Orbital AI Data Centers — Economic Competitiveness Timeline > Sources > Orbital Eclipse/Shade Model Audit for a Proposed LEO “Orbital AI Datacenter”

Orbital Eclipse/Shade Model Audit for a Proposed LEO “Orbital AI Datacenter”

Report from ChatGPT Pro
Prepared in ChatGPT Pro using GPT-5.4 Pro.
Date: 2026-03-25

A. Executive summary

A standard dawn-dusk Sun-synchronous orbit at about 575 km is not year-round eclipse-free. In the standard aerospace sense, “dawn-dusk SSO” means a near-constant 06:00 or 18:00 mean local time of node crossing, not a plane that stays exactly perpendicular to the true Sun vector at all times. In that standard case, a circular ~575 km orbit still develops a single annual eclipse season about 3 months long, with a peak eclipse of about 21 minutes per orbit, about 95 eclipse days per year, and an annual sunlight fraction of about 95.4% in a first-principles cylindrical-shadow model. Real mission documents for similar dawn-dusk SSOs describe them as eclipse-free for roughly 8–9 months per year, not eclipse-free all year. (NASA NTRS, 2008)

For the quoted numbers:

Those numbers are right when “dawn-dusk SSO” means near-circular SSO, LTAN/LTDN ≈ 06:00/18:00, referenced to the mean Sun, with ordinary seasonal solar declination. They are wrong if someone silently assumes the stronger condition that the orbit normal remains exactly aligned with the true Sun all year; that idealized condition would indeed be eclipse-free, but it is not what an ordinary Earth SSO at 500–650 km provides. (NASA NTRS, 2008)

Direct answers

B. Independent derivation

1) Orbit family and assumptions

I model the candidate as a near-circular LEO SSO with altitude h = 500–650 km, eccentricity e ≈ 0, and a fixed mean local time of ascending node near 06:00 or 18:00. In standard mission design, SSO means choosing a and i so that the J2 nodal precession matches the mean Sun’s apparent motion of about 0.9856 deg/day. NASA defines MLTAN relative to the mean Sun, not the true/apparent Sun. (NASA NTRS, 2008)

The standard J2 nodal-precession relation is

Ω̇ = -(3/2) J2 n (Re/a)^2 cos(i) / (1 - e^2)^2
n = sqrt(μ / a^3)

and SSO sets Ω̇ ≈ +0.985647°/day for a retrograde orbit. Using Re = 6378.137 km, μ = 398600.4418 km^3/s^2, and e = 0, the required inclinations are approximately:

The orbital period is

T = 2π sqrt(a^3 / μ)

giving about 94.6 min, 96.2 min, and 97.7 min respectively. These are calculated from the standard SSO/J2 relation used in NASA mission-design references. (NASA Orbit Primer)

2) Beta angle and why dawn-dusk is special

Let the orbit-normal unit vector be ĥ and the Sun unit vector be ŝ. The beta angle β is the angle between the Sun vector and the orbit plane; equivalently, it is 90° minus the angle between the Sun vector and the orbit normal. NASA’s thermal-design notes use this same geometry. (NASA TFAWS beta-angle notes)

With right ascension of ascending node Ω, inclination i, solar right ascension α, and solar declination δ,

sin β = sin i cos δ sin(Ω - α) + cos i sin δ

For a standard dawn-dusk orbit, Ω - α is near ±90°, corresponding to LTAN ≈ 18:00 or 06:00. That is why dawn-dusk maximizes |β|, while noon-midnight SSO (Ω - α ≈ 0 or 180°) drives β toward zero and gives the worst eclipses. NASA mission-design literature explicitly describes these as “twilight” orbits selected to maximize solar-cell illumination. (NASA NTRS, 1974)

For the 18:00 ascending-node branch, Ω - α ≈ +90°, so

sin β = sin i cos δ + cos i sin δ = sin(i + δ)

A very useful simplification follows if you define the orbit normal’s declination as

δh = 90° - i

Then for exact 18:00 dawn-dusk,

β = 90° - |δ - δh|

This is the key point: in a real SSO, δh is fixed by the orbit inclination and is about -7.4° to -8.0° in this altitude band, while the Sun’s declination δ still swings between ±23.44° over the year. So even an exact 06:00/18:00 SSO does not keep β = 90° year-round. The seasonal minimum is

βmin = 180° - i - ε

where ε ≈ 23.44° is Earth’s obliquity. Numerically:

So the orbit is nowhere near a permanent β = 90° geometry. That is the fundamental reason seasonal eclipses persist. NASA’s thermal notes also emphasize that β varies with season and with orbit precession. (NASA TFAWS beta-angle notes)

3) Earth-shadow geometry and eclipse duration

For LEO mission design, a very common first-order model is a cylindrical umbra and a circular orbit. NASA explicitly notes that this is a good approximation near Earth because the Earth’s umbra is about 106 Earth diameters long, so its width changes very little in the vicinity of Earth. (NASA NTRS, 1963)

Let r = Re + h. The no-eclipse threshold follows directly from the geometry:

|β| ≥ β* = arcsin(Re / r)

Equivalently, in the notation used in NASA’s 1974 design report, if η is the angle between the Sun line and orbit normal, then the continuous-sunlight condition is η ≤ ηc with

cos ηc = Re / (Re + h)

which is the same result because |β| = 90° - η. (NASA NTRS, 1974)

The critical beta angles are:

Since βmin is only about 59°, and β* is still 65–68° in this altitude band, eclipse is unavoidable for part of the year. In fact, solving βmin = β* in this same simple model gives an altitude of about 1390 km before an exact dawn-dusk SSO would become year-round eclipse-free. That is far above the proposed 500–650 km band.

For |β| < β*, the eclipse half-angle in the orbit is

ψ = arccos( sqrt(1 - (Re/r)^2) / cos β )

and total eclipse duration is

te = (T / π) ψ

That produces the familiar behavior: eclipse goes to zero as |β| → β*, and is longest when β → 0. This is the same low-circular-orbit cylindrical-shadow geometry used in NASA thermal-analysis material. (NASA TFAWS beta-angle notes)

4) Eclipse season length

For the exact 18:00 ascending branch, eclipse begins when β falls below β*, which is equivalent to the Sun declination exceeding

δth = (90° - i) + arccos(Re / r)

Because δ = arcsin(sin ε sin λ), where λ is solar ecliptic longitude measured from the vernal equinox, the eclipse-season length is

Lseason = (Y / 2π) [ π - 2 arcsin( sin δth / sin ε ) ]

with Y = 365.2422 d. The 06:00 ascending branch has the same season length and peak eclipse, just shifted by about half a year. The annual sunlight fraction is then obtained by integrating te(β(λ)) over the year.

Using that model, the central results are:

Altitude SSO inclination Period Critical β* Dawn-dusk βmin Max eclipse Eclipse season Annual sunlight
500 km 97.40° 94.62 min 68.02° 59.16° 22.66 min 103.0 d 94.56%
575 km 97.69° 96.17 min 66.54° 58.87° 21.17 min 95.1 d 95.40%
650 km 97.99° 97.73 min 65.16° 58.57° 19.73 min 87.7 d 96.13%

These are the baseline audit numbers for this orbit family. The source-backed assumptions are the standard SSO/J2 condition, mean-Sun local-time definition, and cylindrical-shadow geometry. (NASA NTRS, 2008)

5) Sensitivity to local-time offsets

This family is very sensitive to local time. At 575 km:

So the “dawn-dusk” choice really is the right SSO local-time family if solar availability is important. It is just not a no-eclipse one.

There is a smaller second-order issue: MLTAN is tied to the mean Sun, while the apparent Sun can differ from the mean Sun by up to 16 minutes, or about 4 degrees on the sky. That changes the 575 km eclipse result only modestly: on the order of a few tenths of a minute in peak eclipse and a few days in season length. It does not eliminate the seasonal eclipses. (U.S. Naval Observatory, equation of time)

C. Comparison to external sources and real missions

The best cross-check is to compare the model against real dawn-dusk missions.

TerraSAR-X

TerraSAR-X is almost a direct check on the low end of the candidate band: DLR reports a 514.8 km, 18:00 ± 0.25 h ascending-pass dawn-dusk SSO, and a later battery-operations paper reports an eclipse season from April 29 to August 14 with maximum duration 23 min around the June solstice. The central model for a 514.8 km exact dawn-dusk orbit gives about 22.3 min max eclipse and about 101 days of eclipse season. That is very close. (DLR TerraSAR-X mission status, DLR battery operations paper)

Sentinel-1

Sentinel-1 is a strong mid/high-altitude check. ESA/Copernicus lists 693 km, 18:00 ascending node, 98.6 min period, and max eclipse duration 19 min. The same model at 693 km gives about 18.9 min, essentially a direct hit. (Copernicus Sentinel-1 mission document)

MicroSCOPE

MicroSCOPE chose a dawn-dusk SSO at ~710 km, RAAN ~18 h. Its mission paper states the orbit is fully sunlit except for roughly three months, from May 9 to August 4, when eclipses occur. The same model in that altitude range gives an eclipse season of about 83–87 days, which is consistent with the reported window. (MicroSCOPE mission paper)

PROBA-2

PROBA-2 is a good higher-altitude dawn-dusk reference. ESA lists semi-major axis 7135 km (about 757 km altitude), 98.445° inclination, 06:00 ± 1 min MLTAN, and says the chosen orbit is eclipse-free for more than nine months each year, with eclipses rising to a maximum of about 20 minutes in the eclipse season. A mission paper gives similar design targets: 700–800 km, LTAN 6:00 ± 15 min, targeted to be eclipse-free for 9 months/year with max eclipse <20 min. The model at 757 km gives about 17.7 min max eclipse and about 287.5 eclipse-free days/year. Again, that is consistent. (ESA PROBA-2 spacecraft page, ILRS/ESA paper)

IRIS

IRIS is especially relevant because it sits close to the proposed family. NASA describes it as a sunrise-line polar SSO at about 620 × 670 km, giving eight months of continuous observations per year and maximizing eclipse-free solar viewing. That is exactly the kind of wording that fits this orbit class: long eclipse-free seasons, not year-round eclipse freedom. (NASA IRIS brochure)

Cross-check conclusion

Taken together, the real-mission evidence strongly supports the first-principles conclusion: dawn-dusk SSO in this altitude regime is a high-sunlight orbit with one seasonal eclipse season, typically about 18–23 minutes maximum eclipse and about 8–9 months eclipse-free, depending on altitude and local-time control. (DLR battery operations paper)

D. Error analysis

  1. The biggest modeling mistake is equating “dawn-dusk” with “eclipse-free.”
    In ordinary Earth mission design, dawn-dusk SSO means fixed MLTAN/LTDN relative to the mean Sun, not a plane whose normal stays exactly aligned with the true Sun. The latter idealization would be eclipse-free; the former is not. NASA’s own definitions and real mission documents make that distinction clear. (NASA NTRS, 2008)

  2. A standard dawn-dusk SSO still has substantial seasonal beta-angle motion.
    The orbit normal declination is only about -8°, while solar declination reaches ±23.44°. That forces β down to about 59° at one solstice, below the 65–68° no-eclipse threshold in 500–650 km LEO. This is the core physical reason seasonal eclipses remain. NASA’s thermal notes also explicitly state that beta angle varies with seasons and orbit precession. (NASA TFAWS beta-angle notes)

  3. Mean Sun vs true Sun is a real assumption, but not the dominant one.
    MLTAN is defined with respect to the mean Sun, while the apparent Sun can differ by as much as 16 minutes or about 4°. That changes the exact timing and slightly changes the eclipse extrema, but it does not turn a 575 km dawn-dusk orbit from eclipsed into eclipse-free. (NASA NTRS, 2008)

  4. Local-time tolerance matters and should be stated explicitly.
    PROBA-2’s design literature, for example, distinguishes 06:00 ± 15 min targeted control from a wider ±45 min acceptable band without propulsion. In the 575 km model, ±15 min only nudges the peak eclipse to about 21.4 min, but ±45 min pushes it to about 23.0 min and lengthens the eclipse season to about 113 days/year. So a claim like “20 min max” is only right if the local-time control assumption is also right. (ILRS/ESA paper)

  5. “Eclipse days per year” is ambiguous.
    In this orbit family it means roughly one eclipse on every orbit during the eclipse season. For a 575 km orbit with a 96.2 min period and about 95 eclipse days, that is roughly 1,425 eclipsed orbits/year, not merely “95 isolated events.” If someone uses “100 days” as though it implied a small number of rare outages, that is misleading.

  6. Annual sunlight fraction is not the correct power-design driver.
    A 95% annual sunlight fraction sounds benign, but battery and array sizing are driven by the worst orbit in eclipse season, not the annual average. A design that sizes power around annual-average illumination instead of the ~20–23 minute worst-case eclipse will be underbuilt.

  7. The cylindrical-shadow approximation is acceptable here, but it is still an approximation.
    NASA treats it as valid near Earth because the Earth’s umbra is very long and changes little in width near Earth. In this altitude range, the finite solar angular size shrinks the Earth’s true umbra radius by only about 2–3 km, which changes the worst-case eclipse duration by only a few seconds. So cylindrical-shadow error is not what will decide this trade. (NASA NTRS, 1963)

  8. Circular orbit is another hidden assumption.
    Real missions in this family usually have small but nonzero eccentricity. That slightly changes the period, critical beta, and eclipse duration, but usually at the level of seconds to perhaps a few tens of seconds, not many minutes, unless eccentricity is intentionally significant.

  9. “Lower altitude = very low radiation” is only partly true.
    Lower altitude helps, but SSO’s high inclination is the larger environmental fact: it crosses the SAA and high-latitude trapped-particle regions. NASA’s radiation study uses 500 km, zero inclination as its benign LEO case at only about 10 rad/year, while 500 km, 51° is about 200 rad/year, and 500 km polar is described as highly similar to that ISS-like case, with more electron activity near the poles. So dawn-dusk SSO is low-radiation compared with MEO/GEO, but it is not the minimum-radiation part of LEO. (NASA radiation modeling study)

  10. Ignoring AO and drag can bias the altitude trade.
    NASA notes that between 180 and 650 km, atomic oxygen is the most abundant species and erodes susceptible materials; drag also becomes more important as altitude drops. At the extreme low end, ESA’s GOCE at about 255 km needed electric propulsion continuously to compensate drag. A giant-array compute platform at 500 km will not be GOCE-bad, but it will feel drag and AO more than one at 650 km. (NASA atomic oxygen review)

E. Engineering impact

1) Battery sizing impact

For a continuous spacecraft load P, the raw eclipse energy that must be supplied is

Eraw = P × te,max

Per 1 kW of continuous load, the raw worst-case eclipse energy is:

That is only the ideal usable energy. With a realistic battery design allowance, for example 80% depth of discharge and about 90–95% electrical efficiency through discharge/conversion, the required nameplate battery energy becomes roughly:

For early mass modeling, 0.5 kWh per kW continuous load is a good central battery-sizing input for a ~575 km dawn-dusk SSO.

NASA’s power state-of-the-art notes that Li-ion cells are well above 150 Wh/kg, with current top energy cells around 265–276 Wh/kg, but complete batteries are lower because of BMS, interconnects, and thermal hardware. Using a conservative pack-level assumption of roughly 120–180 Wh/kg, consistent with those NASA cell-level figures plus packaging overhead, the battery mass works out to about 2.5–4.3 kg per kW continuous load. (NASA power state of the art 2021)

That implies roughly:

This is before extra margin for lifetime, redundancy, cold-temperature derating, or high-cycle aging.

2) Solar-array sizing impact

During eclipse season the solar array must both:

  1. run the spacecraft during the sunlit part of the orbit, and
  2. recharge the battery before the next eclipse.

If ts = T - te is sunlit time and ηrt is round-trip battery efficiency, then the required sunlit array power is approximately

Parray,sun ≈ P × (1 + te / (ηrt × ts))

Using ηrt ≈ 0.9, the required sunlit power factor at season peak is about:

So a platform that must deliver 1 MW continuous electrical load through the worst part of the eclipse season needs about 1.28–1.35 MW of usable array output during sunlit arcs before normal engineering margins. Once degradation, cosine loss, temperature loss, regulation margin, and contingency are added, a practical first-cut design factor is more like 1.4–1.5 kW of array output per 1 kW continuous load.

NASA’s recent power chapter shows current space missions clustered around roughly 30 W/kg solar-array specific power, with the empirical high end around 200 W/kg, while an Aerospace Corporation review gives ~40 W/kg for current state-of-the-art arrays and 115–130 W/kg for next-generation concepts. On that basis, the array mass per 1 kW continuous load is roughly:

So:

That already swamps the battery mass. For a solar-powered orbital compute platform, the array mass penalty is larger than the battery mass penalty, even in dawn-dusk SSO. (NASA power state of the art 2024)

3) Operational consequences for continuous AI inference

A 575 km dawn-dusk SSO has a period of about 96.2 minutes, so it makes about 15 orbits/day. During the roughly 95-day eclipse season, essentially every orbit contains an eclipse, so the platform sees on the order of 1,400 battery cycles per year from eclipse service alone. That is not catastrophic for Li-ion, but it is not negligible either.

Operationally, this means:

Real missions in this orbit class reflect that. TerraSAR-X/TanDEM-X explicitly discuss the operational consequences of their recurring 23-minute eclipse season on Li-ion battery management. Sentinel-1, despite its dawn-dusk SSO and only 19-minute maximum eclipse, still carries a substantial 324 Ah battery and 5.9 kW EOL solar-array capability. (DLR battery operations paper)

4) Thermal impact

“More sunlight” is not a free thermal win. ESA notes that PROBA-2’s permanent Sun-pointing geometry produced a large thermal gradient that made purely passive thermal control difficult. For a high-power compute spacecraft, the constant solar load plus internally dissipated heat means radiator placement and view factors may become as important as eclipse avoidance. This audit does not include radiator sizing, but for an “orbital AI datacenter” it is likely another first-order mass driver. (ESA PROBA-2 spacecraft page)

Radiation.
LEO below 1000 km is relatively benign compared with higher orbits, but SSO is not the minimum-radiation LEO case. NASA’s 500 km examples show a zero-inclination orbit as the benign case at about 10 rad/year, while inclined and polar orbits encounter the SAA and polar trapped-particle regions. So lowering from 650 to 500 km helps somewhat, but the polar inclination is the bigger reason SSO is not ultra-benign. (NASA radiation modeling study)

Atomic oxygen and drag.
NASA’s AO review states that atomic oxygen is the dominant atmospheric species from 180 to 650 km, exactly the candidate band. Lower altitude improves disposal and reduces some radiation exposure, but increases AO erosion and drag. NISAR’s dawn-dusk SSO operations explicitly budget drag make-up maneuvers and inclination adjust maneuvers, and GOCE at very low altitude needed continuous electric propulsion to counter drag. Large power arrays magnify this trade because they drive up cross-sectional area and disturbance torques. (NASA atomic oxygen review, NISAR mission description)

Deorbit and debris.
NASA’s deorbit review says 400 km naturally decays well within 25 years, while beyond 800 km natural 25-year decay is not guaranteed. A CNES/STELA-based RemoveDEBRIS study gives a more specific SSO illustration: about 5–6 years lifetime at 500 km SSO, 14–15 years at 550 km, and about 26–27 years at 600 km, noting that compliance above 600 km is not guaranteed without propulsion or drag augmentation. That makes 650 km attractive for eclipse reduction but weaker for passive disposal. (NASA deorbit state of the art)

Communications.
Dawn-dusk polar SSO does not inherently provide continuous cloud-like connectivity. It provides frequent high-latitude opportunities and works well with polar ground stations, but continuous low-latency service would still need a relay architecture or large gateway network. NISAR, for example, plans 30–45 minutes of data downlink per orbit at 3.5 Gbps through polar ground stations—excellent for EO, but still a scheduled store-and-forward model, not continuous interactive service. (NISAR mission description)

For a near-circular 500–650 km dawn-dusk SSO intended for solar-powered continuous service, the recommended planning ranges are:

Case Assumptions Max eclipse Eclipse days/year Annual sunlight Battery sizing input
Optimistic ~650 km, LTAN/LTDN within ±15 min of 06:00/18:00 19.5–20.5 min 85–90 96.0–96.2% 0.43–0.48 kWh per kW
Central ~575–600 km, exact or near-exact dawn-dusk 20.5–21.5 min 92–97 95.4–95.7% 0.46–0.52 kWh per kW
Pessimistic ~500–575 km, or LTAN off by ±30–45 min 22.5–23.5 min 100–115 94.0–94.8% 0.50–0.58 kWh per kW

Orbit recommendation

The orbit recommendation depends on which constraint dominates:

The one recommendation this audit rejects is the claim that “575 km dawn-dusk SSO is basically eclipse-free.” It is not. It is the best practical high-sunlight SSO family in LEO, but for power-system design it should be treated as a ~3-month annual eclipse-season orbit with ~20–23 minute peak eclipses, not as continuous-solar.

Appendix A. Prompt

Research the orbital eclipse/shade model for a proposed LEO “orbital AI datacenter” and produce a rigorous technical audit focused on factual and analytical correctness.

Context
We are evaluating a high-power satellite or satellite cluster intended to run AI inference workloads in low Earth orbit. The design goal is to choose an orbit that is:

- low enough in altitude to reduce radiation exposure and related hardware-reliability risks,
- but also configured to minimize eclipse time, so solar power availability is as continuous as possible and battery mass is minimized.

The concept is most interested in very low-radiation, high-sunlight LEO options, especially dawn-dusk sun-synchronous orbits, roughly in the 500–650 km altitude range. A representative candidate is around 575 km in a dawn-dusk SSO near 06:00/18:00 local time.

Task
I want a deep red-team review of the eclipse properties of this orbit family and whether a dawn-dusk SSO at roughly this altitude is actually close to eclipse-free.

Questions to answer
- For a true dawn-dusk sun-synchronous orbit around 500–650 km, what eclipse behavior should we expect?
- Should such an orbit be effectively eclipse-free, nearly eclipse-free, or can it still have substantial seasonal eclipses?
- Are values like “~20-minute maximum eclipse,” “~100 eclipse days per year,” or “~95% annual sunlight fraction” physically plausible for this kind of orbit?
- Under what exact assumptions would those numbers be right or wrong?

What to do
1. Reconstruct the orbital problem from first principles
- Define the orbit family precisely: altitude, inclination, local time of ascending/descending node, beta angle, Earth shadow geometry, and sun-synchronous precession.
- Explain how dawn-dusk SSO differs from other SSO local times.
- Show the equations for eclipse duration as a function of altitude and beta angle.
- Compute or approximate:
  - orbital period,
  - maximum eclipse duration,
  - annual eclipse season length,
  - annual sunlight fraction,
  - how these vary across 500, 575, and 650 km if useful.

2. Verify against authoritative sources
Use primary or highly authoritative sources wherever possible:
- NASA, ESA, JAXA, CNES
- peer-reviewed papers
- astrodynamics textbooks / university orbital mechanics notes
- credible aerospace mission-design references
Cite every important factual claim.

3. Compare against real missions
Use real dawn-dusk or near-dawn-dusk SSO missions as checks.
I want concrete examples of missions in similar altitudes/inclinations and what their eclipse behavior is reported to be.

4. Quantify engineering implications
Estimate how the corrected eclipse model affects system design for a solar-powered orbital compute platform, especially:
- required battery energy and battery mass,
- solar array sizing,
- power-system mass,
- total spacecraft mass,
- operational consequences for continuous AI inference service.

You do not need a full spacecraft design, but I do want order-of-magnitude implications.

5. Check related tradeoffs
Also assess whether the “best” orbit is really dawn-dusk SSO once these constraints are considered:
- radiation exposure / total ionizing dose,
- atomic oxygen and drag at lower altitudes,
- constellation operations and station-keeping,
- deorbit and debris considerations,
- communications implications if they materially interact with orbit choice.

Output format
A. Executive summary
- direct answer on whether a ~575 km dawn-dusk SSO is close to eclipse-free
- direct answer on whether ~20 min max eclipse / ~100 eclipse days / ~95% sunlight are plausible

B. Independent derivation
- equations, assumptions, intermediate steps, computed values

C. Comparison to external sources and real missions

D. Error analysis
- every factual or modeling mistake, ambiguity, or hidden assumption you find in the eclipse reasoning

E. Engineering impact
- how corrected eclipse assumptions would change battery sizing, solar sizing, and total system mass

F. Recommended parameter ranges
- reasonable central / pessimistic / optimistic values for:
  - max eclipse duration
  - eclipse days per year
  - annual sunlight fraction
  - battery sizing inputs for this orbit family

Requirements
- Browse thoroughly.
- Prefer primary sources.
- Show the math.
- Be explicit about uncertainty.
- Do not just provide a generic overview of SSO; make this a technical audit aimed at selecting the best low-radiation, low-eclipse orbit for an orbital AI datacenter.
- Be skeptical and resolve contradictions rather than smoothing them over.