BRENNEN

Industries / Transport and vehicles / Aerospace and defense / Space and satellites

Building satellites

Building satellites is the work of designing, making and testing spacecraft that must survive launch and then keep working for years in vacuum, heat, cold and radiation, with no chance of repair.

5

Milestones achieved, from 1990 to 2025.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1990

    Hubble's mirror tested once

    The Hubble Space Telescope's main mirror was ground to the wrong shape, which blurred its images in orbit. NASA's board of investigation found that the maker, Perkin-Elmer, used one instrument, a reflective null corrector, as its only test of the mirror's shape, and that a lens in it was 1.3 mm out of place. Other tests had shown signs of the error at the time but were dismissed as flawed, and the board said cost and schedule worries had discouraged independent tests.

    LessonA single test, however precise, needs an independent check, and results that disagree need an explanation before launch.

    ntrs.nasa.gov

    Credit

    the Hubble Space Telescope Optical Systems Board of Investigation

  2. 1998

    Tin whiskers short satellites

    Tin whiskers are hair-thin metal crystals that can grow out of pure tin plating over years and touch nearby parts. NASA Goddard's record lists two suspected tin-whisker failures in pure-tin-plated relays inside satellite control processors. On PanAmSat's Galaxy VII the main processor failed in June 1998 and the backup in November 2000. On Mexico's Solidaridad 1 the processor failed in April 1999, and the satellite was lost in August 2000.

    LessonA metal finish that is harmless on the ground can grow a short circuit after years in orbit.

    nepp.nasa.gov

    Credit

    NASA Goddard Space Flight Center, which keeps the record

  3. 2015

    All-electric satellites reach orbit

    Electric propulsion uses power from the solar panels to push out a gas, here xenon, at very high speed. It needs far less propellant than burning fuel but pushes gently. On 1 March 2015 (2 March UTC) a Falcon 9 launched two Boeing 702SP satellites, ABS-3A and Eutelsat 115 West B, stacked one on top of the other; Boeing calls them the first all-electric propulsion satellites to launch. ABS-3A climbed to its slot on its xenon thrusters and was handed over to its owner, ABS, on 31 August 2015.

    boeing.mediaroom.comnewatlas.com

    Credit

    Boeing; owners ABS and Eutelsat

  4. 2019

    Satellites on assembly lines

    In July 2019 OneWeb Satellites, a joint company of OneWeb and Airbus, opened a factory at Merritt Island, Florida, with two production lines designed to build two satellites a day. Satellite Today noted that custom-built satellites had taken more than a year each. In January 2020 OneWeb's founder, Greg Wyler, wrote that the factory was turning out two a day. McDowell's catalog counted 660 OneWeb satellites in orbit, 638 of them working, on 18 November 2026.

    satellitetoday.comadvanced-television.complanet4589.org

    Credit

    OneWeb Satellites (OneWeb and Airbus)

  5. 2025

    Thousands built, few failing

    McDowell counted 4,526 payloads launched in 2025, against 221 in 2016. SpaceX's Starlink satellites made up 66% of the 14,132 satellites he assessed as working on 13 December 2025. Of the 10,801 Starlink satellites launched by 25 December 2025, he assesses that 134, or 1.2%, failed before they reentered, with 11 more dead in orbit and slowly falling.

    planet4589.org

    Credit

    SpaceX and other builders; counts by Jonathan McDowell

Watch for

  1. Data-center chips in orbit

    Starcloud, a startup backed in part by Nvidia, says it runs an Nvidia H100, a data-center graphics chip, on its Starcloud-1 satellite and was the first to train an AI model on that chip in orbit. It plans two 8 kW compute satellites on rideshare flights in 2027. Separately, Google says its own proton-beam test of its Trillium TPU chip found no hard failures up to 15 krad(Si), a unit of absorbed radiation dose, about 20 times the 750 rad(Si) it expects behind shielding over five years; the chip's high-bandwidth memory showed the first errors, at 2 krad(Si). Google plans two prototype satellites with Planet by early 2027.

    Achieved whensomeone outside the builders publishes the chips' error rates and working hours after at least a year in orbit, or a paying customer reports work run on them.

    techcrunch.comresearch.google

    Claimed by

    Google, November 2025; Starcloud, August 2026

Open

  1. 01

    Radiation data for every part

    A tool that reads a satellite circuit board's bill of materials, the list of every part on it, and finds published radiation test results for each part: the total dose it survived and its single-event effects, errors or lock-ups caused by one charged particle. It records the source, the test conditions and how closely the tested part matches (same part number, same maker, or only a similar part), and flags parts with no data.

    The brief
    Credit · unclaimed
  2. 02

    An open satellite failure ledger

    A public table of satellite failures and serious anomalies since 2000, built from public records: operators' annual reports and investor filings, regulators' filings, insurers' statements and news. Each row gives the satellite, its maker and bus (the standard platform it was built on), the date, the subsystem that failed (power, propulsion, pointing control, payload or software), whether service was lost, the cause if one was stated, and the source.

    The brief
    Credit · unclaimed

Open 01 · the brief

Radiation data for every part

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A tool that reads a satellite circuit board's bill of materials, the list of every part on it, and finds published radiation test results for each part: the total dose it survived and its single-event effects, errors or lock-ups caused by one charged particle. It records the source, the test conditions and how closely the tested part matches (same part number, same maker, or only a similar part), and flags parts with no data.

Why it is still open

Small teams still build these lists by hand. The AcubeSAT student team in Greece described "creating a database and gathering existing information online" on dose and single-event levels for its parts, then testing parts without data in a heavy-ion beam. NASA Goddard says its radiation database holds more than 30 years of test data, but it is searched part by part. Zero-G Radiation Assurance sells a reviewed database of test results for commercial parts that, by its own account, cuts a parts-list review from months to a few days; its page gives no price and does not describe automatic matching. A web search and a GitHub search found no free or open tool that matches a parts list against the public databases.

arxiv.orgetd.gsfc.nasa.govsatsearch.co

Built with what exists

KiCad's command-line parts-list export; NASA Goddard's radiation database and ESA's radiation database in its ESCIES parts system; papers from the IEEE Radiation Effects Data Workshop; an AI model that reads PDF test reports and their tables; Python.

escies.org

  1. First step

    Export the parts list of one open CubeSat board and search NASA's and ESA's databases by hand for its ten most complex chips, to build the answer key.

  2. Done looks like

    On the parts lists of two open-source CubeSat boards, the tool finds a published result for at least 90% of the parts for which a person searching by hand finds one. An audit of 50 of its matches finds no wrong-part match, and every result in its output table carries a link.

Open 02 · the brief

An open satellite failure ledger

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table of satellite failures and serious anomalies since 2000, built from public records: operators' annual reports and investor filings, regulators' filings, insurers' statements and news. Each row gives the satellite, its maker and bus (the standard platform it was built on), the date, the subsystem that failed (power, propulsion, pointing control, payload or software), whether service was lost, the cause if one was stated, and the source.

Why it is still open

A 2014 RAND report found "no centralized, up-to-date, detailed, and broadly available database of anomalies covering many different satellites", because operators are reluctant to share and no one funds a trusted third party to run one. The closest thing is commercial: Seradata, now part of Slingshot Aerospace, offers records of more than 50,000 spacecraft, with anomaly and failure data, to operators, insurers and investors as a commercial product. ESA's open anomaly dataset covers telemetry from three of its own missions, not causes across many satellites, and ESA calls it a possible catalyst for "a future anomaly repository." A November 2026 student project infers failures in low orbit from orbit data alone, gives no causes, and was checked on four satellites.

rand.orgslingshot.spaceesoc.esa.intgithub.com

Built with what exists

The SEC's EDGAR full-text search for filings by satellite operators; McDowell's GCAT satellite catalog for identities, makers and status; an AI model to read filings and news and draft rows, with a person checking each one (AI reviewers of records are covered in the Note "Seven more flagged"); Python.

  1. First step

    Run an EDGAR full-text search for "satellite anomaly" in annual reports from 2015 to 2025 and list each satellite named.

  2. Done looks like

    At least 300 events, each with a link to a public source. Subsystem codes assigned separately by two people agree on at least 90% of a 50-event sample. The share of failures by subsystem is set beside published studies, with differences explained.