Tracked, so it can be dodged
Radar networks such as the US Space Fence track these objects, and operators can manoeuvre around them.
Orbit's missing dataset.
A million objects, moving ten times faster than a bullet, too small to track and too big to shield, and almost no one is watching.
Request early data accessPre-launch. Mission modelled end to end. The animation is an illustration, not flight data.
First stereo pair in orbit, targeted for 2028See the roadmap
The population
Too big to shield. Too small to dodge. Almost no one is watching.
| Size | Objects in orbit (ESA)1 | Detected per year, 750–900 km: Speculai, 1 pair → 36 satellites19 | Detected per year: NASA-style radar14 | Observing hours a year |
|---|---|---|---|---|
| > 10 cm | ~68,450 | About 47,000 are tracked by radar networks, and operators can manoeuvre around them. | ||
| 1–10 cm | ~1.5 million | up to 22% → 78% | 1.5–2.2% | NASA radar: ~500 hSpeculai: ~8,800 h per pair, ~158,000 h with 36 satellites |
| 5–10 cm | ~22,000 in the band | up to 99.6% → all | 1.5–2.2% | |
| 2–5 cm | ~110,000 in the band | up to 47% → 99% | 1.5–2.2% | |
| 1–2 cm | ~280,000 in the band | up to 6% → 70% | 1.5–2.2% | |
| Below 1 cm | ~230 million | up to 0.4% → 6% (0.5–1 cm) | 1.5–2.2% | Radar: ~500 h. Not our band: shielding stops most of these. |
~500 ha year of small-debris radar sampling (NASA)
~8,800 h → ~158,000 ha year of Speculai observing, one pair → 36 satellites
We are a data company first. AI can now pull far more from data than ever before, but in orbit there is almost no data on small objects to pull it from. Speculai flies stereo pairs of small telescopes through the debris band to collect it: counts, ranges and sizes of the 1–10 cm fragments that nobody measures continuously today.
1 M+fragments of 1–10 cm, too small to track and too big to shield (ESA model estimate)1
~10 km/saverage collision speed in low Earth orbit, about ten times a rifle bullet5
$42.3 Bupper estimate of the cost of unaddressed debris to industry, 2025–2035; the range is $25.8–42.3 B9
~300,000avoidance manoeuvres by Starlink alone in 2025, every one for a tracked object22
The problem
Satellites have two defences against debris. Shielding stops fragments up to about 1 cm. Avoidance manoeuvres work for objects above about 10 cm, because radars track those. Everything in between has neither45, and by industry estimates it carries most of the risk of a mission-ending impact2.
Shielding stops most of it
Whipple shields and design margins are built for this size.
Nobody has measured it
The count is extrapolated from models (ESA MASTER, NASA ORDEM). Industry estimates put more than 95% of the mission-ending collision risk in this band2.
Why it hurts: debris moves at about 7–8 km/s in orbit and about 10 km/s on average when two objects meet. At that speed a 1 cm aluminium fragment carries about 70 kJ, roughly a family car at 20 mph, concentrated on a spot the size of a fingertip30.
Why it is getting worse
More satellites. More debris. A hidden tax.
The space economy is booming: the World Economic Forum and McKinsey project about $1.8 trillion by 203520. More satellites means more debris, concentrated in the busiest band. Starlink alone made about 300,000 avoidance manoeuvres in 2025, roughly 820 a day and about 50% more than in 202422. Its latest filing counts 207,152 more in the six months to May 202621. Every one of them was for a tracked object. None were for the untracked million.
$25.8–42.3 billion, a "hidden tax"
Debris costs industry this much over 2025–2035 even with no major collision. A destructive collision would push it far higher9.
5–10% of mission cost in GEO
Debris-related costs already run at 5–10% of total mission costs in geostationary orbit, potentially hundreds of millions of dollars per mission13.
Risk index about 4× the safe level
ESA's health index puts the environment at about four times its long-term sustainability threshold, and projects that debris would keep growing even if launches stopped today23.
Where it sits
The densest debris is where the most valuable satellites fly.
Debris is densest at roughly 750–900 km, shaped by two events in that band: China's 2007 Fengyun-1C anti-satellite test at about 865 km, which created thousands of tracked fragments26, and the 2009 Iridium 33–Cosmos 2251 collision at about 790 km, the first accidental collision between two satellites27. The WEF model finds clusters at about 775, 840 and 1,000 km, with a chance of a serious collision that could reach 29% in some zones by 2032 without mitigation9. The same altitudes host the weather and Earth-observation satellites economies depend on: Europe values its weather services at €52.8 billion a year, and credits the single polar satellite Metop-A with €4.9 billion a year of forecast benefit10.
A ~1 cm fragment hit a European radar satellite
An untracked fragment struck ESA's Sentinel-1A at about 11 km/s, leaving a 40 cm damaged area and cutting solar-array power by about 5%. It struck a solar panel, and the satellite kept working67. Its orbit is at about 693 km28, just below the densest band.
A 5 mm hole in the station's robotic arm
Inspectors found the hole during a routine check. NASA and the Canadian Space Agency could not say what hit the arm, or when, because the object was too small to track8.
New debris clouds, counted only above ~10 cm
A Starlink satellite (March), a Chinese rocket body (August), USA 32 (September) and Meteor 2-7 (October) broke up in low Earth orbit. Each breakup leaves far more small fragments than anyone counts17.
What we are not saying: many failures in low Earth orbit go unexplained, and the untracked population is one suspect nobody can rule out. We do not claim any failure was caused by it. The environment is getting worse regardless: even the International Space Station, below the densest debris, dodged tracked objects about 1.4 times a year from 1999 to 2017, and about 3.5–4 times a year from late 2022 to 202518. It can only dodge what is tracked.
Why the data doesn't exist yet
The models exist. The data doesn't.
Robotics and AI companies are racing to collect real-world data, because a model is only as good as its data. Orbit is the same. ESA's and NASA's debris models disagree for small fragments because so little has been measured. We collect the dataset that the whole expanding space economy will need.
Ground radars look for a few hundred hours a year
NASA's HUSIR and Goldstone radars together give about 500 hours of small-debris data a year14. After one Goldstone antenna was retired, its detection rate fell from 45–50 to 8–10 per hour15. The measurement the world relies on is shrinking.
The models disagree, and lack data to settle it
ESA's MASTER and NASA's ORDEM agree on total debris mass but diverge sharply for small fragments. Their validation data are limited in altitude, coverage and time16.
Every US licence includes a small-debris number
The FCC requires every satellite applicant to state whether the chance of being disabled by small debris is 1 in 100 or less, computed with NASA's model12. Operators spend 5–10% of mission cost on debris protection and mitigation (OECD, for geostationary missions)13.
The tools finally exist, and governments are asking
Rideshare launch, iodine ion thrusters and space-qualified AI processors make a 12U telescope affordable. US intelligence research agency IARPA is funding sensors and automated analysis for "lethal but untrackable" debris, with special interest near 800 km3.
Who uses the data
One dataset. Six kinds of customer.
Collision risk and constellation design
Know the measured small-debris risk of an orbit before choosing altitude, shielding or fleet size, and back the small-debris figure in licence filings with an independent check12.
Pricing low Earth orbit risk
Few satellites in low orbit are insured. A measured small-debris flux is the input a premium needs (see below).
Compliance and Zero Debris 2030
Independent measurements to test the models behind licensing rules and behind ESA's Zero Debris goal for 203029.
Space domain awareness
A persistent picture of small-fragment flux by altitude, and change reports within days of a breakup, with leads that cue radars to the right patch of sky.
Where removal pays most
Flux by altitude and size shows which orbital shells carry the most small-fragment risk. Picking individual objects to remove needs ground radars alongside us.
Ground truth for environment models
A persistent, calibrated record to validate and improve the debris models everyone else uses.
What we deliver
Small-debris risk by orbit
Measured density of 1–10 cm debris by altitude, size and week, with error bars and the ratio to NASA and ESA model predictions.
For operators and manufacturers: shielding, orbit choice and the small-debris number in licence filings.
Breakup and change reports
Counts of small fragments within days of a breakup, and alerts when a band's debris flux jumps.
For operators flying in that band, insurers and agencies.
Ground truth for the models
A persistent, calibrated record to validate and improve the debris models everyone else uses.
For NASA's Orbital Debris Program Office, ESA's Space Debris Office and model builders.
Where to look next
Positions and short-term orbits of new fragments, so ground radars can follow up and add them to catalogues.
For radar networks, space-domain-awareness providers and defence.
No data, no pricing, no market
~$550–600 Mglobal space-insurance premiums a year, 2025 forecast24
< 1%of active low-orbit satellites carry in-orbit insurance: about 300 of roughly 10,00025
Self-insuredmost large constellations carry their own risk, and SpaceX buys no third-party in-orbit cover for Starlink25
Our algorithms
With the data, we build the models.
Detection is the first layer. On top of it we build models of how these fragments behave: how many there are by altitude and size, how their orbits evolve, and how a breakup changes the picture week by week. A persistent, calibrated record is what lets those models be trained and checked instead of assumed.
Find streaks at the noise floor
Onboard star removal and matched-filter streak search turn about 1.1 TB of raw frames a day into about 125 MB of detections19.
Range and size from two views
Stereo parallax gives range. Range plus brightness gives size, with stated error bars for the unknown surface reflectivity19.
Populations, orbits, evolution
Debris density by altitude, size and orbital plane, compared with NASA and ESA predictions, and updated as the environment changes.
Why we get there and the ground can't
To see 1–10 cm you must be close, above the atmosphere, and always watching.
Radar signal fades with the fourth power of range
A debris object at 800 km altitude is 800 km from a radar directly beneath it, and often 1,000–2,000 km away along the line of sight at typical elevation angles. A telescope flying in the debris band can be tens to hundreds of kilometres away.
Orbit has no clouds and no turbulence
Ground telescopes need clear, dark skies and fight atmospheric turbulence. In orbit there is no atmosphere, no cloud and no daylight cut-off.
Ground sites are fixed and few
They see each object only on brief passes and leave large gaps, such as the oceans and the southern hemisphere. A constellation in orbit watches the band continuously.
Two eyes turn brightness into size
Stereo pairs in orbit measure range directly, so brightness becomes size, with error bars for an assumed reflectivity (about ×/÷1.6)19. Ground optical telescopes get angles only.
17× to 315× today's radar hours
One pair adds about 17 times the observing hours that today's radar campaigns give; a fleet of 36 adds about 300 times19. Each pair is expected to log tens to hundreds of 1–10 cm detections a day.
How we do it
Go to the debris. Use two eyes. Think on board.
01 · Proximity
Look from hundreds of kilometres, not thousands.
A 15 cm telescope flying through the debris band sees a 5 cm fragment at 140–440 km. Ground radars look from thousands of kilometres away; a CubeSat-sized radar would see the same fragment at only about 1–3.5 km19.
02 · Stereo
Range turns a dot of light into a size.
One camera can't tell a 1 cm fragment nearby from a 5 cm one far away. Our satellites fly in pairs about 3 km apart. The parallax between the two views gives range, and range plus brightness gives size. That is what makes small fragments measurable.
03 · Onboard AI
Find faint streaks at the noise floor.
Small fragments cross the field of view in a fraction of a second, close to the noise. Detection runs on the satellite: about 1.1 TB of raw frames a day becomes about 125 MB of detections, small enough to send down every day19.
The spacecraft
A 12U CubeSat built to look, not to listen.
- Platform
- 12U CubeSat, 16.7 kg at launch
- Sensor
- 15 cm passive telescope, 4° × 4° field, no moving parts
- Formation
- Pairs about 3 km apart along the orbit
- Orbit
- Dawn-dusk sun-synchronous, raised by an iodine ion thruster to 450 × 1,000 km so each pass sweeps the whole debris band
- Processing
- Onboard star removal and streak detection; detections, not images, go to the ground
- End of life
- Thruster lowers the orbit and a drag sail deploys: re-entry in under a year
Performance figures come from our first-order engineering models and will be measured on orbit19.
In orbit vs. ground-based
What a ground-based system can't do.
Ground-based
~500 ha year of small-debris radar sampling at NASA's two research radars14
800–2,000 kmtypical distance to a debris object (our estimate)
Ground-based radar and optical vs. in-orbit sensing
| Ground radar | Ground optical | In orbit (Speculai) | |
|---|---|---|---|
| Smallest objects | About 10 cm is routinely catalogued. Research radars sample down to a few millimetres, as statistics14 | About 10 cm and up in low orbit (our estimate) | Detects down to about 1 cm and measures range and size from about 2 cm (model). Millimetres are radar's niche, not ours19 |
| Observing hours | Limited to passes over a site | Night and clear sky only | Continuous: about 8,800 h a year per pair at 50% duty19 |
| Range to target | 800–2,000 km (our estimate) | 800–2,000 km (our estimate) | Tens to hundreds of km |
| Measures size | Radar cross-section, converted to size by a model | No: angles only | Yes: stereo range plus brightness, assumed reflectivity (about ×/÷1.6 error) |
| Weather and daylight | Partly limited | Limited | Not limited |
| Geographic gaps | Yes | Yes | No: the whole 450–1,000 km band |
| Detection rate, 1–10 cm | No public count of sub-10 cm objects held by any commercial radar, and no independent validation (our reading of public sources) | Not applicable | 32–308 detections a day per satellite; the range is our conservative and optimistic optics models19 |
| Revisit | Hours to days for catalogued objects; sub-10 cm objects are not catalogued (our estimate) | Passes at night | A 20% change in a band's flux shows within about 0.3–3 days with 12 satellites. Individual small objects are revisited over months, so the census comes first19 |
| Cost per observation | No public figure | No public figure | About $3–30 per 1–10 cm detection, over three years including launch and operations (modelled)19 |
Why it compounds
A record nobody can collect after the fact.
The debris environment changes every week. Breakups add fragments without warning, and atmospheric drag removes them at a rate that varies with the solar cycle. Training images can be simulated; the environment cannot. It has to be measured.
Every week we observe adds to a calibrated time series that a later entrant cannot collect retroactively. That record trains the models that predict what comes next, and the better the models, the more the record is worth.
Roadmap
Census first. Then a catalogue.
- 2026
Mission design
Detection, stereo, orbit, propulsion, power, cost and disposal modelled end to end.
- 2027
Ground sky test and first partners
Our camera and onboard pipeline observe real objects in orbit from the ground. Data partners define the first products.
- 2028 (target)
First stereo pair in orbit
First measured sizes and ranges of 1–10 cm fragments from space.
- Next
6, then 12, then 36 satellites
A weekly flux map, change detection within days, and population-scale coverage.
What we sell first
Counts, sizes and distribution of the 1–10 cm population by altitude, from the first pair onward. With one pair: first measured sizes of 1–10 cm fragments from orbit, calibrated against hundreds of catalogued objects a day. With 6–12 satellites: a living map, updated weekly. With 36: an estimated 15–77% of the small-fragment population observed at least once a year19.
Buyers: insurers, regulators, researchers, defence.
Where we go next
As the constellation grows, we can follow individual objects and build a catalogue of the small fragments nobody tracks today. We make no per-object claims for the first pair.
Buyers: satellite operators.
What we are careful to claim
Precise about what we do, and what we don't.
- Performance figures on this page come from our first-order engineering models. Measuring them on orbit is our first milestone.
- The first pair measures the 1–10 cm population: counts, sizes and distribution by altitude. It does not yet maintain a catalogue of individual fragments.
- We do not issue collision warnings for individual small fragments. Our single-pass orbits are good for tens of minutes, enough to cue a radar.
- We detect down to about 1 cm and measure range and size from about 2 cm. Millimetre-scale fragments are the realm of large radars and impact sensors, not ours.
- Comparisons with ground-based systems are our estimates from public sources. Sizes assume a surface reflectivity and carry an error of about ×/÷1.6.
Early data access
Fly in the 500–1,000 km band? We want to talk.
We are choosing a small group of satellite operators, insurers and agencies to shape our first data products. Tell us which orbits you care about and what decision the data would change.
Sources
Where the numbers come from.
- ESA Space Debris Office, Space Environment Statistics (MASTER-8 reference population, February 2026; the page was updated on 14 October 2026).
- D. McKnight (LeoLabs), Managing Orbital Debris (2024) and GNOSIS talk (2021): lethal non-trackable debris (~1–10 cm) carries an estimated 95–98% of mission-terminating risk. Industry estimate.
- IARPA, SINTRA: A Bold New Space Debris Enterprise (2023), citing NASA OIG (2021); Breaking Defense (Aug 2023).
- Tagawa et al., Space-Based Short Range Observation for LEO Debris, 6th European Conference on Space Debris (2013).
- NASA Orbital Debris Program Office, Frequently Asked Questions.
- ESA, Copernicus Sentinel-1A satellite hit by space particle (2016).
- IADC/IAA meeting minutes, Sentinel-1A impact analysis (2017): ~1 cm, 0.2 g impactor.
- EarthSky, Space debris smacked ISS robotic arm (June 2021), reporting the Canadian Space Agency.
- World Economic Forum, Clear Orbit, Secure Future (Jan 2026); E&T summary.
- EUMETSAT, socio-economic benefits study (2023); AMS paper on Metop-A benefits (2013).
- Orbits: ESA Sentinel-2; WMO OSCAR, NOAA-21; Landsat 9.
- US eCFR, 47 CFR § 5.64; FCC debris-mitigation checklist.
- ESA, The cost of space debris, summarising OECD analysis.
- ESA 9th Space Debris Conference, Monitoring the small debris environment with ground-based radar (2025).
- Micro-meteoroid and orbital debris radar from Goldstone radar observations, Journal of Space Safety Engineering.
- Astronomy & Astrophysics (2026); ESA 9th Space Debris Conference, paper 367 (2025).
- Space.com (Oct 2026); EarthSky (2026).
- NASA, USA Space Debris Environment, Operations, and Research Updates (2018): 25 manoeuvres since 1999; NewSpace Economy (2026): ~41 by April 2025.
- Speculai engineering models (first-order; optical detection, stereo geometry, swarm yield, cost, and a 2026 detection-versus-radar run for the 750–900 km band). Conservative and optimistic cases differ by about 10× in detections; available to partners on request.
- World Economic Forum and McKinsey, Space: The $1.8 trillion opportunity for global economic growth (2024).
- Space.com, Every SpaceX Starlink satellite has to dodge a collision almost weekly (2026), citing SpaceX's semiannual FCC report: 207,152 manoeuvres, December 2025 to May 2026.
- N2YO, citing New Scientist, SpaceX's Starlink dodged 300,000 satellite collisions in 2025: about 300,000 manoeuvres in 2025, up about 50% on 2024, from SpaceX's FCC filing. Secondary source.
- ESA, Sounding the alarm: ESA introduces space environment health index; Space Environment Report 2025: index of about 4 against a long-term sustainability threshold of 1.
- Seradata, Space insurance: promising start to the year (2025): gross premium about $580 million in 2024, forecast $550–600 million for 2025.
- NewSpace Economy, The orbital insurance market: how underwriters are pricing constellation-scale risk (Mar 2026).
- Fengyun-1C anti-satellite test, 2007 (about 865 km): Wikipedia, summarising NASA Orbital Debris Program Office data.
- Iridium 33–Cosmos 2251 collision, 2009 (about 790 km): Wikipedia, summarising NASA data.
- Sentinel-1 orbit (about 693 km): Wikipedia, summarising ESA.
- ESA, Space Debris, including the Zero Debris approach (goal: 2030).
- Our calculation: a 1 cm aluminium sphere (2.7 g/cm³) has a mass of about 1.4 g; at 10 km/s its kinetic energy is about 70 kJ, the same as a 1,500 kg car at about 22 mph.