Cost Parity Timeline

When (if ever) does orbital compute TCO reach parity with terrestrial?

Answer

Orbital compute does not reach cost parity with terrestrial compute in any scenario within the 2026-2040 analysis window. Even the optimistic scenario -- combining the most favorable assumptions for orbital (low launch costs, lightweight satellites, cheap platform manufacturing, low opex) with the most favorable assumptions for orbital's comparison case (high terrestrial energy costs) -- shows the TCO ratio bottoming out at approximately 1.5 around 2035, meaning orbital remains ~48% more expensive than terrestrial at best.

The central scenario shows orbital compute at 2.3x terrestrial cost by 2040, and the conservative scenario at 3.5x. The ratio declines over time as launch costs fall, but converges toward a persistent floor driven by orbital opex and the GPU cost premium.

Beyond cost, orbital compute is constrained to Tier 1 and Tier 2 inference workloads (models up to ~405B parameters that fit within a single satellite or small cluster). Frontier MoE models requiring wide expert parallelism across 64+ GPUs cannot be served across satellites with current or near-term inter-satellite link technology.

Inputs

Analysis

The TCO Ratio Over Time

The optimistic ratio drops rapidly from 1.9 in 2026 to 1.5 by 2030, then essentially plateaus -- reaching 1.5 by 2035 and actually ticking up slightly to 1.5 by 2040. This uptick occurs because terrestrial energy costs continue falling in the optimistic scenario while orbital opex is constant, causing the terrestrial baseline to improve faster than orbital in the late 2030s.

The central ratio follows a similar pattern with a slower decline, reaching 2.3 by 2040 -- orbital remains 2.3x more expensive than terrestrial.

Why Parity Is Never Reached: The Structural Cost Gap

The failure to reach parity is not a marginal miss that better assumptions could fix. It reflects three structural economic realities:

1. GPU cost is common to both sides and dominates both TCOs.

GPU hardware represents 74% of central terrestrial TCO and approximately 50-60% of central orbital TCO (once GPU cost and GPU-driven replacement costs are combined). Since GPU cost per kW_IT is essentially the same in both deployment contexts (plus a modest 5-30% orbital premium), this dominant shared cost component cannot create a cost advantage for either side. The competition between orbital and terrestrial reduces to a comparison of their non-GPU costs -- and the non-GPU cost gap strongly favors terrestrial.

2. Orbital energy savings are small in absolute terms.

The primary theoretical advantage of orbital compute is free solar power, eliminating electricity costs. But terrestrial energy cost is only 723 $/kW_IT/year (central, 2026) -- roughly 8% of total terrestrial TCO. Even eliminating this entirely saves only ~$700-$770/kW_IT/year. This saving must offset:

• Orbital opex of $9,800/kW_IT/year (central) -- predominantly failure-driven replacement cost that has no terrestrial equivalent
• Platform manufacturing amortization of $2,400/kW_IT/year (central, over 5 years)
• Launch cost amortization (declining but still $369/kW_IT/year in central 2040)

The ~$700/kW_IT/year energy savings is overwhelmed by the ~$12,600/kW_IT/year in orbital-specific costs. Orbital compute does eliminate cooling costs too (these are captured in the zero-PUE-overhead for space), but modern terrestrial PUE of 1.10 means cooling overhead is only ~5% of power -- a trivial advantage.

3. Failure-driven replacement creates a persistent opex floor.

The inability to perform in-orbit repairs means that every hardware failure (GPU, bus, power system) requires manufacturing and launching an entire replacement satellite. At a central estimate of 8% annual satellite attrition, this drives $6,500/kW_IT/year in replacement costs alone. This single line item is roughly 9x the annual terrestrial energy cost it replaces. Until in-orbit servicing becomes viable or satellite failure rates drop dramatically, this asymmetry persists.

Sensitivity: What Would Need to Change for Parity?

To explore what parameter shifts could theoretically achieve parity, consider the central 2040 gap:

• Orbital TCO: 20,044 $/kW_IT/year
• Terrestrial TCO: 8,758 $/kW_IT/year
• Gap: ~$11,286/kW_IT/year that must be closed

The most sensitive levers:

Orbital opex reduction (highest impact). Reducing orbital opex from $9,800 to $0 (impossible, but illustrative) would cut orbital TCO to ~$10,244, still above terrestrial's $8,758. Even eliminating ALL orbital opex does not reach parity because amortized capex alone exceeds terrestrial TCO. To reach parity, opex would need to go negative -- i.e., orbital would need to generate revenue from some source other than compute to subsidize its cost.

Terrestrial energy cost increase (moderate impact). If terrestrial energy cost doubled from $0.07 to $0.14/kWh, terrestrial TCO would rise by ~$675/kW_IT/year to ~$9,432 -- still far below orbital's $20,044. Energy cost would need to rise to approximately $1.50/kWh (20x current levels) to close the gap, which is physically implausible.

GPU cost premium elimination (low impact). Removing the 15% orbital GPU premium saves only ~$975/kW_IT/year in amortized capex. This is helpful but minor.

Orbital lifetime extension (moderate impact). Extending satellite life from 5 to 10 years would halve amortized capex from ~$10,244 to ~$5,122/kW_IT/year, reducing orbital TCO to ~$14,922. This is a larger improvement but still well above terrestrial TCO, and a 10-year satellite life exceeds FCC deorbit rules and GPU obsolescence constraints.

Launch cost (negligible at this point). By 2040, launch cost is already 1,846 $/kW_IT -- less than 4% of total capex. Reducing it to zero saves only ~$369/kW_IT/year.

The analysis confirms that no single parameter change achieves parity. Multiple simultaneous improvements would be needed: dramatically lower opex (through much lower failure rates AND much cheaper replacement), longer satellite lifetimes, AND lower platform costs. Even combining all three optimistic-direction shifts, the floor of GPU cost + some irreducible opex keeps orbital above terrestrial.

The Networking Constraint: Workload Limitation

Even if cost parity were achieved, orbital compute faces a workload feasibility constraint. The inference-networking-requirements analysis identifies three tiers:

Tier 1 (single-satellite, high feasibility): Models up to ~70B parameters with quantization, running on 1-8 GPUs within one satellite. No inter-satellite networking needed. This covers most practical inference workloads today, including distilled frontier models.

Tier 2 (small-cluster, moderate feasibility): Models up to ~405B parameters distributed across 2-4 satellites via pipeline parallelism, using 800 Gbps+ optical inter-satellite links. Adds latency but achievable for throughput-oriented workloads.

Tier 3 (wide-EP, low feasibility): Frontier MoE models (DeepSeek R1, Kimi K2) requiring 64+ GPU NVLink domains for wide expert parallelism. The ~2,000x bandwidth gap between NVLink (1.8 TB/s per GPU) and demonstrated optical ISLs (800 Gbps per link pair) makes this infeasible with current or near-term technology.

This means orbital compute is limited to serving models that are 1-2 generations behind the terrestrial frontier in terms of capability. The frontier itself constantly advances, so orbital always offers "last year's model" -- which is still highly capable (frontier models become runnable on single consumer GPUs within 6-12 months), but precludes orbital from competing head-to-head with terrestrial for the highest-value frontier inference.

What the Ratio Floor Implies for the Business Case

The optimistic ratio floor of ~1.48x does not categorically rule out orbital compute, but it does constrain the viable use cases to scenarios where orbital offers non-cost advantages:

1. Energy sovereignty. Nations or organizations without reliable grid access might pay a 50% premium for self-contained compute capacity.

2. Regulatory arbitrage. Orbital compute may face different regulatory frameworks than terrestrial, potentially advantageous for certain workloads.

3. Capacity deployment speed. If terrestrial data center buildout is bottlenecked by power grid interconnection (currently 8-year queues in PJM), orbital could provide faster capacity deployment even at higher cost.

4. Siting constraints. If terrestrial land/power availability becomes severely constrained, the effective terrestrial cost could rise toward parity through scarcity pricing.

However, the TCO ratio analysis shows that the cost-competitiveness thesis -- the claim that orbital compute will be cheaper than terrestrial -- is not supported by the evidence across any scenario in the 2026-2040 timeframe. The fundamental problem is that GPU hardware dominates AI compute costs, energy is a small fraction of terrestrial TCO, and orbital operations introduce large new costs (replacement, platform, opex) that dwarf the energy savings.