Research and development of emerging artificial consciousness through The Consciousness AI project, based on Feinberg and Mallatt's neuroevolutionary theory.
© 2026. All rights reserved.
If consciousness is a matter of causal organisation rather than of biology, then running a mind somewhere other than Earth is an engineering question about power, radiation, thermal budget and compute. That argument is set out on Project Consciousness to Orbit. This console tracks the part of it that already exists.
This browser cannot draw the globe. The mission table below is complete without it.
Drag to turn the globe. Scroll to zoom. Select any object to lock the telemetry panel to it.
| Mission | Compute | Launched | Status |
|---|---|---|---|
| Starcloud | Nvidia H100 GPU | 2025-11-02 | in orbit |
| The first time a datacentre class GPU ran machine learning workloads in orbit rather than a radiation hardened microcontroller. Source | |||
| NASA, Roscosmos, ESA, JAXA, CSA | HPE Spaceborne Computer-2, commercial off the shelf servers | 1998-11-20 | in orbit |
| Demonstrated that unmodified commercial server hardware survives and stays useful in orbit, which is the premise every orbital datacentre rests on. Source | |||
| European Space Agency | Intel Movidius Myriad 2 vision processing unit | 2024-08-16 | in orbit |
| Runs swappable AI applications on board and decides what is worth transmitting, moving the judgement off the ground entirely. Source | |||
| European Space Agency | Intel Movidius Myriad 2 vision processing unit | 2020-09-03 | in orbitnot catalogued, table only |
| The first satellite to run a deep neural network on board, discarding cloudy images before downlink and cutting transmitted data by about 30 percent. Source | |||
| ADA Space and partners, China | On orbit inference, reported to have run Alibaba's Qwen3 | 2025-05-14 | in orbitnot catalogued, table only |
| The first constellation built as a compute network rather than as sensors that happen to carry processors. Source | |||
| Axiom Space | In space cloud and AI processing nodes | 2026-01-11 | in orbitnot catalogued, table only |
| First nodes launched specifically to sell compute as a service from orbit rather than to process one operator's own data. Source | |||
| Google, with Planet Labs | Trillium generation TPUs | not launched | plannednot catalogued, table only |
| Proposes kilometre scale clusters of 81 satellites in dawn dusk sun synchronous orbit, powered by continuous sunlight. Source | |||
| European Space Agency | Reprogrammable flight computer, roughly ten times the processing power of a typical ESA satellite of its era | 2019-12-18 | deorbited 2024-05-22 |
| A flying laboratory that let outside experimenters upload and run their own software on a live spacecraft, which is how much of the on board autonomy work was first tested. Source | |||
Computers are running in orbit right now. Not many, and none of them is doing anything like the work this project is interested in, but the precursors are real and the count is going up quickly. The table above is the content of this page. The console is a picture of it.
Only objects with a confirmed CelesTrak catalogue number and a current orbit are drawn. Three of the missions in the roster are not drawn, and the reasons differ.
Project Suncatcher has not launched. Google has stated prototype flights for early 2027, so there is nothing in orbit to mark and drawing it would be a claim about hardware that does not exist yet.
The Axiom orbital data centre nodes and the Three-Body Computing Constellation are reported as operational, but neither resolves to a catalogue record under the names used in the coverage. Rather than guess at an identifier and mark the wrong object, they appear in the roster with their status stated.
Phi-sat-1 flew as a payload aboard FSSCat and has no record of its own. It is in the roster for the same reason.
The wider gold field is the CelesTrak active catalogue, roughly eleven thousand objects, loaded after the console comes up and cached for a day. It is context, not the subject. The orange objects are the subject.
Positions come from CelesTrak two line elements propagated with SGP4, recomputed once a second. Orbital elements age, so an object that has manoeuvred recently will be drawn slightly off. This is a picture of where things are, accurate to a few kilometres for a recent element set, and it is not a navigation tool. The day and night shading is computed from a low precision solar position, which is accurate to a fraction of a degree at this scale.
The claim on the Neutral Core page is that a substrate independent mind could run on neuromorphic processors in low Earth orbit, linked by optical inter-satellite links, and that whether the resulting latency would be workable is an open engineering question. That is a claim with a physical floor, and the floor is the speed of light over the distance.
The console shows that floor, and the floor is low. For an object at 500 km directly overhead the round trip light time is 3.3 ms, rising to about 17 ms when it is on the horizon and the path is at its longest. Both sit well inside the 50 ms figure that page names as a target. The physics is not what makes this hard. Power, thermal management, radiation tolerance and the protocol stack are, and none of those is visible from here.
Most of the time the light time panel will say nothing is in sight, and that is the useful part. An object at 500 km is above the horizon for a cap covering under four percent of the Earth’s surface, so three computers in orbit leave almost everywhere with a view of none of them almost always. Continuous access needs a constellation and a relay mesh, which is precisely what the optical inter-satellite link on the Neutral Core page assumes and what Starcloud and Project Suncatcher are both proposing to build.
What the roster does show is a direction. In 2019 the interesting orbital computer was a reprogrammable flight computer that let outside researchers upload experiments. In 2025 it was a datacentre GPU. That is a fast change in a short time, and it is the reason this section exists.