Orbital Compute Total Cost of Ownership

What is the amortized TCO per kW_IT/year for orbital compute?

What is the amortized total cost of ownership per kW_IT per year ($/kW_IT/year) for orbital AI compute?

Answer

Orbital TCO ranges from 6,614 $/kW_IT/year (optimistic, 2040) to 162,138 $/kW_IT/year (conservative, 2026). The central estimate declines from 51,619 $/kW_IT/year in 2026 to 16,367 $/kW_IT/year by 2040, driven primarily by declining launch costs reducing the amortized capex component.

The TCO has two components: amortized capex (orbital_capex × CRF) and fixed annual opex ($100-$400/kW_IT/year for ground stations, operations, and regulatory costs), divided by orbital availability (the fraction of the year the satellite delivers compute, accounting for eclipse downtime net of battery ride-through) to express cost per operating kW_IT-year. This availability adjustment accounts for eclipse downtime — the model selects the cost-optimal battery sizing for each scenario and year, balancing battery mass (which increases capex) against lost operating time (which reduces delivered compute). By amortizing capex over a shorter effective lifetime (optimistic: 5.9 years, central: 3.8 years, conservative: 2.2 years) rather than physical lifetime, the model captures the cost impact of failures, degradation, and deployment delay without needing a separate replacement opex line item.

Inputs

Analysis

TCO Formula

Orbital TCO = (orbital_capex × CRF(orbital_wacc, effective_lifetime) + orbital_fixed_opex) / availability

Where CRF is the Capital Recovery Factor — the annuity factor that converts capex into an equivalent annual cost accounting for the cost of capital. At the central 2026 orbital WACC of 13.5% and 3.8-year effective lifetime, CRF = 0.35 (vs 0.26 for simple 1/n amortization). The WACC declines over time (central: 13.5%→10% by 2040) as operational history accumulates, which progressively reduces the CRF. See the orbital WACC page for the derivation, including the double-counting adjustment that reduced the central from a naive 15% to 13.5%.

The effective lifetime is a capacity-weighted measure that accounts for both physical degradation (radiation damage to solar cells, thermal cycling fatigue) and failure-driven capacity loss. The availability divisor (from the eclipse duration analysis) adjusts for eclipse downtime so that both orbital and terrestrial TCOs express cost per operating kW_IT-year on a comparable basis.

Amortized Capex Component

The amortized capex declines steeply in the early years as launch costs fall, then flattens as GPU and platform costs become dominant. By 2040, the optimistic amortized capex reaches 6,514 $/kW_IT/year, while the central case settles at 16,167 $/kW_IT/year.

The conservative scenario is penalized twice: higher capex (more expensive launch, heavier satellites, costlier platform) AND shorter effective lifetime (2.2 years vs 3.8 or 5.9). This double penalty pushes conservative amortized capex to 159,422 $/kW_IT/year in 2026 -- nearly 10x the optimistic figure.

Fixed Opex: A Minor Component

Orbital fixed opex covers ground station operations, fleet management, regulatory compliance, and spectrum licensing. These costs are small relative to amortized capex:

• Optimistic: $100/kW_IT/year
• Central: $200/kW_IT/year
• Conservative: $400/kW_IT/year

Fixed opex represents only 1-3% of total orbital TCO across all scenarios and years. The cost of failure-driven satellite replacement -- previously modeled as a large opex item -- is now captured in the effective lifetime parameter, which reduces the number of capacity-years each satellite delivers and thereby increases the amortized capex per year.

This restructuring is more analytically sound: failure-driven replacement is fundamentally a capital recycling cost (manufacturing and launching new satellites to maintain fleet capacity), not an operational expense. By folding it into effective lifetime, the model correctly represents it as a capex amortization penalty rather than an ongoing operating cost.

Total TCO Trajectories

The optimistic trajectory falls 46% from 13,015 (2026) to 6,614 (2040), with most of the decline concentrated in 2026-2030 as launch costs collapse. The pace of improvement slows after 2030 as the TCO approaches a floor set by amortized GPU and platform costs over the effective lifetime. The central trajectory shows a 66% decline over the period, from 51,619 to 16,367. The conservative trajectory declines but remains extremely high in absolute terms due to the combined penalty of expensive hardware, heavy satellites, and a short 2.2-year effective lifetime.

Key Structural Observations

1. Orbital TCO is dominated by amortized capex, amplified by cost of capital. Because fixed opex is small ($100-$400/kW_IT/year), nearly all of orbital TCO comes from the CRF-weighted amortization of capex. The CRF amplifies the lifetime effect: at 13.5% WACC (2026) and 3.8-year lifetime, CRF is 0.35 (vs 0.26 for simple amortization). By 2040, WACC compression to 10% reduces the CRF to 0.33.

2. Orbital TCO has a floor set by amortized GPU and platform costs, which itself declines over time. In the optimistic 2040 scenario with near-zero launch costs and 8% WACC, the TCO floor is approximately ($27,000 + $2,360) × CRF(0.08, 5.9) + $100 ≈ $6,508/kW_IT/year. The 2040 optimistic value of 6,614 is close to this floor, confirming that further launch cost reduction yields negligible improvement.

3. The effective lifetime penalty is the primary remaining cost driver. Terrestrial GPU hardware is depreciated over 4-6 years at 5-10% WACC (central: 7%), but orbital compute hardware delivers only 2.2-5.9 effective capacity-years at higher WACC. The combination of shorter lifetime and higher cost of capital creates a compounding annual capex penalty, partially offset by WACC compression over time.

4. Extending effective lifetime is the highest-impact path to cost reduction. Improving effective lifetime from 3.8 to 5.9 years reduces the CRF from 0.35 to ~0.23 at 2026 central WACC, cutting annual amortized capex by ~35%. This remains the single most impactful parameter change, as confirmed by the OAT sensitivity analysis.

5. Availability is an annual average; seasonal variation is not modeled. The availability divisor uses the annual-average eclipse downtime, but eclipses are concentrated in a ~30-150 day seasonal window around the solstice (see eclipse duration analysis). For batch and deferrable workloads, annual-average availability is the correct cost metric. For always-on latency-sensitive products, seasonal clustering of downtime may impose a quality-of-service penalty not captured by the annual average.