Most defense-tech “markets” are three companies and a long tail.

Six domains are effectively closed. If you are not one of three companies, you are not raising serious money in them.

Share of each domain's total primary equity held by its three largest raisers, over 24 months.

Sector-wide the same shape holds. Of 395 tracked companies, only 229 raised primary equity at all in the window, and the ten largest raisers took 47% of the $34.3B. The top twenty took 66%.

In five domains a single company holds more than seventy percent of everything raised. Naming the domain and naming the company is close to the same act.

Anduril is excluded throughout. It was the largest raiser and the largest hirer in four of these domains simultaneously, and including it would have made this table a description of one company rather than of the sector.

Space and aerospace is the single domain where capital is genuinely distributed. It is also the largest, at $9.87B.

The practical read: in space, a company can raise a competitive round without being the category leader. In the other nine domains the evidence says otherwise, and a founder or an operator should plan for that rather than assume a level field.

Autonomy teams are growing three times faster than cyber teams. This is the median company, so no single giant is driving it.

Median headcount growth over 12 months, shown only for the five domains with at least 25 companies. Every figure here is unchanged to within two points when the largest company in the domain is removed, which is why these five are shown and the smaller domains are not.

Ten labels, but the real structure is a handful of fused ones. Sensing has merged into space, and autonomy has merged into AI.

58 of 400 tracked companies genuinely operate in more than one domain. Counting each company once and reading the pairs shows two dominant fusions and a long tail of smaller ones. A company selling autonomy without an AI stack, or satellites without sensing, is now the exception.

Cohort. 400 tracked defense-tech and national-security companies, of which 395 carry a usable domain and appear here. Anduril is excluded for the reason given in section 02. Companies whose only classification was a services or unclear label are excluded.

Capital. Disclosed primary equity only, 24 months to September 2026, taken from full round history rather than the most recent round. That distinction matters: 97 companies raised more than once in the window, and counting only the latest round understated the total by 32.5%. Debt, secondary sales, grants, corporate rounds and post-IPO raises are all excluded, because none of them is new money into the business. Undisclosed rounds are invisible to us, so every figure is a floor.

Domains. Every company carries one primary domain and, where genuinely accurate, additional ones. All capital totals use the primary domain only. Under full attribution a large company's entire headcount and capital land in every domain it touches, and in testing, 89% of the sensors category's apparent hiring turned out to be borrowed from companies that are primarily something else. Multiple domains are used only for the overlap map in section 05.

What we cut, and why. We built and then dropped a capital-per-hire figure for each domain. It did not survive its own robustness test: removing the single largest company moved several domains by more than half, because six of the ten hold fewer than twenty companies and are dominated by one or two rounds. The concentration this issue reports is the reason that metric failed, so we published the cause rather than the artefact. Growth rates are shown only where they hold steady when the largest company is removed.

Hiring. Net headcount change over 12 months from LinkedIn observations. LinkedIn lags real start dates and counts profile updates rather than payroll, so recent months are understated. Net change is not gross hiring: a company that hired forty and lost ten shows thirty.