Military AI Companies in 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: August 28, 2026 | License: CC BY 4.0
Two companies hold 98.04% of the disclosed U.S. military AI contract ceiling. Axis Intelligence Research finds that across ten scored defense AI vendors, Anduril’s $20.0 billion and Palantir’s $10.0 billion Army enterprise contracts account for $30.0 billion of $30.6 billion in publicly disclosed ceiling — while eight of the ten frontier AI companies now cleared for classified networks have no disclosed AI contract ceiling at all.
Quick Answer
The largest military AI companies in 2026 are Anduril Industries (Defense AI Anchoring Score 83.3, $20.0 billion Army contract ceiling, $61 billion valuation) and Palantir Technologies (DAAS 94.1, $10.0 billion Army ceiling, $809 million in U.S. government revenue in Q2 2026 alone). Behind them sit the frontier labs — OpenAI, Google, xAI — each capped at a $200 million prototype ceiling. The U.S. Department of War requested $58.5 billion for AI and CJADC2 in FY2027, of which 78.63% is compute infrastructure, not vendor software.
Key Findings
- According to Axis Intelligence Research, Anduril Industries and Palantir Technologies together hold 98.04% of the $30.6 billion in disclosed U.S. Department of War AI and autonomy contract ceiling across the ten vendors in the Axis DAAS cohort, as of August 28, 2026.
- The Department of War requested $58.5 billion for artificial intelligence and Combined Joint All-Domain Command and Control in fiscal year 2027, of which $46.0 billion — 78.63% by Axis Intelligence Research’s calculation — funds the government-owned “AI Arsenal” compute infrastructure rather than commercial AI software.
- Palantir reported $809 million in U.S. government revenue in the second quarter of 2026, up 90% year over year and equal to 41.8% of the company’s $1.935 billion total quarterly revenue.
- Eight companies — SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Amazon Web Services and Oracle — signed agreements on May 1, 2026 to deploy frontier AI on Department of War Impact Level 6 and Impact Level 7 classified networks; Anthropic was excluded.
- Axis Intelligence Research finds that the Pentagon’s single-year FY2027 AI request is 1.91 times larger than the entire multi-year disclosed contract ceiling of the ten military AI vendors tracked in this dataset.
Which Companies Actually Hold U.S. Military AI Contracts?
The phrase “defense AI company” now covers three groups that behave nothing alike: autonomy primes with nine-figure hardware lines, frontier model labs selling inference into secure enclaves, and cloud providers selling the racks everything runs on. Grouping them into one list produces a ranking that means nothing. So this analysis scores them on what a contracting officer can actually verify.
The Defense AI Anchoring Score (DAAS) is an Axis Intelligence Research metric that measures how deeply a vendor is embedded in U.S. military procurement, rather than how much venture capital it has attracted. It has three disclosed components, all traceable to primary award documents:
| Component | Weight | What it measures |
|---|---|---|
| Contract anchoring | 45 points | Largest single disclosed U.S. Department of War contract ceiling, log-normalised |
| Institutional permanence | 30 points | Program-of-record status, enterprise agreement, production deployment, or prototype only |
| Classified-network depth | 25 points | Confirmed operation at Impact Level 6/7, enterprise IL5, fielded-unclassified, or none |
The formula is stated in full in the Methodology section, and every input sits in the downloadable CSV.
Source: Axis Intelligence Research — CC BY 4.0
Defense AI Anchoring Score, August 2026 baseline
| Rank | Company | DAAS | Largest disclosed DoW ceiling | Permanence | Classified depth | Source |
|---|---|---|---|---|---|---|
| 1 | Palantir Technologies | 94.1 | $10.0B (Army EA) | Program of record | IL6 / multi-domain | U.S. Army; DefenseScoop |
| 2 | Anduril Industries | 83.3 | $20.0B (Army EA) | Enterprise agreement | Fielded, unclassified | U.S. Army |
| 3 | OpenAI | 50.9 | $200M (CDAO OTA) | Enterprise deployment | IL6/IL7 agreement | CDAO; Dept. of War |
| 3 | Google Public Sector | 50.9 | $200M (CDAO OTA) | Enterprise deployment | IL6/IL7 agreement | CDAO; Dept. of War |
| 5 | xAI | 42.6 | $200M (CDAO OTA) | Enterprise deployment | GenAI.mil (IL5) | CDAO |
| 6 | BigBear.ai | 36.7 | Not disclosed | Fielded programs | GenAI.mil via Ask Sage | BigBear.ai; CDAO |
| 7 | Reflection AI | 35.0 | Not disclosed | Prototype / new entrant | IL6/IL7 agreement | Dept. of War |
| 8 | Shield AI | 28.3 | Not disclosed | CCA autonomy provider | Fielded, unclassified | Shield AI |
| 9 | Scale AI | 18.3 | Not disclosed | Prototype OTA | Fielded, unclassified | Defense Innovation Unit |
| 10 | Anthropic | 0.0 | Contracts cancelled | None active | Access removed | CNN; DefenseScoop |
Snapshot as of August 28, 2026. DAAS scores only vendors whose primary product is AI or autonomy software; cloud and compute suppliers are listed separately below. “Not disclosed” scores zero on contract anchoring — an absence of public ceiling, not an absence of revenue.
The gap between rank 2 and rank 3 is the story. Anduril and Palantir are separated from everyone else not by model quality but by contract architecture: both converted scattered task orders into decade-long enterprise vehicles. The U.S. Army consolidated more than 120 separate procurement actions for Anduril into a single 10-year contract with a $20 billion ceiling on March 13, 2026 — a five-year base with a five-year option, centred on the Lattice command-and-control layer. Seven months earlier the Army had folded 75 Palantir contracts — 15 prime and 60 related — into a 10-year agreement with a $10 billion ceiling.
A ceiling is not revenue. The Army says so explicitly: the $20 billion “represents the maximum potential value, not an obligated amount.” Treating these as booked contracts is the single most common error in coverage of this sector, and it inflates the apparent scale of defense AI by an order of magnitude.
Sarah Mitchell: The interesting thing about an enterprise agreement is that it changes what a vendor has to win. Under 120 separate procurement actions, you have to be the best answer 120 times. Under one agreement, you have to be the incumbent once. The Army was candid that the point is pre-negotiated pricing and volume discounts — a procurement efficiency argument, not a capability argument. What it also does, quietly, is move the competitive question from “which model performs better on this task” to “which vehicle is already open.” That is a different game, and the frontier labs, with their $200 million prototype ceilings, are not yet playing it.
How Much Is the Pentagon Actually Spending on Military AI?
The FY2027 request is the first defense budget where AI is a headline line item rather than a subheading. The Department of War’s FY2027 Budget Overview requests $1.45 trillion in total budgetary resources, a $440.9 billion (44%) increase over the $1,009.1 billion enacted for FY2026. Within it:
FY2027 AI and autonomy request, by line
| Line item | FY2027 request | Share of AI request | Source |
|---|---|---|---|
| AI and CJADC2, total | $58.5B | 100% | Dept. of War Overview Book |
| Sovereign AI Arsenal (multi-year mandatory) | $46.0B | 78.63% | Dept. of War Overview Book |
| AI integration into platforms and workflows | $6.8B | 11.62% | Dept. of War Overview Book |
| Maven Smart System + Joint Fires Network | $2.3B | 3.93% | Dept. of War Overview Book |
| Leading-edge AI research and development | $2.2B | 3.76% | Dept. of War Overview Book |
| Unallocated residual | $1.2B | 2.05% | Axis calculation |
| Autonomous and remotely operated systems | $54.0B | — | Dept. of War Overview Book |
| Drone Dominance (mandatory) | $53.6B | — | Dept. of War Overview Book |
| Counter-Unmanned Systems | $20.6B | — | Dept. of War Overview Book |
| Defense Innovation Unit | $956M | — | Dept. of War Overview Book |
Axis Intelligence Research finds that 78.63% of the FY2027 AI request is infrastructure, not vendor software. The AI Arsenal funds “government-owned AI supercomputing infrastructure within secure, hardened data centers, combined with flexible commercial surge capacity.” Add the application-layer lines together — integration, Maven and Joint Fires Network, and AI research — and they total $11.3 billion, or 19.32% of the request.
One honest note on the arithmetic: the four disclosed components sum to $57.3 billion against a stated total of $58.5 billion, leaving a $1.2 billion residual. The Overview Book introduces the list with “including,” so it is not presented as exhaustive; we flag the gap rather than silently reconciling it.
The counter-drone line deserves its own caveat. The Department states $20.6 billion for Counter-Unmanned Systems, “an increase of $16.6 billion, or 424 percent over the FY 2026 enacted level of $3.9 billion.” Subtracting $3.9 billion from $20.6 billion gives $16.7 billion, and 428.2%. The discrepancy is rounding inside the source document, and we cite the Department’s own figures rather than our recomputation. For the drone-side view of that budget line, see our military drone statistics dataset.
Sarah Mitchell: Read the AI Arsenal line the way you would read a hyperscaler’s capex disclosure, because that is what it is. The Pentagon has decided that the scarce resource is not models — models are commodities it can license four at a time for $200 million each — but sovereign compute it controls at IL6 and IL7. Every frontier lab pitching the Department is pitching into a budget where roughly four fifths of every AI dollar is being spent on the floor beneath them. That is a structurally uncomfortable place to sit, and it is the reason the classified-network agreements matter more than the contract ceilings do.
Which AI Companies Are Cleared for Classified Military Networks?
On May 1, 2026 the Department of War announced agreements with eight frontier AI companies to deploy on Impact Level 6 and Impact Level 7 environments. IL6 is the compliance standard for classified cloud workloads; IL7 covers top secret and the most sensitive national security information. This list is the closest thing that exists to an official register of who is trusted with military AI at the high side.
Frontier AI on DoW classified networks, as of May 1, 2026
| Company | Category | IL6/IL7 agreement | Also a CDAO $200M awardee |
|---|---|---|---|
| SpaceX | Space systems / connectivity | Yes | No |
| OpenAI | Frontier models | Yes | Yes |
| Frontier models + cloud | Yes | Yes | |
| NVIDIA | AI compute | Yes | No |
| Reflection | Frontier models | Yes | No |
| Microsoft | Cloud + models | Yes | No |
| Amazon Web Services | Cloud | Yes | No |
| Oracle | Cloud | Yes | No |
| xAI | Frontier models | No | Yes |
| Anthropic | Frontier models | No | Yes (cancelled) |
Source: U.S. Department of War, “Classified Networks AI Agreements,” May 1, 2026; CDAO, July 14, 2026.
The same release disclosed the adoption number that best captures where military AI actually is in 2026: over 1.3 million Department personnel have used GenAI.mil, generating tens of millions of prompts and deploying hundreds of thousands of agents, in five months. That is not a pilot. It is the largest disclosed enterprise deployment of generative AI inside any government, and it happened in under half a year.
The July 14, 2025 CDAO awards — Anthropic, Google, OpenAI and xAI, each with a $200 million ceiling for agentic AI workflows — look small next to the enterprise agreements, and they are. Four awards at $200 million equal $800 million of ceiling against Anduril’s $20 billion alone. The frontier labs entered defense through the narrowest door in the building.
The Anthropic exception
Anthropic is the only company in this dataset to go from a $200 million ceiling to zero. The Department designated it a supply chain risk in March 2026 — a label normally reserved for foreign-linked vendors — after the company declined to permit its models to be used for mass surveillance of Americans or for fully autonomous weapons without human authorisation over targeting. Anthropic sued on March 9, 2026. On August 27, 2026, a federal judge ruled the designation unlawful and ordered it removed, finding First and Fifth Amendment violations. Parallel litigation in Washington, D.C. remains pending as of this dataset’s snapshot date, so the designation is not fully resolved.
We score Anthropic at 0.0 because that is what the August 2026 snapshot shows: no active award, no classified access. It is also the score most likely to move at the next review. DefenseScoop reported that Anthropic’s models had previously been the only ones integrated into the Department’s classified workflows, via a partnership with Palantir — which makes the exclusion a genuine capability subtraction, not a symbolic one.
Which Military AI Startups Have Raised the Most?
Private capital and contract anchoring have decoupled. Several of the best-funded defense AI startups sit in the bottom half of the DAAS table, because valuation is priced on forward autonomy demand while contract ceilings are awarded on delivered systems.
Defense AI funding rounds, 2026
| Company | HQ | Round | Amount | Post-money valuation | Date | Source |
|---|---|---|---|---|---|---|
| Anduril Industries | United States | Series H | $5.0B | $61.0B | 2026-05-13 | Anduril |
| Helsing | Germany | Series E | $1.8B | $18.0B | 2026-07-13 | Defense News |
| Saronic Technologies | United States | Series D | $1.75B | $9.25B | 2026-03-31 | CNBC |
| Shield AI | United States | Series G | $1.5B | $12.7B | 2026-03-26 | Shield AI |
| Quantum Systems | Germany | Series D | $1.2B | ~$8.0B | 2026-07-02 | CNBC reporting |
Anduril’s Series H doubled its valuation from the $30.5 billion set in June 2025, against 2025 revenue of $2.2 billion — a valuation-to-revenue multiple of 27.7 by Axis Intelligence Research’s calculation. Its $20 billion Army ceiling equals 32.8% of that valuation, the highest contract-to-capital ratio in the private cohort.
Shield AI’s Series G raised $1.5 billion in equity at a $12.7 billion post-money valuation, with a further $500 million in fixed-return preferred equity from Blackstone funds and a $250 million delayed draw facility. That is a 139.6% valuation increase over the $5.3 billion mark set twelve months earlier — and it happened without a disclosed nine-figure contract ceiling attached to the Hivemind autonomy stack. Saronic’s $1.75 billion round shows the same pattern in the maritime domain.
Among the publicly traded pure-plays, scale is an order of magnitude smaller. BigBear.ai reported Q2 2026 revenue of $36.7 million, up 13% year over year, with $269.6 million of backlog and $409.8 million in cash and investments, and full-year guidance of $135–165 million. Its Ask Sage acquisition matters more than its revenue line: Ask Sage powers the Army’s Enterprise Large Language Model Workspace, which the CDAO named as one of the routes by which generative AI reaches combatant commands.
How Does Palantir’s Government Business Compare?
Palantir is the only company in this cohort that discloses audited military AI revenue, which makes it the sector’s only real benchmark. Its Q2 2026 results filed with the SEC show:
| Metric | Q2 2026 | Change | Note |
|---|---|---|---|
| Total revenue | $1,935.464M | +93% YoY | |
| U.S. government revenue | $809M | +90% YoY | |
| U.S. government share of revenue | 41.8% | — | Axis calculation |
| Six-month revenue | $3,568.047M | +89.0% YoY | Axis calculation vs $1,887.552M |
| FY2026 revenue guidance | $8,150–8,158M | Raised |
The Maven Smart System is the vehicle behind much of that government line. It began as a $480 million five-year IDIQ in May 2024; the Department raised the ceiling by $795 million in May 2025, taking it to roughly $1.3 billion through 2029. In a March 9, 2026 memorandum, Deputy Secretary of War Steve Feinberg designated Maven a formal program of record — the status that converts a successful prototype into a funded, standardised capability across services. The FY2027 budget then requests $2.3 billion for Maven and the Joint Fires Network combined.
Sarah Mitchell: Program-of-record status is the thing to watch, not the ceiling. A ceiling is an option the government may never exercise; a program of record is a line in the budget that has to be actively killed to disappear. Palantir spent nine years converting one from the other. Every frontier lab currently holding a two-year prototype OTA is at the stage Palantir was at in 2017 — and the historical base rate for prototypes reaching program-of-record status is not encouraging, whatever the eval scores say.
Who Are the Leading European Defense AI Companies?
Europe’s defense AI cohort is younger, smaller in contract terms, and constrained by parliaments in a way the U.S. cohort is not.
Helsing, founded in Munich in 2021, raised $1.8 billion at an $18 billion valuation on July 13, 2026 — the largest private defense-technology round in European history. Its portfolio spans the Altra battlefield-operations software, the HX-2 loitering munition, and the CA-1 Europa uncrewed aircraft concept.
The German contract that underwrites that valuation is instructive about how European procurement differs. In February 2026 the Bundestag budget committee approved an initial framework worth roughly €269 million for HX-2 munitions. Procurement documents had shown option values reaching €1.46 billion for Helsing and €2.86 billion for Stark Defence. Lawmakers then capped the framework at €1 billion per producer and €2 billion combined, requiring fresh parliamentary approval beyond that, on the reasoning that unchecked options would create de facto multi-billion-euro budget commitments.
That is the structural difference worth naming. A U.S. enterprise agreement sets a $20 billion ceiling and lets the Army order against it. A German framework sets a ceiling and then a budget committee removes it. Any comparison of European and American defense AI companies that treats disclosed ceilings as equivalent is comparing two different legal objects.
Quantum Systems, also Munich-based, raised $1.2 billion at roughly $8 billion post-money on July 2, 2026, co-led by Blackstone, Noteus, Airbus and Advent — a hardware-weighted counterpart to Helsing’s software-first thesis, competing for the same NATO procurement.
What the Data Does Not Show
Three limits of scope worth stating plainly, because they bound how this dataset should be cited.
Classified and obligated spending is invisible. Contract ceilings are announced; obligations largely are not, and classified AI programs appear nowhere in public budget exhibits. DAAS measures disclosed anchoring, which is a proxy for embeddedness, not a measure of revenue.
Absence of a ceiling is not absence of business. Shield AI, Scale AI and Applied Intuition all hold Department work whose value has not been publicly disclosed. A zero on contract anchoring reflects the public record on August 28, 2026, not a judgment about the company.
Enterprise-agreement ceilings are not comparable to program ceilings. Anduril’s $20 billion consolidates more than 120 pre-existing actions; Palantir’s $10 billion consolidates 75. Neither represents new money. The CDAO’s $200 million awards, by contrast, are genuinely new prototype authority. DAAS log-normalises to compress this distortion, but it does not eliminate it.
Methodology
Collection. All figures were retrieved from primary documents between August 27 and August 28, 2026: the Department of War FY2027 Budget Overview Book and budget press release, war.gov release announcements, army.mil contract announcements, ai.mil CDAO releases, diu.mil program announcements, Congressional Research Service report R49023, SEC EDGAR filings, and company investor-relations releases. No figure in this article is drawn from model memory or from secondary data aggregators. Every number in the prose has a corresponding row in the downloadable CSV with its source URL and retrieval date.
The Defense AI Anchoring Score. DAAS is an Axis Intelligence Research metric scored 0–100 on this formula:
DAAS = 45 × min(1, max(0, (log10(C) − 8) / 2.3010))
+ 30 × (P / 3)
+ 25 × (N / 3)
Where:
- C = the largest single disclosed U.S. Department of War contract ceiling attributable to the vendor, in USD. The normalisation floor is $100 million (log10 = 8) and the cap is $20 billion (log10 ≈ 10.3010), the largest ceiling in the cohort; the denominator 2.3010 is the difference. A vendor with no disclosed ceiling scores 0 on this component.
- P = institutional permanence, ordinal 0–3. 3 = designated program of record or multi-year enterprise agreement with the U.S. Army or Department of War. 2 = department-scale production or enterprise deployment. 1 = prototype other-transaction agreement, pilot, or limited award. 0 = no active Department award as of the snapshot date.
- N = classified-network depth, ordinal 0–3. 3 = party to the May 1, 2026 IL6/IL7 classified-network agreements, or documented operation across multiple classification domains. 2 = documented department-wide IL5 enterprise deployment. 1 = documented fielded Department use with no disclosed classified accreditation. 0 = none disclosed, or access removed.
Worked example — Palantir Technologies: C = $10.0 billion, so log10(C) = 10.0 and the anchoring component is 45 × ((10.0 − 8) / 2.3010) = 39.11. P = 3 (Maven program of record plus the Army enterprise agreement), giving 30.00. N = 3 (Maven operates across three classification domains), giving 25.00. Total: 94.1.
All DAAS arithmetic was computed and independently re-verified in Python using both floating-point and exact rational arithmetic before publication.
Scope of the DAAS cohort. DAAS scores only vendors whose primary product is AI or autonomy software sold to militaries. Cloud, compute and launch infrastructure suppliers — Microsoft, Amazon Web Services, Oracle, NVIDIA, SpaceX — are excluded from scoring and appear only in the classified-network table, because their Department revenue is dominated by non-AI infrastructure and would not be measuring the same thing. Google Public Sector is scored because its CDAO award and GenAI.mil model deployment are AI-specific.
Cross-source calculations. Cohort ceiling concentration is computed as ($20.0B + $10.0B) ÷ ($20.0B + $10.0B + $0.2B + $0.2B + $0.2B) = 30.0 ÷ 30.6 = 98.04%. The AI-request ratio is $58.5B ÷ $30.6B = 1.91. Budget shares are computed directly against the $58.5 billion stated total. Palantir’s government revenue share is $809M ÷ $1,935.464M = 41.8%.
Currency. All contract ceilings in the DAAS index are U.S.-dollar Department of War awards, so no foreign-exchange conversion is applied. European contract values are reported in euros as awarded and are not converted, to avoid introducing an undisclosed exchange-rate assumption.
About This Dataset
Title: Military AI Companies: Contracts, Funding and the Defense AI Anchoring Score, 2026
Creator and publisher: Axis Intelligence Research
Coverage: 14 named companies plus 6 infrastructure suppliers, across the United States and Europe; U.S. federal fiscal years 2024–2027; snapshot date August 28, 2026
Rows: 151 observations, each with source organisation, source document, source URL and retrieval date
Format: UTF-8 CSV, one observation per row
Licence: Creative Commons Attribution 4.0 International (CC BY 4.0). You may reuse, republish and build on this dataset commercially, with attribution.
Download: military-ai-companies-2026.csv
Cite this dataset
APA: Axis Intelligence Research. (2026). Military AI companies: Contracts, funding and the Defense AI Anchoring Score, 2026. https://axis-intelligence.com/military-ai-companies/
MLA: Axis Intelligence Research. “Military AI Companies: Contracts, Funding and the Defense AI Anchoring Score, 2026.” Axis Intelligence Research, 28 Aug. 2026, axis-intelligence.com/military-ai-companies/.
Chicago: Axis Intelligence Research. “Military AI Companies: Contracts, Funding and the Defense AI Anchoring Score, 2026.” Axis Intelligence Research. August 28, 2026. https://axis-intelligence.com/military-ai-companies/.
Frequently Asked Questions
Which military AI company holds the largest U.S. contract ceiling?
Anduril Industries, at $20.0 billion. The U.S. Army awarded the 10-year enterprise contract on March 13, 2026, consolidating more than 120 separate procurement actions into a single vehicle centred on the Lattice software platform. Palantir holds the second largest at $10.0 billion, awarded July 31, 2025.
Does a $20 billion contract ceiling mean the company will receive $20 billion?
No, and the Army says so directly: the figure “represents the maximum potential value, not an obligated amount.” A ceiling authorises spending; task orders obligate it. Anduril’s disclosed 2025 revenue across all customers was $2.2 billion. Reading ceilings as bookings is the most common way defense AI market size gets overstated.
Which AI companies can run on classified Department of War networks?
As of May 1, 2026: SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Amazon Web Services and Oracle signed agreements to deploy on Impact Level 6 and Impact Level 7 environments. IL6 covers classified cloud workloads; IL7 covers top secret and the most sensitive national security information. Palantir’s Maven Smart System already operates across multiple classification domains.
Why were the Pentagon’s Anthropic contracts cancelled?
The Department designated Anthropic a supply chain risk in March 2026 after the company declined to permit its models to be used for mass surveillance of Americans or for fully autonomous weapons without human authorisation over targeting decisions. Anthropic sued on March 9, 2026, and on August 27, 2026 a federal judge ruled the designation unlawful and ordered it removed. Separate litigation in Washington, D.C. was still pending as of August 28, 2026.
Are defense AI startups actually displacing the traditional primes?
On contract ceiling, partially: Anduril’s $20 billion single-vendor ceiling exceeds what many legacy primes accumulate in a decade. On delivered revenue, not yet — Anduril’s $2.2 billion in 2025 is a fraction of any top-five prime’s annual sales. What has changed structurally is the contract vehicle: enterprise agreements let a software-first vendor sell across a service without competing task order by task order.
How much of the Pentagon’s AI budget actually reaches AI software vendors?
Less than a fifth. Of the $58.5 billion requested for AI and CJADC2 in FY2027, $46.0 billion (78.63%) funds the government-owned AI Arsenal compute infrastructure. The application-layer lines — AI integration, Maven Smart System and Joint Fires Network, and AI research and development — total $11.3 billion, or 19.32% of the request.
What is a program of record, and why does it matter for an AI vendor?
A program of record is a formally designated, budget-line-funded capability rather than an experiment. It carries stable multi-year funding and standardised adoption across services, and it has to be actively cancelled to disappear. Palantir’s Maven Smart System received the designation in a March 9, 2026 memorandum. No frontier AI lab currently holds one; all four CDAO awardees hold two-year prototype other-transaction agreements.
How do European defense AI companies compare with U.S. ones on contract scale?
They are an order of magnitude smaller and legally constrained differently. Helsing’s German HX-2 framework began at roughly €269 million with option values reported up to €1.46 billion, before the Bundestag budget committee capped the framework at €1 billion per producer and required fresh parliamentary approval beyond €2 billion combined. A U.S. enterprise agreement’s ceiling is a standing authorisation; a German framework ceiling can be removed by a budget committee after award.
Does venture funding predict military AI contract wins?
Not reliably in this dataset. Shield AI raised at a $12.7 billion valuation, a 139.6% increase in twelve months, without a disclosed nine-figure Department contract ceiling. Helsing raised at $18 billion against a €269 million initial award. The two companies with the largest disclosed ceilings — Anduril and Palantir — are also the two with the longest delivery histories, which suggests the causation runs from fielded systems to capital rather than the reverse.
Related Axis Intelligence Research datasets: Military AI statistics · Military drone statistics · Research index · Autonomous Weapons Regulation Statistics