A sky-view model of how proposed space-based AI datacenters — large, sun-pointed solar arrays in low Earth orbit — would appear from a chosen spot on the ground.
Read it with caution: apparent brightness is partly subjective and screen-dependent, so the display will not match a real sky, and the model rests on many assumptions — some adjustable in the console, others fixed and described below. Naked-eye visibility also says little about the effect on astronomy, since telescopes record objects far fainter than the eye can see.
SpaceX's FCC filing describes datacenter satellites in two kinds of orbit, and you can show either or both here. They are not interchangeable — they trade power against coverage in opposite ways.
"Sun-synchronous" is whatever near-polar, slightly retrograde inclination (about 97.6° at 550 km, rising with altitude) makes the orbit plane precess once per year so it holds a fixed angle to the Sun. Parked along the day/night terminator — the dawn–dusk configuration — such a satellite is sunlit for most of the year, falling into Earth's shadow on each orbit only during a roughly three-month season around the northern-summer (June) solstice. That near-constant power suits steady, around-the-clock compute. From the ground it shows up as a roughly north–south arc, and only during your own dawn and dusk, because that is when its plane is above your horizon while your sky is dark.
A 30° orbit stays within ±30° latitude and does not track the Sun, so each satellite passes through Earth's shadow on part of every orbit — its power comes and goes. In exchange it is far cheaper to launch into, can be packed in enormous numbers, and concentrates over the populated low and mid latitudes; the filing frames it as capacity for demand peaks. Crucially, its visibility depends on your latitude: within ±30° (the tropics) the satellites pass directly overhead across the whole sky; toward higher latitudes they sit lower in the equator-facing sky; and a 550 km satellite never rises at all above about 53° latitude (a 2,000 km one, above about 70°).
Try the same date and time from Quito (overhead passes) versus Ithaca (a low southern band) versus Fairbanks (little or nothing) to see the latitude dependence directly.
The display shows the real night sky for your location, date and time — about 8,400 catalogue stars, the visible planets, and the Moon with its correct phase — with the datacenter swarms overlaid. The two shells are tinted differently only so you can tell them apart: warm white for sun-synchronous, cool blue for the 30° shell (the real satellites would all look the same near-white).
Both shells are propagated as circular orbits at the altitude you set, with a proper cylindrical Earth-shadow test deciding whether each satellite is sunlit, and full topocentric geometry converting positions to your local altitude/azimuth.
Apparent magnitude is anchored to the peer-reviewed estimate by Marcy (MNRAS, 2026): a 5 GW datacenter with a ~4 km × 4 km solar array shines near magnitude −6 (his range, −5 to −7) at ~550 km. From that anchor:
√P; angular size is √area / range.One honest limitation: this diffuse-sphere term is direction-tolerant and ignores that a real array is a flat, strongly directional panel — brightest seen face-on, nearly vanishing edge-on. That smooths over a genuine difference between the shells (the 30° shell exposes observers to bright, near-face-on geometry the dawn–dusk shell mostly avoids). The base brightness here is therefore a fair guide to where and when, and a conservative one for the 30° shell's peak brightness.
A flat array is partly mirror-like, so when its specular reflection of the Sun briefly lines up with you, it can flash far brighter than its steady glow — the old Iridium-flare effect. With the flare toggle on, each satellite is given a specular glint on top of its diffuse brightness.
Use it for intuition only. The flare layer answers "if a surface this reflective lined up, how bright and how often?" — not "how bright will these actually flare," which no public data can yet support.