SIPRI Military Expenditure Database

One question, asked continuously since 1949: who spends how much on their military, and what has changed since. The SIPRI Military Expenditure Database is the standard answer - annual spending for over 170 countries plus regional aggregates, running to 2025, which means up to 77 years of uninterrupted history for a single country. Nothing else in the aerospace defense data hub reaches that deep on budgets.

Two design choices make the series usable rather than merely long. First, comparability: every figure is compiled from official government documents, NATO and IMF statistics and national yearbooks, so a country that switched currency, borders or accounting conventions still sits in one consistent frame. Second, redundancy of measurement: each observation arrives in six measures at once - national currency at current prices, constant (2024) US dollars, current US dollars, share of GDP, per-capita spending, and share of total government expenditure - so choosing a lens never requires a second collection pass.

The latest edition covers 1949-2025 as file version 1.2, revised 27 April 2026, and supersedes every earlier version. Get a sample of this dataset and inspect real records before committing pipeline time.

A single observation for the anchor country, exactly as it lands:

COUNTRY    : United States        YEAR: 2024
CONST USD  : 997                  BASIS: constant (2024) US dollars, millions, calendar year
% OF GDP   : 3.4                  BASIS: share of GDP, calendar year

MEASURES   : local_currency_current | constant_2024_usd | current_usd | share_of_gdp | per_capita | share_of_government_expenditure

KEYS       : country, year
SPAN       : 1949-2025            COUNTRIES: 170+ plus regional aggregates
WORLD TOTAL: published from 1988 onward (Soviet-era data unavailable before)

Four rows and the whole analytical contract is visible. Every figure travels with its basis - constant versus current dollars, financial versus calendar year - so a percent-of-GDP column can never be averaged into a per-capita one by accident. The keys are plain country and year, which makes the panel joinable to anything else built on nation-state geography. And the coverage rules ship inside the data rather than in a footnote: the span, the country count, and the honest admission that world totals begin only in 1988.

Eight fields define every observation - the who, the when and the six ways of saying how much. Definitions below were verified against the custodian's documentation during research; the finer layout details flagged underneath the table are folded into your sample on request.

• Geography: over 170 countries plus regional aggregates. Availability varies - series exist for every country independent at the relevant time, and most begin in the late 1950s or earlier, so post-colonial histories are covered from independence rather than dropped.
• Temporal: 1949 through 2025. Regional aggregates extend back using country-level data, but no world total exists before 1988 because Soviet-era figures are unavailable - treat pre-1988 global sums as partial rather than complete.
• Granularity: annual, one observation per country per year. The unit of analysis is the nation-state in a given year, which suits burden-sharing league tables, country-risk panels and budget-cycle backtests equally.

Scale check: 170-plus countries times up to 77 years times six measures is a corpus small enough to load whole into memory, yet dense enough that every defense-economics chart you have seen since graduate school is probably a subset of it.

API, files, or your warehouse. Daily, weekly, or hourly.

You pick the channel and the cadence; the field dictionary above travels unchanged across all three. Files suit analysts who want the full 1949-2025 panel loaded once per planning cycle, API calls suit products that surface a single country-year figure inside a dashboard or briefing tool, and warehouse delivery suits teams running burden-sharing benchmarks in SQL next to their own revenue data.

Revision discipline comes standard. A new edition can restate any year back to 1949, not just the newest one - the 27 April 2026 revision superseded all previous versions outright - so later deliveries carry corrections forward instead of leaving stale values in your warehouse. Each delivery is versioned and preserves superseded values, and every figure keeps its measure basis attached, so reproducible pipelines survive the restatement instead of silently drifting.

• Defense market sizing - the constant-dollar series separates real growth from currency noise, which is the difference between a market that grew and a market that repriced. TAM models for anyone selling into national defense start here.
• Burden-sharing analysis - expenditure as a share of GDP turns a political argument into a sorted table; the share-of-government-expenditure measure adds what a budget competes against domestically.
• Country-risk scoring - decades of consistent observations make spending trajectory a feature, not an anecdote, for sovereign and credit models.
• Budget-cycle backtesting - quants line contractor revenue and equity factors up against 77 years of spending history to see which claims survive contact with the cycle.
• Territory planning - sales teams screen which national budgets are expanding before allocating coverage, rather than after the quarterly call.
• Citable journalism and research - the series is the citation of record for military spending, with methods published openly enough to survive peer review.

Investors and quant researchers get the longest consistent spending panel in the catalog - 77 years, six measures, 170-plus countries - which is the series a defense thesis gets stress-tested against. Market researchers and consultants size and rank national defense markets with per-capita and GDP-share measures already computed. Data scientists and ML engineers get a cleanly keyed country-year table that joins outward to macroeconomic series and company fundamentals on identifiers both sides already share. Journalists, academics and students cite the standard academic source with institutional custody behind every figure. Sales and growth teams read the same panel as a lead-scoring table for which budgets are growing. Five jobs, one spine.

Three things worth knowing upfront. First, coverage is broad but not uniform: some countries carry gaps or estimates, uncertain figures are marked rather than silently dropped, and the per-country method notes sit outside the data itself - your sample includes the availability window for every country you name, so panel construction takes minutes instead of an afternoon. Second, restatement is routine: editions may revise any historical year, which is why every delivery is versioned and corrections propagate forward. Third, when you set these figures beside NATO or European Defence Agency numbers you are comparing different membership sets and slightly different definitions - plan joins on country and year and expect small definitional deltas rather than contradictions.

Notes that pair well with this page: