Seven fiscal years of publicly traceable US federal AI procurement add up to $4.1 billion. European defence-tech startups raised $7.4 billion in the first nine months of 2026 alone.
The CSIS Futures Lab report “Understanding the U.S. Federal Government’s AI Spending”, published 22 September 2026 by Yasir Atalan, Erik Tiersten-Nyman and Benjamin Jensen, identified $4.1 billion obligated across 2,255 AI-related awards from FY2019 through FY2025. The Department of Defense took roughly 83% of the dollars, about $3.4 billion, while holding only about 64% of the contracts by count. FY2025 alone accounted for $1.162 billion of that seven-year total.
64% of contracts, 83% of dollars
The gap between those two shares is the most informative line in the CSIS data. DoD writes fewer AI contracts than its dollar share implies, so each one is substantially larger than the civilian average. Civilian agencies run the opposite pattern: HHS leads by count with 296 awards (131 of them at NIH), followed by NASA at 219, Transportation at 136, Commerce at 129, and Homeland Security at 122. Lots of awards, small tickets.
That has a practical consequence for anyone selling. The civilian side is a volume business with short procurement cycles and modest ceilings. The defence side is a small number of large competitions where losing one is a bad quarter and winning one is a two-year roadmap.
What the $4.1B does not includeCSIS built the dataset by combining SAM.gov solicitation notices with USAspending.gov awarded-contract data and filtering for procurements that mention AI terminology. The authors are explicit that classified contract data is generally not fully visible in public procurement datasets. So $4.1 billion is the publicly observable floor rather than the actual ceiling, and the real DoD share is almost certainly higher than 83%.
There is a second limit worth naming. Keyword filtering catches contracts that say “AI” and misses contracts that deliver AI under other labels: autonomy, sensor fusion, decision support, predictive maintenance. The direction of that error runs one way. It undercounts.
How US federal AI spending compares to European defence-tech funding
Dealroom’s State of Defence Tech 2026, published 5 October 2026, puts European defence-tech startup funding at $7.4 billion year to date as of 20 September 2026, against $2.6 billion for all of 2025. That is roughly 2.85x in under nine months. Dealroom projects about $10.5 billion for the full year, close to four times 2025.
| Measure | US federal AI procurement | European defence tech VC |
|---|---|---|
| Total capital | $4.1B over FY2019 to FY2025 | $7.4B in 9 months of 2026 |
| Prior-period comparison | Contract count more than tripled FY2019 to FY2025 | $2.6B full-year 2025, ~2.85x growth |
| Concentration | 83% of dollars in one buyer (DoD) | 85%+ of dollars in rounds above $100M |
| Largest single line | n/a (not broken out) | Helsing $1.8B Series E at $18B valuation |
| Source | CSIS, 22 Sep 2026 | Dealroom, 5 Oct 2026 |
These are not the same kind of money, and I want to be careful about that. One is obligations against delivered work. The other is equity bought against future revenue. The comparison still bites, because the equity side is being priced against expected government demand, and the publicly visible government demand is smaller than the equity.
Helsing at $18B, Quantum Systems at $8B
The concentration numbers explain where European capital actually went. Rounds above $100 million took more than 85% of 2026 defence-tech funding. Helsing raised a $1.8 billion Series E in July 2026 at an $18 billion valuation. Quantum Systems raised $1.2 billion at $8 billion. Defence tech reached 11.1% of all European VC and 15% of EU-27 VC in 2026.
Bloomberg reported that US investors participated in each of the four largest European defence rounds and overtook local European funds for the first time. The capital is American, the companies are European, and the buyer with the deepest publicly traceable AI budget is the US DoD. Work out what that implies about where these companies expect to sell.
Helsing’s $18 billion valuation is roughly five times the entire publicly visible seven-year AI spend of the US Department of Defense.
The curve matters more than the total
I would not read $4.1 billion as a ceiling on the market. Read the slope. Federal AI contract count more than tripled between FY2019 and FY2025, and FY2025’s $1.162 billion carried over a quarter of the seven-year sum. If FY2026 repeats that growth rate, the annual figure alone starts approaching half the historical total.
The same concentration problem shows up on both sides of the Atlantic, which is a pattern I also found in Q3 2026 AI funding overall: big rounds eat the market. Here it is 85%-plus above $100 million. The mid-market round barely exists in European defence tech right now.
The valuations are priced on a demand curve rather than a demand level. That is defensible if procurement keeps tripling, and dangerous if it plateaus. The specific risk I would watch is the gap between DoD’s dollar concentration (83%) and its contract count (64%): a market where a handful of large awards decide outcomes is a market where one lost competition can reprice an $18 billion company. Unconfirmed, but I expect the FY2026 CSIS update to show a widening dollar gap between DoD and civilian agencies rather than a narrowing one, because civilian AI procurement is mostly small studies and pilots.
Funding size tells you about runway, not product
If you are evaluating defence or dual-use AI vendors, the funding numbers are a weak signal about product and a strong signal about runway. A company that raised at $8 billion is not going to disappear next year, and it is also not going to price flexibly for you. Procurement leverage sits with whoever is not in a hurry.
For civilian agency work and the suppliers serving it, the CSIS count data is the more useful planning input: HHS, NASA and Transportation together account for 651 of the 2,255 awards. Small contracts, repeat buyers, and a keyword-driven solicitation pipeline you can actually monitor on SAM.gov. That is a different go-to-market from chasing DoD programmes, and I think most teams underestimate how different.
One thing I cannot resolve from this data: how much of the DoD’s $3.4 billion is new AI capability versus AI language attached to existing systems integration work. The CSIS methodology cannot separate those, and neither can I.
Treat $4.1 billion as the visible floor of a market growing faster than its headline total suggests, and treat $7.4 billion in European private capital as a bet on that slope continuing. If you are sizing a build-or-buy decision against defence or dual-use AI vendors, price in the possibility that today’s valuations assume procurement growth that has not been contracted yet. Book a call →
