Same planet.
Different orbital geometry.

Your location: “Use my location” requests a single browser location fix only after you click. This page keeps the returned latitude and longitude in memory, with no analytics, remote geocoding, or persistent storage. The browser or operating system may use an external location service and may return only an approximate position; the reported accuracy is shown. Manual coordinates also work offline. Coordinates remain fixed on the globe when you explore other times. “Set to now” uses your computer’s UTC clock, longitude, and NOAA’s approximate solar equations for apparent solar time and declination. It takes a snapshot, not a live clock; changing time or season switches back to a simulation. Globe coastlines and the spherical Earth model limit placement accuracy.

Orbits: spherical Earth (radius 6,378.137 km), circular orbits, and a dawn–dusk Sun-synchronous plane with an 18:00 ascending node. Inclination is calculated from Earth’s J₂ precession, with a target nodal rate of one revolution per tropical year. The Sun-locked frame factors out that annual precession. June and December change the Sun’s declination by ±23.44°. The blue mesh uses a small sample of 53° and 70° planes, not actual Starlink orbital elements or V3 inclination assignments.

Relay route: the ground terminal first reaches a hypothetical Starlink satellite directly overhead at 350 km. An AI destination is placed at the closest possible point on the gold orbit, assuming a sufficiently populated orbit. Intermediate relays follow a great-circle direction at Starlink altitude until they can see that destination. Each laser segment clears Earth by at least 80 km. The purple nodes are hypothetical relay locations, independent of the animated blue satellites. This is a geometry demonstration, not a prediction of Starlink routing, coverage, link availability, or measured AI1 latency.

Delay: RTT = 2 × total one-way path length / 299,792.458 km/s, assuming the return follows the same route. The ideal direct-overhead floor at 600 km is about 4 ms. The illustrated relay path is not necessarily the mathematical shortest path. Radio scheduling, forwarding, bandwidth, congestion, Internet gateways, GPU queues, and inference are excluded.

Sunlight: eclipses use parallel sunlight and Earth’s geometric shadow, ignoring atmospheric absorption, the Sun’s finite size, and lunar occultations. The large percentage is the illuminated fraction of each orbit at the selected season, not a yearly average. “June solstice” sets solar declination to +23.44°, not an average across June. The full-year average separately samples each day of the reference year 2026 using NOAA’s approximate solar declination and this same orbit geometry. The annual worst case checks the seasonal range. A 600 km dawn–dusk orbit has seasonal eclipses; a suitable higher orbit can avoid Earth’s shadow year-round in this model. Sun-synchronous does not by itself mean continuously sunlit.

600 km calculation, verified: the modeled circular orbit has inclination 97.7877° and period 96.6872 minutes. At the June solstice its Sun-to-orbit-plane angle is 58.7723°. Earth blocks the Sun for 20.6860 minutes per orbit, leaving 78.6052% sunlight at that season. The full-year average is approximately 95.5%. The eclipse equation was checked against NASA’s circular-orbit eclipse geometry and independently by testing Sun-ray intersections with Earth at 500,000 evenly spaced orbital positions. An additional annual calculation including mean-Sun versus apparent-Sun node alignment gives about 95.51%, so this correction does not raise the result to 98%. These are geometric estimates for the stated orbit, not measured AI1 solar-power availability.

AI1: SpaceX describes Sun-synchronous AI satellites and laser connections to Starlink. This page explores a dawn–dusk configuration; it does not establish AI1’s final altitude, number of planes, network design, deployment status, or actual inference performance. Geographic outlines are schematic.