 km, geosynchronous satellites hover near 36,000 km, and high elliptical orbits swing out to 40,000 km. Those satellites creep across the sky, stay in view for hours, and can be built bigger because you need far fewer of them. Put a data centre on one and it runs the operations centre for a whole LEO constellation, computing routing tables and pulling back telemetry on battery levels, antenna health and clock drift. The orbits talk over laser or V-band links, which work well in vacuum.

Then the claims. The system watches traffic between the constellation and terminals on Earth, meaning phones and the vSAT dishes bolted to aircraft, ships and remote sites. When a rule fires, it hands your connection up to a higher orbit and rewrites the routing so the session survives the jump. The two orbits can be run by different network operators.

The ground station keeps receiving data, but loses the job of deciding what the network does next.

There are more than 16,300 tracked satellites in orbit. Every one needs commands up and data down, and the ground has not kept pace.

Around 30 to 50 optical-capable ground stations operate worldwide, against the 200 to 500 sites SatNews reports constellations will need by 2030.

Novaspace values the ground segment at US$106 billion over the ten years to 2034. That money buys antennas and terminals but does little for buying back the eight-minute window.

Thousands of satellites still take their orders from a few dozen buildings.

Two races are running under one headline. One is about putting AI compute in orbit and the other is about who gets to run the network (which has far fewer entrants).

Starcloud, in Redmond, Washington, put an Nvidia H100 in orbit in November 2025 and processed radar imagery from Capella Space with it, per TechCrunch. Starcloud-2 goes up this year carrying a Blackwell chip, an AWS server blade and, for reasons of its own, a bitcoin miner. Starcloud sells processing time. The fleet still takes its orders from the ground.

Google's Project Suncatcher is the most serious version technically. Data Center Dynamics describes Trillium TPUs in formation-flying clusters of 81 satellites, two prototypes with Planet Labs in early 2027, and a 1.6 Tbps optical link already demonstrated in a lab. Suncatcher flies at 650 km, and SpaceX's proposed million-satellite compute constellation sits between 500 and 2,000 km. Both are building the thing Intel's patent describes controlling.

Kepler Communications in Toronto is closest to the real plumbing, having launched the first tranche of an optical relay constellation in early 2026 so satellites can pass data between themselves. Relay moves the traffic. The commands still come from Earth.

SES already sells the outcome, running GEO and MEO fleets together since closing the Intelsat acquisition in July 2025 and selling multi-orbit handover to airlines, shipping and governments.

SpaceX told the FCC in January 2026 it wants up to one million orbital data centre satellites, projecting 100 gigawatts of added AI compute a year if it can launch a million tonnes of hardware annually, according to Data Center Dynamics. It also asked the FCC to waive the rule requiring half a constellation deployed within six years. The waiver request is the more informative document.

MarketsandMarkets puts satellite non-terrestrial networks at US$0.56 billion in 2025, reaching US$2.79 billion by 2030 at 38% a year, which is fast growth on a modest base.

The other market this patent touches is bigger and older. That US$106 billion of ground segment spending through 2034 is shifting, away from antenna builders and toward whoever supplies the compute and network software riding on the spacecraft.

At the consumer layer, Omdia expects smartphone satellite direct-to-device services to bring in close to US$12 billion by 2030. Omdia separately expects defence and government buyers to spend another US$15 billion on LEO connectivity by the same year.

Intel builds no spacecraft, so its leverage sits elsewhere. We're looking at the signal around what they're doing with the Network and Edge Group (NEX), which turned over US$5.8 billion with US$931 million in operating income, according to SiliconANGLE. In July 2025 Intel said it would spin NEX out and find outside investors. On 3 December 2025 it reversed, saying the business was best positioned inside Intel. Twenty-six days later, it filed this continuation.

Which leaves Intel's real asset in orbital networking. The priority claim runs back to 21 February 2022, before Starcloud existed, before Suncatcher was announced, before SpaceX had filed for anything like a space data centre. Does a 2022 description of a control plane still read onto an industry that spent the next four years building different architecture, that will continue to develop still?

Paragraph 67 of the filing deserves a second read. A space-based data centre, it says, may be fully autonomous and not need any ground-based control.

Much of satellite regulation assumes a ground station. Spectrum licences, lawful intercept and outage reporting attach to the point where a network touches soil. Move the operations centre to 36,000 km and that point gets harder to name. Add claim 9 and a single call can cross between two companies' legal obligations mid-sentence, with no settled rule about which set applies. The United States still has no single cybersecurity regulator for space operators, as Mayer Brown sets out.

Concentration is the sharper problem. On 24 February 2022, malware later attributed to Russian military intelligence wiped around 27,000 Viasat KA-SAT modems hours before the invasion of Ukraine, and took 5,800 Enercon wind turbines in Germany off remote monitoring for weeks as spillover, per the NATO CCDCOE. The attackers came in through a misconfigured VPN, and they were only touching the edge. A high-orbit node holding mission planning and routing for thousands of satellites is one object, in a known position, that barely manoeuvres. Take it and you own the fleet's routing. Break it and you have blinded the constellation without going near a LEO satellite.

The Scar

Microsoft built ten six-metre tracking antennas across five continents for Azure Orbital Ground Station, announced in September 2020, then shut the service on 18 December 2024 and sold the dishes to Space Leasing International, as Data Center Dynamics reported. A company with effectively unlimited capital tried to own the link between space and cloud, and quit inside four years.