Independent & unofficial — not affiliated with, endorsed by, or operated by SpaceX or xAI. Brand names are used for commentary and identification only.
Independent explainer · starmind.technology
STARMIND

SpaceX wants to lift the data centre off the planet. Explore the "Star" family of space products and the xAI products they were merged to feed — all bridged by Starmind.

How the products link together

One company, two halves, bridged by Starmind

In February 2026 SpaceX acquired xAI — and Musk named orbital data centres as a main reason. On the left, the SpaceX hardware stack builds, launches and connects. On the right, xAI s products generate the AI demand and run on the ground-based Colossus supercomputer. Starmind is the bridge: it offers SpaceX s launch and connectivity as orbital compute to relieve exactly the terrestrial limits Colossus runs into. Tap any node to open its page.

SPACEX HARDWARExAI COMPUTEMERGED FEB 2026StarbaseBuild & launchStarshipTo orbitStarlinkConnectivityStarshieldSecure layerStarfallCargo returnStarmindOrbital computeColossusGround computeGrokAI assistantX platformData & reachxAI APIDev platformFalconWorkhorseStargazeTraffic control
The combined stack

SpaceX + xAI, joined at Starmind

Hover or tap a node to see how it connects, and click through to its page. Amber = launch & build; cyan = data & orbital compute; dashed = shared / secure; orange = the xAI product chain.

Launch & build Data & orbital compute Shared / secure xAI product chain
— xAI products (now part of SpaceX) —

In February 2026 SpaceX acquired xAI, folding Grok and the X platform into the company — with orbital data centres cited as a main reason for the deal. These are the AI products Starmind is ultimately built to feed.

Original illustration · orbital AI node
Launch / Starmind explainer
The orbital AI data centre

Starmind Up to a million AI satellites computing in low Earth orbit, then beaming the answers back down.

Status check, June 2026: Starmind is a confirmed name (Musk verified the trademark on X), an accepted FCC filing, and announced hardware (AI1) — not a product you can buy. Two prototypes are slated for early 2027. Everything below is sourced; speculative figures from low-reliability blogs are left out.
01 · The idea

A server, not a pipe

Starlink moves data: a satellite catches a signal and relays it. Starmind is built on similar orbital infrastructure but does the opposite job — its satellites process data, running AI inference on onboard chips powered by large solar arrays, then send results down through Starlink.

Where a Starlink satellite is a very fast pipe, a Starmind satellite is a server: run a query in orbit, beam the answer to a user anywhere on Earth, without the data ever touching a ground-based data centre.

SpaceX's reasoning: terrestrial data centres are hitting hard limits — land, community opposition, and enormous power and water demands. Orbit offers near-constant solar power, the vacuum as a heat sink, and no zoning boards.

Musk has said he expects space to become the lowest-cost place to deploy AI compute within two to three years — an aggressive timeline that regulatory and engineering reality may stretch.

02 · The hardware

AI1 — the first-generation node

SpaceX unveiled AI1 on 8 June 2026. A single satellite is a giant: roughly the height of a six-storey building, with a wingspan of solar arrays wider than a Boeing 747-8. One node is pitched as carrying compute comparable to a small data centre.

AI1 at a glance

Height~20 metres
Deployed wingspan~70 m (wider than a 747-8)
Average compute payload~120 kW
Peak payload~150 kW
NetworkingOptical inter-satellite links
DownlinkVia Gen-1 / Gen-2 Starlink
Per Starship launch~30–50 AI1 satellites
First prototypesTwo, targeted early 2027
Volume productionLate 2027, at "Gigasat"

How a query would flow

A request reaches the constellation and is routed by laser link to a satellite with spare capacity. That node runs the inference on its onboard processors, hands the result to the Starlink layer, and it's relayed to a ground station near wherever the request came from.

Each laser hop in LEO adds ~10–20 ms of round-trip latency, so analysts note the constellation suits batch inference and training far better than real-time work. It's a compute layer in the sky, not a gaming server.

03 · Where it would fly

The orbital shells

SpaceX's filing describes operation between roughly 500 and 2,000 km altitude. Tap a node to read what each band is for.

Select a node

The constellation, by altitude

Tap any glowing satellite above. The filing covers a 500–2,000 km band; the trade-off is always the same — lower means faster but more crowded, higher means broader but slower.

04 · The hard part

Why this might not work (yet)

Plenty of serious people are sceptical — OpenAI's CEO reportedly called orbital data centres "ridiculous." The objections are real engineering ones.

01

Cooling by radiation only

In a vacuum, the only exit for heat is radiating it into space — far less efficient than air or water. Dumping 120+ kW per satellite means very large radiators, and that physics dominated the public reaction.

02

Radiation-hardened silicon

The AI chips that thrive in ground data centres aren't built for orbit's radiation. Hardening them without killing performance or ballooning mass is unsolved at this scale.

03

Latency for a laser mesh

Each inter-satellite hop adds delay, confining the constellation to batch and training workloads rather than real-time applications.

04

A million-object traffic problem

At the proposed density, conjunction analysts would face an orbital-population challenge with no precedent — collision avoidance, debris, sustainability of LEO.

Counterpoint: Google's 2025 feasibility study argued orbital data centres could become cost-competitive if launch to LEO falls to ~$200/kg — projected around 2035 if Starship scales to ~180 launches/year. The economics aren't absurd; they're just not here yet.

06 · Receipts

Sources

• Space.com — "Another 'Star' is born: SpaceX names AI megaconstellation 'Starmind'"
• Wikipedia — "Space-based data center" (FCC filing, Google feasibility study)
• FCC Space Bureau — SpaceX Orbital Data Center application (accepted Feb 2026; 500–2,000 km)
• Teslarati — Starmind vs Starlink; AI1 prototypes, Gigasat, per-launch capacity
• Notebookcheck / Space.com — AI1 specs (~120 kW avg / ~150 kW peak, 20 m, 70 m wingspan)
Original illustration · ascent
Launch / The Star family / Starship
The launch vehicle

Starship

The biggest, most powerful rocket ever built — and the reason lofting a million heavy satellites is even a conversation. If Starmind is the payload, Starship is the freight elevator.

Active · V3 flying test missions
What it is

A fully reusable super-heavy rocket

124.4 m
height, V3
~100 t
payload to LEO
33
booster engines
2
reusable stages

Starship is a two-stage, fully reusable launch system: a Super Heavy booster topped by the Starship upper stage ("Ship"). The current generation, Version 3 (V3), stands about 124 metres tall and is the first to fly SpaceX's new Raptor 3 engine.

V3 debuted on Flight 12 on 22 May 2026 — the first Starship launch in over seven months and the first from Starbase's new Pad 2. The Ship completed its suborbital flight and deployed 22 dummy Starlink satellites; the booster had a faulty boostback burn and crash-landed in the ocean. SpaceX still counted the test a success. Flight 13 is targeted for no earlier than July 2026.

The numbers

Specifications & status

VehicleStarship V3 (Super Heavy + Ship)
Height124.4 m (~408 ft)
Diameter9 m
Payload to LEO~100,000 kg
Liftoff thrust~80,800 kN
Booster engines33 × Raptor 3
First V3 flight22 May 2026 (Flight 12)
Next flightFlight 13, NET July 2026
  • Each flight is expected to carry ~30–50 Starmind AI1 satellites — dozens of server racks per launch.
  • V3 added an improved booster fuel-transfer system and docking ports for in-orbit refuelling.
  • NASA selected Starship as the crewed lunar lander for Artemis; Artemis 3 targets mid-2027.
  • Deep-space trips need a dozen or more tanker flights to refuel one Ship in orbit.
Context

How it connects to Starmind

Starmind's entire economic case rests on cheap launch. Google's 2025 feasibility study suggested orbital data centres become cost-competitive only if launch costs to LEO fall to around $200/kg — which depends on Starship scaling to roughly 180 flights a year. No reusable heavy lifter, no million-satellite constellation.

Receipts

Sources

• Space.com — Starship Flight 12 updates & "What's next for Starship V3" (May 2026)
• CNN — Starship V3 debut and scrub coverage (May 2026)
• nextspaceflight / Space Launch Live — Flight 12 results, Flight 13 schedule
• Space.com / Wikipedia — orbital data-centre launch-cost threshold
Original illustration · low Earth orbit
Launch / The Star family / Starshield
The secure government layer

Starshield

Starlink's national-security sibling: a secured service layer for government and intelligence customers, built on the same orbital infrastructure but locked down.

Active · government/defence service
What it is

Security-hardened orbital networking

LEO
same orbital base
Gov / mil
target customers
Layer
not a separate fleet

Starshield is a modified, security-focused version of Starlink that flies national-security missions for the US government and intelligence agencies. SpaceX describes it as a service layer built on top of Starlink's LEO infrastructure rather than a separate constellation — so it scales with Starlink while adding the access controls and encryption its customers need.

As with several defence-adjacent programs, SpaceX has said very little publicly about Starshield — a reticence it shares with the newer Starfall capsule.

The numbers

Specifications & status

TypeSecured service layer on Starlink
CustomersUS government, defence, intelligence
ArchitectureEncryption & access controls atop Starlink LEO
ScalingGrows with the Starlink fleet
Public detailLimited by design
Relation to StarlinkShared infrastructure, separate access
  • Same orbital backbone as consumer Starlink, hardened for sensitive missions.
  • Being a layer not a fleet, capability grows automatically as Starlink launches more sats.
  • Signals SpaceX's shift from launch provider to orbital infrastructure operator.
Context

How it connects to Starmind

Starshield shows the pattern Starmind follows: take the Starlink platform and specialise it. Starshield adds secure comms for government; Starmind adds an AI compute layer. Both reuse the constellation rather than rebuilding from zero — and a secured-access model is exactly what a defence buyer of orbital inference would expect.

Receipts

Sources

• Gear Musk / Space.com — Starshield's role in the Star lineup
• basenor.com — Starshield as a service layer on Starlink
• Space.com — note on SpaceX's limited public disclosure for defence programs
Original illustration · ascent
Launch / The Star family / Starbase
The factory & launch city

Starbase

The South Texas home of Starship — manufacturing site, launch complex, and since 2025 an incorporated city. The ground end of the whole orbital story.

Active · incorporated city since 2025
What it is

Where the rockets are built and launched

2025
became a city
Pad 2
new launch complex
Boca Chica
South Texas coast

Starbase is SpaceX's development, manufacturing and launch hub at Boca Chica on the South Texas coast — where Starship is designed, welded, stacked and flown. In May 2025 the site formally became an incorporated city named Starbase.

Its facilities grow with the program: Flight 12 in May 2026 was the first launch from the new Pad 2 (Orbital Launch Pad 2), and teams there are already preparing hardware for Flight 13.

The numbers

Specifications & status

LocationBoca Chica, South Texas
FunctionStarship R&D, manufacturing, launch
City statusIncorporated May 2025
Launch padsIncluding new Orbital Launch Pad 2
Notable 2026 eventFirst Pad 2 launch (Flight 12)
ManufacturingStarfactory advanced production
  • The ground anchor for everything orbital — including Starmind's AI1 satellites, once stacked for launch.
  • On-site Starfactory is a large advanced-manufacturing facility for Starship vehicles.
  • Volume production of Starmind AI1 satellites is slated for a separate new facility called Gigasat.
Context

How it connects to Starmind

Starmind satellites must be built somewhere and launched from somewhere. Starbase is the launch side, and SpaceX's vertical integration — design, manufacture and fly from one coastal complex — is part of why it believes it can deploy orbital compute cheaply. Reporting points to a dedicated Gigasat facility for mass-producing AI1.

Receipts

Sources

• Space.com — Starbase incorporated as a city (May 2025)
• Space.com / nextspaceflight — first Flight 12 launch from Pad 2
• Teslarati — Gigasat facility for AI1 volume production
Original illustration · low Earth orbit
Launch / The Star family / Starfall
The cargo return capsule

Starfall

A saucer-shaped capsule built to bring cargo down from orbit — microgravity-made pharmaceuticals and materials today, point-to-point delivery anywhere on Earth tomorrow.

Demo flown · 23 June 2026
What it is

A disk that brings things home

3.1 m
diameter
0.75 m
height
~2,100 kg
empty mass
~1,000 kg
payload

Starfall is an uncrewed, mass-producible reentry capsule — a flat disk, often likened to a giant hockey puck, about 3.1 m across and 0.75 m tall. Its job is to return payloads from orbit safely: first for microgravity research and in-space manufacturing, later for rapid point-to-point cargo delivery on Earth.

SpaceX developed it unusually quietly; most early detail came from an FAA environmental assessment finalised 15 May 2026, which cleared up to two demonstration reentries. The first Starfall Demo launched on a Falcon 9 on 23 June 2026 from Cape Canaveral and splashed down in the Pacific. It has no main propulsion — only cold-gas (nitrogen) attitude thrusters — and separates into a top plate and heat shield before a parachute splashdown.

The numbers

Specifications & status

ShapeDisk / "hockey puck"
Diameter3.1 m (10.2 ft)
Height0.75 m (2.5 ft)
Empty mass~2,100 kg
Payloadup to ~1,000 kg
PropulsionCold-gas (N₂) attitude control only
Launch onFalcon 9 or Starship
First flightStarfall Demo, 23 Jun 2026
FAA approvalFinal EA 15 May 2026, two demos
Splashdown~1,300 km off US west coast
  • FAA documents cast Starfall as a possible "proliferated successor" to the ISS for in-space manufacturing.
  • The demo tested Starlink antennas through the plasma-blackout phase of reentry.
  • Military interest is explicit: orbiting capsules dropped on command could resupply remote sites fast.
  • Microgravity enables products hard to make on Earth — certain crystals, optical fibre, even bio-printed tissue.
Context

How it connects to Starmind

Starfall and Starmind point at the same future: orbit as a place where valuable things happen, not just where data passes through. Starmind computes in space; Starfall manufactures and returns physical goods. Both treat low Earth orbit as industrial real estate — and both lean on Starlink connectivity, right down to keeping a data link alive through reentry plasma.

Receipts

Sources

• Space.com — "What is Starfall?" and the 23 June 2026 launch report
• Spaceflight Now / NASASpaceflight — Starfall demo coverage & FAA EA
• Wikipedia — "SpaceX Starfall" (dimensions, mass, payload)
• New Atlas — Starfall military / orbital supply-depot applications
Original illustration · ground compute
Launch / xAI products / Grok
The AI assistant

Grok

xAI's chatbot and model family, wired into the X platform and trained on the Colossus supercomputer. After the merger, Grok is the AI that Starmind's orbital compute is ultimately meant to serve.

Active · Grok 4.x family
What it is

A frontier model with a live feed

Grok 4.x
current family
Real-time
X / web search
Voice
+ image, video
API
grok.com & x.ai

Grok is xAI's conversational AI — a frontier large-language-model family with native tool use, real-time search across the web and the X platform, and voice, image and video generation. It's available on grok.com, inside X, on mobile apps, and through the xAI API.

The lineage runs Grok-1 (open-weighted in 2024) through the Grok 3 and Grok 4 families to the current Grok 4.x releases, with a coding-focused "Grok Build" agent and Microsoft 365 add-ins among the 2026 additions. xAI has talked openly about pushing toward much larger models and an aggressive release cadence.

The numbers

Specifications & status

MakerxAI (now part of SpaceX)
Current familyGrok 4.x
ModalitiesText, voice, image, video
Live dataReal-time web + X search
Accessgrok.com, X, mobile, xAI API
Coding agentGrok Build (CLI)
Office add-insGrok for Word / PowerPoint / Excel
OriginFounded 2023; name from "Stranger in a Strange Land"
  • After the Feb 2026 merger, Grok and X became part of SpaceX — the same company building Starmind.
  • Grok runs on Colossus today; orbital compute (Starmind) is pitched as a future way to scale that further.
  • xAI's model cadence has been unusually fast, with multiple Grok 4.x point releases in 2026.
  • Grok has also drawn serious safety controversy and regulatory scrutiny over generated content.
Context

How it connects to Starmind

Grok is the demand side of the Starmind story. Musk's argument for orbital data centres is that terrestrial compute can't scale fast or cheaply enough to feed models like Grok — so moving inference and training to orbit is framed as the way to keep growing. Starmind is, in effect, future infrastructure for Grok-scale AI.

Receipts

Sources

• x.ai — official product news and model pages
• CNN / CNBC — SpaceX–xAI merger (Feb 2026)
• Releasebot — xAI June 2026 release notes (Grok Build, Office add-ins)
• Independent statistics roundups — Grok model history & cadence
Original illustration · ground compute
Launch / xAI products / Colossus
The supercomputer

Colossus

xAI's giant GPU supercluster in Memphis — the ground-based engine that trains Grok today. It's the terrestrial extreme of exactly the problem Starmind proposes to move off the planet.

Active · scaling toward 1M GPUs
What it is

The world's largest AI training cluster

Memphis
Tennessee site
Hundreds of K
NVIDIA GPUs
Gigawatt
scale power draw
1M GPUs
stated target

Colossus is xAI's flagship supercomputer, built in Memphis, Tennessee, to train the Grok models. xAI says the first build went up in 122 days and was then roughly doubled, with a public roadmap toward one million GPUs and a second site, Colossus 2, brought online at gigawatt scale.

Reported figures vary by source and move fast, so treat exact GPU counts and power numbers as approximate — but the direction is consistent: hundreds of thousands of NVIDIA GPUs drawing on the order of a gigawatt or more.

The numbers

Specifications & status

OperatorxAI (now part of SpaceX)
LocationMemphis, Tennessee (+ Colossus 2)
HardwareNVIDIA GPUs (H100 / GB200-class)
ScaleHundreds of thousands of GPUs
Power~gigawatt-class draw
RoadmapToward 1,000,000 GPUs
First buildReported ~122 days
PurposeTraining the Grok model family
  • Colossus is the ground-based answer to AI compute demand; Starmind is the proposed orbital alternative.
  • Its power and cooling needs are exactly the "hard terrestrial limits" Musk cites to justify orbital data centres.
  • Reported GPU counts and power figures differ between sources and change quickly.
  • NVIDIA is both a key supplier and, in xAI's history, an investor.
Context

How it connects to Starmind

Colossus and Starmind are two answers to one question: where does the compute to train and run frontier AI live? Colossus is the terrestrial maximum — enormous, power-hungry, hard to permit. Starmind is the bet that the next order of magnitude is cheaper in orbit, where solar power is constant and there are no zoning boards. The contrast is the clearest single illustration of why SpaceX is even attempting orbital data centres.

Receipts

Sources

• x.ai/colossus — official Colossus page
• NxCode / EPC Group — Colossus 2 reporting (Memphis, gigawatt scale)
• Independent statistics roundups — GPU-count estimates (vary by source)
• Musk public statements on terrestrial compute limits
Original illustration · ground compute
Launch / xAI products / X platform
The platform

X platform

The social platform (formerly Twitter), folded into xAI and then into SpaceX. It's both Grok's real-time data source and its largest distribution channel.

Active · merged into the company
What it is

The real-time data and distribution layer

X
the platform
Real-time
data feed for Grok
Distribution
Grok in-app

X — the platform formerly known as Twitter — was merged into xAI in 2025 and then carried into SpaceX with the 2026 acquisition. For the AI side of the business it plays two roles: it's a live stream of real-time data that Grok can search, and it's the largest place Grok is put in front of users.

That tight coupling is a defining feature of Musk's combined company: the model, the data feed, and the distribution channel all sit under one roof.

The numbers

Specifications & status

TypeSocial platform
Former nameTwitter
Merged into xAI2025
Now part ofSpaceX (via xAI acquisition)
Role for GrokReal-time data + distribution
IntegrationGrok built into the X apps
  • X gives Grok a real-time pulse of public conversation to search and cite.
  • It's also Grok's biggest shop window — millions of users meet the AI inside the app.
  • The X + xAI + SpaceX stack is the "vertically integrated" engine Musk described at the merger.
Context

How it connects to Starmind

X is the consumer-facing end of the same chain that, far upstream, leads to Starmind. Data and queries originate with users on X and elsewhere; Grok answers them; the compute behind those answers is what SpaceX wants to scale — first on Colossus, eventually in orbit on Starmind.

Receipts

Sources

• CNN / CNBC — merger coverage noting X folded into the combined company
• Motley Fool / Yahoo Finance — X–xAI–SpaceX combination history
• x.ai — Grok integration with the X platform
Original illustration · ground compute
Launch / xAI products / xAI API & tools
The developer platform

xAI API & tools

The xAI API and Grok's coding tools — how developers build on the same models. After SpaceX's $60B Cursor acquisition, the coding side of the business grew sharply.

Active · public API & coding tools
What it is

Grok for builders

api.x.ai
endpoint
Grok Build
coding agent
Bedrock
also on AWS
Cursor
$60B acquisition

Beyond the consumer app, xAI exposes its models through a developer API (api.x.ai) with usage-based pricing, plus a growing set of coding tools — the Grok Build CLI agent, a plugin marketplace, and Office add-ins. Grok models have also been made available through Amazon Bedrock.

SpaceX leaned further into developer tooling after its June 2026 IPO by acquiring the AI coding startup Cursor in a large all-stock deal, signalling ambitions to compete more directly in AI-assisted software development.

The numbers

Specifications & status

APIapi.x.ai (OpenAI-compatible)
PricingUsage-based per token
Coding agentGrok Build (CLI, /goal mode)
MarketplaceGrok Build plugins
Cloud availabilityAlso via Amazon Bedrock
Office add-insWord / PowerPoint / Excel
Notable acquisitionCursor (post-IPO, 2026)
  • The API lets third-party products run on the same Grok models the apps use.
  • Grok Build is a terminal-based coding agent with a long-running autonomous mode.
  • The Cursor acquisition is reported as one of the largest startup buyouts ever.
  • Coding has been an area where xAI openly says it wants to catch up to rivals.
Context

How it connects to Starmind

The developer platform is how Grok-scale demand multiplies: every app built on the API is more inference to run. That compounding demand is the business case Starmind is meant to answer — if orbital compute really is cheaper at scale, the API and tools are what would fill it.

Receipts

Sources

• x.ai — API docs and developer platform
• Releasebot — Grok Build, plugin marketplace, Bedrock availability
• techjournal.org — SpaceX's post-IPO Cursor acquisition
Original illustration · launch & landing
Launch / SpaceX products / Falcon
The workhorse rocket

Falcon

The rocket doing the real work today. Falcon 9 is the first reusable orbital-class launcher — it lifts Starlink, Dragon and the Starfall demo, and proved the reusability that makes everything else affordable.

Active · ~140 launches planned in 2026
What it is

Reusable, relentless, everywhere

667
Falcon launches
~140
planned in 2026
~650
booster landings
Block 5
current variant

Falcon 9 is SpaceX's medium-lift launch vehicle and the most-flown rocket in the world. Its defining breakthrough is the reusable first stage: boosters land on a ground pad or an ocean droneship and fly again, collapsing the cost of reaching orbit. Falcon Heavy adds two extra boosters for heavy-lift missions.

As of June 2026 the Falcon family has flown 667 times with a near-perfect success record, and SpaceX president Gwynne Shotwell has said the company expects around 140 Falcon 9 launches across 2026 — roughly three a week. It carries Starlink batches, Dragon crew and cargo runs, national-security payloads, and flew the Starfall demo capsule.

The numbers

Specifications & status

VehiclesFalcon 9 Block 5, Falcon Heavy
Total Falcon launches667 (June 2026)
Planned 2026 cadence~140 Falcon 9
Booster landings~650 to date
ReusabilityFirst stage lands & reflies
Engines9 × Merlin (first stage)
First reuse2017 (SES-10)
Launch sitesCape Canaveral, KSC 39A, Vandenberg
RoleLaunches Starlink, Dragon, Starfall
  • Falcon 9 flew the Starfall demo on 23 June 2026 — the same vehicle that lifts almost everything SpaceX does.
  • Reusability proven on Falcon is the foundation Starship scales up, and the reason cheap orbital compute is even plausible.
  • At ~3 launches a week, Falcon is the highest-cadence orbital rocket ever operated.
  • Falcon Heavy handles the heaviest payloads, from Europa Clipper to large GTO satellites.
Context

How it connects to Starmind

Before Starship can loft Starmind satellites by the dozen, Falcon is what made reusable launch real and routine. It still flies the constellation Starmind depends on (Starlink) and demonstrated the return technology (via the Starfall demo) that the wider orbital economy needs. Falcon is the proof of concept the whole stack is built on.

Receipts

Sources

• Wikipedia — "List of Falcon 9 and Falcon Heavy launches" (667 launches, June 2026)
• Wikipedia — "SpaceX" / "SpaceX launch vehicles" (reusability, cadence)
• Spaceflight Now — Starfall demo launched on Falcon 9 (June 2026)
• FAA — SpaceX Falcon Program (sites, configuration)
Original illustration · orbital tracking
Launch / SpaceX products / Stargaze
The traffic-control layer

Stargaze

SpaceX's space-situational-awareness system: it turns Starlink's tens of thousands of star trackers into a planet-scale collision-detection network — and directly answers the biggest objection to a million-satellite constellation.

Active · free SSA platform
What it is

Collision avoidance at constellation scale

~30,000
star trackers
~30M
object transits/day
Minutes
conjunction screening
Free
to all operators

Stargaze, unveiled in February 2026, is a Space Situational Awareness (SSA) system built on top of the Starlink fleet itself. It leverages roughly 30,000 star trackers already aboard Starlink satellites to detect on the order of 30 million object transits a day, delivering conjunction screening — the calculation of whether two objects are on a collision course — in minutes rather than the hours typical of ground radar.

SpaceX says it will offer Stargaze free of charge to all satellite operators through its space-traffic-management platform. In one cited close call, a third-party satellite manoeuvred with hours to go, collapsing the miss distance to about 60 metres; Stargaze detected it and a Starlink satellite was able to plan an avoidance manoeuvre within an hour.

The numbers

Specifications & status

TypeSpace Situational Awareness (SSA)
Unveiled18 February 2026
Built onStarlink star trackers (~30,000)
Detection~30 million object transits/day
Screening speedMinutes (vs. hours for radar)
PricingFree to all operators
BetaClosed beta with 12+ operators
PurposeCollision avoidance & orbit estimation
  • Stargaze directly answers the headline criticism of Starmind: orbital traffic at million-satellite density.
  • It reframes Starlink's scale from a congestion liability into an active safety asset for everyone in orbit.
  • Screening in minutes (not hours) matters as low Earth orbit gets more crowded.
  • SpaceX is positioning itself as the operator of the safety layer for all of LEO, not just its own fleet.
Context

How it connects to Starmind

Stargaze is the safety case for Starmind. The single loudest objection to putting up to a million AI satellites is that low Earth orbit can't be managed at that density. Stargaze is SpaceX's answer — a real-time, constellation-wide traffic-control layer. If Starmind is the ambition, Stargaze is the argument that the ambition can be made sustainable.

Receipts

Sources

• Spaceflight Now — SpaceX unveils "Stargaze" SSA system (Jan/Feb 2026)
• basenor.com — Stargaze: ~30,000 star trackers, ~30M transits/day, minutes-scale screening
• SpaceX update — free space-traffic-management platform for all operators