Global Launch Activity Concentration, 2006–2025: Structure, Drivers and Robustness

Scope, Coverage and Analytical Definitions

What is measured, and which boundaries constrain the conclusions?

The unit of analysis is a unique launch identifier, grouped by the exact vehicle name stored in GCAT. The selected window is 2006–2025. Successful, failed and partial-outcome O/D launches remain in scope. N flights and no-liftoff events are outside the main database.

The snapshot contains 254 launches outside the selected complete-year window. The snapshot year 2026 is not annualized or compared directly with a complete year. “Complete” describes the calendar period, not a guarantee of perfect source coverage.

HHI is computed from launch shares, not revenue, payload mass or company ownership. A model-name grouping may split a family or combine variants. All later sections retain this unit unless a sensitivity scenario explicitly changes it.

The Long-Term Concentration Trajectory

How has the distribution of launch counts changed over the selected years?

A higher HHI indicates that more launch activity lies in fewer groups. Its reciprocal expresses the number of equally sized groups that would produce the same HHI. These two metrics are mathematical transformations of the same evidence, not independent confirmations.

Endpoint comparisons summarize a long interval but can conceal reversals. The annual series and pooled windows below allow readers to distinguish a persistent shift from a single-year result. Pooling counts weights years by launch volume; it is not an average of yearly HHI.

Leaders, Challengers and the Long Tail

Is concentration mostly about the largest model, or a wider leading group?

Top-k shares are recalculated independently in every year; membership can change. Comparing the largest group, the largest three and the largest ten separates single-leader concentration from a broader leading cluster.

The long tail consists of every model outside the specified leading set. A large count of low-frequency names can coexist with a highly concentrated distribution. “Challenger” here denotes rank position only, not a claim about competitive strategy.

Where Did the HHI Change Come From?

Which model shares account arithmetically for the endpoint change?

This decomposition compares 2006 with 2025. Each contribution is the difference between squared shares, using the union of names observed at either endpoint. A name absent from one covered endpoint has a zero share there.

The chart shows the 12 largest absolute contributions. The remaining 75 names contribute a net -0.006090. Positive and negative contributions sum exactly to the change in overall HHI before display rounding.

A contribution can change because the model’s own count changes, because other counts change, or both. This identity does not distinguish demand, prices, procurement decisions or technological causes.

Growth Without Diversification?

Does greater launch volume coincide with a broader effective model mix?

Observed group count measures how many names occur at least once. Effective group count measures how evenly the launch activity is distributed. They answer different questions: adding a rarely flown model may barely change concentration.

The scatter plot labels each year rather than fitting a causal model. Time trends, changing model definitions and other unobserved factors can produce association. No correlation significance test or causal inference is claimed.

Entry, Exit and Persistence

How stable are the observed model set and its leading members?

An entry is a name observed this year but not the preceding year; an exit is the reverse. Neither is proof of a real-world debut or retirement. A model can reappear after a gap, and the study boundaries truncate its history.

Top-five membership uses descending counts and alphabetical tie-breaking, making selection reproducible. Jaccard similarity compares the intersection with the union of adjacent sets; ties can affect which models enter that set.

Concentration Within Launch Geographies

How concentrated is the model mix within different launch locations?

Countries refer to liftoff geography, not operator nationality. We calculate pooled concentration inside each geographic group. A minimum of 30 launches is required for the comparative chart; this is an editorial eligibility rule, not a statistical guarantee.

Unknown launch countries remain visible in the audit table but do not become a synthetic country in the comparative ranking. Global HHI is not a weighted average of the within-country indices: the same model can operate across locations.

These results measure where recorded activity occurred. They do not establish licensing flexibility, spare capacity or the ability to substitute one launch site for another.

Orbital vs Deep-Space Activity

Does the concentration pattern differ between source categories?

We use non-overlapping windows and require at least 15 launches per category-window for a concentration comparison. Counts below that threshold remain visible, while their HHI is withheld from the report.

O and D are GCAT categories, not classifications inferred from mission names. A launch can carry several payloads; launch count is not mission count. Small deep-space samples warrant caution even when they pass the threshold.

How Sensitive Are the Results?

Which conclusions depend on the unit and inclusion rules?

The scenario table changes one definition at a time at the study endpoints: retaining O only, retaining GCAT S outcomes only, or splitting exact names by their recorded variants. These are alternative estimands, not competing corrections to the main series.

Variant information is unspecified for 1,256 selected launches. The name-plus-variant scenario retains an explicit unspecified bucket within each model and cannot substitute for a verified family mapping.

Pooled windows in the trajectory section also test temporal aggregation. Differences between pooled and annual indices reflect launch-weighted composition as well as within-year distribution. No confidence interval is inferred from the scenario range.

Dependence on the Leading Model: A Mechanical Scenario

What does the distribution of the remaining activity look like without its largest group?

The largest model is identified independently in each year. We remove its observed launches, renormalize the remaining shares and recompute HHI. The scenario changes the population being measured and does not model replacement activity.

A lower residual HHI does not prove resilience; a higher residual HHI does not predict disruption. Neither result measures unused capacity, compatibility with payloads, prices or time to secure an alternative launch.