Anduril Industries

Anduril sells battlefield software, drones and missiles and plans to make submarine components. Follow the money, the machines and the political relationships to see where its power comes from and what the evidence can prove.

The $20 billion headline

On 13 March 2026 the US Army announced a single contract vehicle through which eligible agencies can order Anduril’s software, hardware, infrastructure and support. The award notice puts a $20 billion upper limit on the vehicle, with five base years and a five-year option. The Army said it previously managed more than 120 separate procurement actions for the company’s offerings. What changed that day was the route by which the government can buy. The amount it ultimately buys depends on the orders and obligations that follow. ^[1] ^[2]

That distinction makes the story more interesting. A broad ordering vehicle can put a supplier in reach of many programmes without anyone having committed the headline amount. Anduril’s most consequential pitch is that the same software can connect the equipment those programmes buy. A customer might first acquire a sensor or aircraft and eventually rely on the supplier’s interface to see what other systems are doing. That is a plausible source of leverage. How much leverage Anduril has already gained depends on the contracts, the technical interfaces and the customer’s ability to switch. ^[2] ^[3]

Follow the money through seven gates

A ceiling sets the maximum potential scope. An order tells us what the customer is buying. An obligation records the government’s legal commitment of funds. Delivery means something was provided. Acceptance means the customer formally accepted it under the applicable terms. Payment records cash disbursed, and recognised revenue is an accounting measure on the company’s side. On a large programme, those stages can happen on different dates and produce different numbers. ^[1] ^[2] ^[4] ^[5] ^[6]

The earlier audit identified three example orders under the Army vehicle. Its captured figures were about $142.04 million in obligations on Lattice order W9128Z26FA005, $14.59 million on NGC2 order W9128Z26FA022 and $1.90 million on counter-drone order FA441726F0120. Those are historical snapshots from the investigation. The linked award records identify the orders, and the figures can move with modifications or deobligations. Taken alone, they establish neither the full amount obligated across the vehicle nor the cash Anduril received. ^[4] ^[5] ^[6]

The wider audit also had two historical federal-obligation compilations around $2.98 billion and $3.008 billion. We exclude both from our headline because the underlying extracts were taken on different snapshots and have not been reconciled. An auditable total needs one cutoff, the contractor and subsidiary entity IDs, the parent-and-order relationships, and the full modification history. The public award pages are a starting point. A $0 outlay field alone cannot settle whether an individual invoice has been paid.

Who owns the upside?

Anduril began in 2017. Its founders were Palmer Luckey, Brian Schimpf, Trae Stephens, Matt Grimm and Joseph Chen. The group brought together experience in consumer hardware, venture capital and government-facing software. May 2026 brought a $5 billion Series H at a company-announced $61 billion valuation, led by Thrive Capital and Andreessen Horowitz. Anduril reported about $2.2 billion in 2025 revenue. Those numbers describe fundraising, implied equity value and sales respectively. The public evidence reviewed here does not establish consolidated profit, cash burn, debt terms or runway. ^[7] ^[8]

In October, Bloomberg reported an employee-share tender at an implied valuation of about $90 billion, citing people familiar with the matter. An employee selling existing shares gets liquidity. The proceeds do not automatically flow into Anduril’s bank account. The tender figure is a separate reported pricing signal, with its own terms and limits. It cannot establish a new primary raise, an IPO timetable or what a public market would pay. ^[9]

The Palantir relationship is more tangible than a shared circle of investors. Palantir disclosed a $40 million investment in Anduril’s Series G, representing less than 1% of outstanding preferred and aggregate capital stock at the time described, and $17.66 million it received from Anduril for Palantir products and services in 2025. Its proxy separately said Peter Thiel may be deemed to share voting, investment and disposal power over more than 10% of Anduril through related entities. One filing establishes commercial ties, an investment and an attributed interest. The current fully diluted cap table, special voting terms and board rights remain open questions. ^[10]

Lattice and the power of the interface

Anduril’s Lattice documentation describes a system organised around entities and tasks. An entity might be a sensor, a vehicle, a tracked object or a geographic feature. A task is an instruction that a connected agent can receive. Put plainly, the software gives different systems a way to describe what they see and what they can do. A track from one sensor can become part of a shared picture rather than living only on a screen beside that sensor. The customer’s ability to integrate and replace individual components is central to whether that picture creates flexibility or dependence. ^[3]

In June 2026 the Army announced a common data baseline for Next Generation Command and Control, following operational validation with two divisions. In October Anduril announced work worth up to $1.8 billion over five years to expand fielding, starting with I Corps. Defense News independently reported the award. That figure describes the stated potential value of that work. Counting it on top of the Army-wide ceiling would require checking how the relevant contract and orders relate. ^[11] ^[12] ^[13]

An API can make new equipment easier to connect. It leaves a different set of questions unanswered. Who can export the data and audit logs? Can a government team operate or rebuild the system without the vendor? Who can deploy updates or restore a prior version? What happens to conflicting tracks when the network splits and reconnects? The public product documentation identifies interfaces. Contract terms and realistic tests would tell us how much control a customer retains. ^[3] ^[11]

The product portfolio has different levels of proof

Fury, Anduril’s aircraft for the US Air Force’s Collaborative Combat Aircraft programme, reached a tangible manufacturing milestone in July 2026 when the first Ohio-built example rolled off the Arsenal-1 line. That proves a first build at the facility. Serial delivery and operational readiness require further evidence, including customer acceptance and sustained throughput. ^[14] ^[15]

Barracuda-M provides a different kind of milestone. The Air Force Research Laboratory verified compliance with the Weapon Open Systems Architecture standard across 14 domains. That gives weight to the claim that parts of the missile’s architecture can be made interoperable. The test concerned the architecture standard. It did not measure combat effectiveness, and a production announcement is a separate source from a count of accepted rounds. ^[16] ^[17]

ALTIUS spans reconnaissance and strike variants, while Ghost is a separate surveillance aircraft family. Reuters reported two ALTIUS test crashes in 2025 and difficulties involving Ghost, including in Ukraine. Anduril described the examples as isolated across hundreds of tests. The incidents deserve attention. To estimate reliability, readers would need a denominator such as comparable launches or mission hours, a common definition of failure and the software or hardware versions involved. The public record reviewed here does not support a fleet-wide rate. ^[18] ^[19]

Can the factories deliver?

Arsenal-1 in Ohio is the test of whether Anduril can repeatedly manufacture and deliver complex hardware at a predictable cost. The first Ohio-built Fury gives us a dated milestone. The planned campus footprint, the rated capacity of a line and the volume of customer-accepted aircraft are three separate measures. Anduril describes ArsenalOS as a digital backbone spanning production and quality workflows. The way to test its advantage is to compare manufacturing yield, rework, lead time, accepted deliveries and cost over multiple periods. ^[14] ^[15] ^[20]

Arsenal-2 in Maryland is a different industrial bet. Maryland announced $3.7 billion in planned private investment at Sparrows Point for a roughly two-million-square-foot shipyard, and Anduril and Reuters reported a $2.9 billion Navy agreement. The work concerns components for Virginia-class submarines. Adding those figures does not produce a verified cash-spend total. Anduril described a 40% government equity interest in a dedicated operating entity associated with the project, a separate ownership question from the parent company. The next meaningful checks are funding milestones, naval qualification, installed equipment and acceptance of the first components. ^[21] ^[22] ^[23]

Three procurement routes, four countries

United Kingdom and Ukraine. In March 2025 the UK Ministry of Defence announced nearly £30 million in contracts with Anduril UK for ALTIUS-600M and ALTIUS-700M for Ukraine, backed by the International Fund for Ukraine. The announcement identifies the buyer arrangement and the variants. The company’s UK filing history is a separate source for the subsidiary. Neither record tells us the number of successful combat missions or the profitability of the US parent. ^[19] ^[24]

Australia. The government announced an A$1.7 billion Ghost Shark programme for the large uncrewed undersea vehicle. Ministers later described the project’s move from prototypes into production. That is a different level of commitment from an early design study. The public material reviewed here still leaves the total customer-accepted operational fleet, maintenance burden and availability unresolved. ^[25] ^[26]

Taiwan. In August 2026 Taiwan News reported that a procurement covered 1,554 ALTIUS-700M and 478 ALTIUS-600ISR systems, or 2,032 in total. The Taipei Times later reported on efforts to build a non-China supply chain for batteries and chargers. That detail is an early clue to what is involved in sustaining a fleet. A country can sign for thousands of aircraft and still have to secure charging equipment, spares, training and support before it can use them at that scale. The quantities cited describe procurement as reported at the time. ^[27] ^[28]

Political access and what it can prove

There are distinct records for lobbying, political contributions and procurement. The earlier audit tracked company and outside-firm lobbying disclosures, and the Anduril PAC can be inspected in FEC records under committee ID C00763623. Public disclosures report $490,000 in Anduril lobbying expenses for the third quarter of 2026. Outside-firm income can overlap with the company’s own expense figure, so adding it mechanically risks counting the same work twice. A founder’s personal contribution and a PAC disbursement are separate transactions. ^[29] ^[30]

The difficult question is how to test influence on a specific decision. A credible chain would match the timing and subject of lobbying to meeting calendars, procurement communications, evaluation criteria, competing bids, recusals and the award rationale. The public records cited here establish that political activity and commercial relationships exist. They do not establish that a payment or relationship secured any named award. The Army’s single-award route is worth examining on its procurement record, including the rationale for consolidating requirements and what competition remains available elsewhere. ^[1] ^[2] ^[10] ^[29] ^[30]

Surveillance and autonomy beyond the battlefield

Persistent surveillance creates practical accountability questions. Which sensors produce identifiable records? Who sees the original material and inferred tracks? How long is either kept, and which organisations can retrieve or challenge it? The UK Information Commissioner decision cited in the earlier audit concerned the Home Office’s handling of a Freedom of Information request. It does not establish a data-protection breach by Anduril. The distinction matters to anyone trying to work out who held the information and which decisions were actually reviewed. ^[31]

Autonomy also covers different tasks. Navigating a vehicle, correlating tracks, recommending an action and authorising a use of force each raise different questions. A meaningful assessment would ask who can authorise action, override a recommendation or halt the system, what happens when identification or communications fail, and whether an independent reviewer can reconstruct the decision afterwards. The public evidence is uneven across products and deployments. Contract-specific data rights, export permissions and operational rules deserve separate scrutiny. ^[3] ^[16]

What would change our judgement?

The public record shows substantial financing, a government route to buy at scale, work moving towards wider software fielding and several measurable manufacturing or test milestones. The open question is whether those advantages add up to durable dependence, profitable production and dependable operation. Each part of that question needs a different record.

Money. A reconciled, dated ledger of orders, modifications, acceptance and payments would show which headline values became real purchases.
Manufacturing. Monthly customer-accepted deliveries, defects, rework and unit costs would show whether first builds are becoming repeatable production.
Control. A government-led demonstration that a competing sensor or platform can be swapped in, including under disrupted communications, would test the portability promise.
Performance. Comparable flight or mission data with denominators, version history and independent incident review would make reliability claims testable.
Influence. Complete procurement rationales, contemporaneous meeting records and any recusals or waivers would let readers assess a specific decision on its merits.

The question we will keep following is this. When a government buys a new aircraft, sensor or interceptor, how much freedom does it have to choose the next one? The answer will tell us more about Anduril’s long-term power than a valuation or an award headline on its own.

Sources and evidence

This investigation uses a research snapshot dated 9 November 2026. Official records establish what an institution filed, announced or tested. Company figures and plans are attributed, reporting is identified, and unanswered questions remain open. Dollar amounts are US dollars unless marked otherwise. The PDF includes the same findings and an at-a-glance contract ledger.