Elbit Systems reported revenues of $7.94 billion for 2025. Israel Aerospace Industries reported $7.38 billion. Rafael Advanced Defense Systems reported $6.8 billion. None of the three appears among the 50 entries in the Israeli Defence-AI Deployment Index, and the reason is a single inclusion rule rather than a view about Israeli defence primes AI work. The rule makes the index sharper. It also means the index cannot describe where most of the money sits.
The inclusion test is about the product, not the size
The index covers companies whose core product is artificial intelligence or autonomy applied to defence. That is the whole test. A company qualifies when the thing it sells is the software, the autonomy stack, the sensor-fusion layer or the machine-learning model. It does not qualify when AI is a component embedded across a broad portfolio of platforms, munitions, avionics and vehicles.
For Elbit, Rafael and IAI, AI is a component. Elbit told investors it invested over $500 million in disruptive research and development programmes including AI enhancements across multiple platforms. That is a description of AI improving a portfolio, not of AI being the portfolio. Applying the index's grading to a company like that would mean grading a conglomerate, and the resulting entry would be uninformative: every prime would score L3 on the strength of some part of its catalogue, and the grade would say nothing about any particular system.
The test is not a size or age filter. The index includes TSG IT Advanced Systems, founded in 1969, ImageSat International, founded in 1997, and Bagira Systems, founded in 1995. It includes Cognyte Software and Windward, both publicly listed. What those companies have in common is that the thing being graded is identifiable as a product.
What the exclusion buys
Two things. First, comparability. When every entry is a company selling one identifiable class of AI or autonomy product, an L2 grade means roughly the same thing across the field: someone announced a purchase of this product. Second, falsifiability. Each of the 50 entries carries one source URL that a reader can open and check against the grading rules. A prime's entry would rest on dozens of sources describing dozens of unrelated systems, and would be unfalsifiable in practice.
The distribution the index reports — 13 companies at L3, 22 at L2, six at L1 and nine at L0 — only holds meaning because the unit of analysis is consistent. That distribution is the subject of the deployment evidence gap.
What the exclusion costs
This is the part that should be stated plainly rather than buried in a methodology note.
The three excluded companies reported roughly $22 billion in combined 2025 revenue. IAI reported an order backlog approaching $29 billion, of which about 71 per cent came from international customers. Rafael reported a backlog of $23.3 billion and said the IDF remained its largest customer at 54 per cent of sales. Israel's Ministry of Defense reported that national defence exports reached a record $19.2 billion in 2025.
Against those figures, an index of 50 startups and mid-size specialists is describing a slice. It cannot tell a reader which Israeli organisations hold the largest AI-relevant order books, which have the deepest integration work under way, or how much of the national export figure has an autonomy component inside it. Anyone who reads the index as a map of the Israeli defence AI market will be reading it wrongly. It is a map of published deployment evidence among AI-first suppliers, and nothing wider. That boundary is set out further in what a deployment index cannot tell you.
The exclusion is visible in the thin categories
Three categories in the index came back with exactly one qualifying company each: active protection (BIRD Aerosystems), targeting and fire control (SMARTSHOOTER), and military logistics and sustainment (Odysight.ai). Read without the inclusion rule in mind, that looks like a claim that Israel has almost nothing in those areas. It is not.
Autonomous military logistics is the clearest example. Raksha Anirveda reported that under the UK Ministry of Defence's Project THESEUS, Marlborough Communications and Israel Aerospace Industries were awarded a contract to deliver demonstrations of a highly automated ground and air resupply network. That is AI-enabled sustainment work being carried out by a prime, in a category where the index finds a single qualifying startup. The emptiness of the category is a finding about where AI-first companies have not gone, not about national capability. The point is developed in three categories came back almost bare and in the military logistics and sustainment pillar.
The boundary is already blurring
The rule is clean today and will not stay clean. Globes reported that Ondas bought the Israeli defence software company Omnisys for $200 million; Ondas also owns the Iron Drone Raider counter-UAS line and, per an investor release, expanded its AI sensing through Insight Intelligent Sensors. Three separate index entries now sit inside one corporate group. If that group keeps acquiring, the index will at some point be grading a portfolio under the same objection it applies to the primes.
Landscape maps handle this differently, and the trade-offs are compared in Israeli defence-tech landscape maps compared. The index's position is that the inclusion test will be restated rather than quietly stretched, and the current version states it on the face of the page.
Limits of this reading
The index publishes no view on whether the primes' AI work is more or less advanced than that of the companies it lists. It holds no data on that question and does not collect any.
Revenue and backlog figures quoted here are company-reported and cover entire businesses, not AI-specific lines. None of the three companies publishes a breakdown of AI or autonomy revenue, so the figures establish scale and nothing about the share of it that is AI. The Ministry of Defense export total is a government figure covering all defence exports and is not decomposable into AI content from public sources.
Frequently asked questions
Why are Elbit, Rafael and IAI not in the deployment index?
The index covers companies whose core product is AI or autonomy applied to defence. For the large Israeli primes, AI is a component distributed across broad portfolios of platforms and munitions rather than the product being sold, so they fall outside the inclusion test.
Does excluding the primes make the index unrepresentative of Israeli defence AI?
It makes it unrepresentative of Israeli defence AI spending. The three primes reported roughly $22 billion of combined 2025 revenue. The index measures published deployment evidence among AI-first suppliers and does not attempt to size the market.
Would a prime be added if it spun out an AI product company?
Yes. The test is applied to the entity being graded. A subsidiary or spin-out whose core product is an AI or autonomy system, with a public source describing a trial, contract or operational use, would qualify on the same terms as any other entry.
Sources
- idf.ai, The Israeli Defense-AI Deployment Index v1.0, 24 August 2026
- Elbit Systems, Elbit Systems reports fourth quarter and full year 2025 results
- Defence Industry Europe, Israel Aerospace Industries reports record 2025 financial results with order backlog nearing $30 billion
- Defence Industry Europe, Rafael reports record 2025 results with $6.8 billion in sales
- Israel Ministry of Defense, All-time defense export record: Israel crosses the $19 billion threshold, June 2026
- Raksha Anirveda, UK MOD Project Theseus: Marlborough Communications and Israel Aerospace Industries awarded contract to deliver demonstrations of highly automated ground and air resupply network capability
- Globes, Ondas buys Israeli defense software co Omnisys for $200m
Independent publication of idf.ai. Not affiliated with, endorsed by, or connected to the Israel Defense Forces, the Israeli Ministry of Defense, or any government body. Compiled entirely from publicly published sources. No classified, restricted or non-public information. Listed companies may dispute any entry: send the published source that contradicts it and the entry will be amended or removed.