The Launch-Cost Equation

Part 14 of 30 · Series date: September 8, 2026

A cheaper ride to orbit changes the equation; useful lifetime output determines whether the equation works.

A low airfare does not tell you the total price of a vacation. There are bags, lodging, transportation, meals, and the cost of changing plans. The quoted price of putting one kilogram in orbit is similarly only the beginning of an orbital business calculation.

Cheap launch can open possibilities. It cannot independently prove that a data center will make money.

Count everything that flies

The payload includes more than chips. Arrays, radiators, structure, communications, power electronics, propellant, protection, and deployment hardware may account for much of the mass.

An operator should state whether a quoted computing density refers to a processor board, a payload module, or the complete spacecraft. Comparing a bare chip's mass with a whole terrestrial facility is meaningless.

NASA's 2024 space-solar assessment identified manufacturing, launch, assembly, and related capability gaps as parts of the system-level challenge. Its findings concern the studied power architectures, but its insistence on complete accounting is equally useful when evaluating compute proposals. NASA: space-solar assessment.

A transparent example

Suppose an imaginary computing spacecraft has a launch mass of 10,000 kilograms. At an assumed transportation charge of $1,000 per kilogram, transportation alone is $10 million. At $200 per kilogram, it is $2 million.

These are scenario inputs, not current prices or forecasts for a named rocket. They show how a lower launch price can help while leaving the hardware and operating bills untouched.

Now suppose the first option delivers twice the useful lifetime output because it includes more supporting equipment. The cheaper launch bill might no longer identify the cheaper service. The denominator matters as much as the numerator.

What should the denominator be?

For electricity, a useful denominator is energy delivered at the agreed point over the system's life. For computing, it is useful work completed to an agreed quality and deadline.

The phrase “GPU-hour” may be helpful within a defined hardware class, but different processors can perform very different amounts of work in that hour. Cost per task can be more meaningful when the task is representative and the output quality is held constant.

A simple planning ratio is total lifetime cost divided by useful lifetime output. A serious financial model must then account for timing, financing, uncertainty, and the value of capital. The simple ratio is a screening tool, not a substitute for that model.

Utilization can dominate

Imagine two identical machines. One performs paid work 80 percent of the time; the other only 20 percent. The second machine spreads its fixed costs across far less output.

In orbit, downtime can arise from empty job queues, communications gaps, thermal constraints, faults, or unavailable data. Not all idle time is a hardware failure. A perfectly healthy satellite can be commercially underused.

This makes an anchor customer valuable, but also creates concentration risk. If the customer leaves, the operator needs other suitable workloads—not merely evidence that AI demand is growing somewhere in the economy.

Cadence is not the same as price

A low theoretical price has limited value if launches are unavailable when equipment is ready. Delays can leave expensive processors aging in storage. A replacement strategy that depends on frequent flights also needs enough actual transportation capacity.

Packaging constraints matter too. A system may run out of fairing volume before reaching the advertised mass limit. The destination orbit and integration requirements can change what a particular flight offers.

Procurement should therefore distinguish an aspirational future cost, a published commercial offer, and a contracted mission price.

Translate a power rating into launched mass

A useful early design measure is the complete system's mass per unit of sustained payload power. It links the thermal and electrical architecture to transportation.

Use an invented example of ten kilograms of total spacecraft mass for each kilowatt of sustained computing power. A one-megawatt computing platform would then have a mass of 10,000 kilograms. If the real design requires thirty kilograms per kilowatt, the mass becomes 30,000 kilograms before changing the computing promise.

At the same assumed launch price, transportation triples. The difference could arise from heavier radiators, additional batteries, structure, or fault protection. It cannot be understood by looking only at the accelerator board.

These values are scenario inputs, not a prediction of achievable spacecraft performance. The point is that mass and useful power must come from the same buildable design. Combining one laboratory component's low mass with another system's high output creates a fictional spacecraft in the spreadsheet.

A fleet is a stream of replacements

A constellation is not purchased once and then frozen in time. Hardware eventually fails, becomes obsolete, or reaches its planned retirement.

Imagine a mature fleet of 1,000 identical nodes with staggered five-year service lives. In a simplified steady state, maintaining the same fleet size requires replacing about 200 nodes per year. Unexpected losses or growth would add to that requirement.

The example assumes uniform life and a smooth schedule; real fleets would be less tidy. It still exposes a major obligation. A launch plan that can establish the fleet once may not be able to sustain it, upgrade it, and expand it simultaneously.

Manufacturing has to match that rhythm. So do integration, ground operations, and disposal. If retired hardware accumulates faster than it can be managed, a transportation success can create an operational failure.

The relevant business model is a continuing flow of equipment and useful output, not a photograph of the fleet at its largest moment.

Starship's role is enabling, not sufficient

For the stack discussed in this series, Starship represents the large-payload transportation ambition. A larger vehicle could allow different packaging and deployment choices, potentially changing which orbital structures are practical. Actual mission capacity, availability, reuse economics, and contracted prices must still be verified for the intended flight. SpaceX: Starship.

Even a dramatic transportation improvement would leave other bills intact. If launch makes up a fifth of a hypothetical service's lifetime cost, cutting that component in half reduces total cost by ten percent, assuming everything else stays unchanged. If launch dominates the total, the same improvement matters much more.

This arithmetic explains why two honest analysts can reach different conclusions about the same rocket. They may be evaluating systems with very different cost structures.

The strongest launch argument therefore begins with the complete orbital design. Show how the new transportation capability changes its mass, assembly, replacement, and output. Then calculate the service benefit.

A rocket opens a door. The equipment and business going through that door still need to justify the journey.

Stress-test the attractive case

Recalculate the business with a shorter hardware life, lower utilization, higher communications charges, and delayed replacement. Change one variable at a time, then examine combinations.

If profitability requires every optimistic assumption to occur together, the proposal is fragile. If it remains useful across a reasonable range, it deserves more attention.

What would prove this?

Publish a cost model with explicit mass, power, lifetime, utilization, launch schedule, and disposal assumptions. Replace assumptions with measured operations as demonstrations proceed.

The decisive milestone is not a record-low price per kilogram. It is a repeatable service whose total bill is justified by what customers actually receive.

Previous article · Next article