Military AI in 2026: Anduril valued at 61 billion, Palantir at the Pentagon, Ukraine as laboratory

The shift that is changing global defence

In 2018, when the press talked about military innovation, it mentioned the F-35 (the most expensive stealth fighter jet in American history, with an estimated total cost of more than 1.7 trillion dollars), the Columbia-class submarines (the new generation of American nuclear-armed ballistic missile submarines), the B-21 Raider bomber. These programmes embodied what was called the "military-industrial complex" (a term popularised by President Eisenhower in 1961, referring to the alliance between the armed forces, the arms industry and political power).

By 2026, the narrative had flipped. People talk about Anduril's Roadrunner-M, the expansion of Palantir's Maven Smart System contract, Shield AI V-BAT operations in Ukraine, Helsing HX-2 drone production lines. The military-industrial complex is becoming the military-AI complex. And Silicon Valley startups are in the process of replacing Lockheed Martin and Boeing in the media narrative, if not yet in the contracts.

This article gives an honest assessment of that shift, which bears directly on our subject of AI. The financial figures are massive, the ethical stakes are major, and Europe is largely absent. Here is the state of the market in 2026.

Anduril: the 61-billion-dollar startup

The case of Anduril Industries is probably the most emblematic of this new economy. Founded in 2017 by Palmer Luckey (an American entrepreneur who had sold Oculus VR to Facebook for 2 billion dollars in 2014) and three former Palantir employees, the company has become one of the most highly valued private startups in the United States in 8 years.

The most recent funding round, announced in May 2026, raised 5 billion dollars at a total valuation of 61 billion. The round was led by Thrive Capital and Andreessen Horowitz (two of the most powerful venture capital firms in Silicon Valley). The company expects to go from 2.2 billion dollars in revenue in 2025 to around 4.3 billion in 2026. It now employs more than 8,000 people.

The products Anduril sells:

• Bolt autonomous drones: small combat drones that launch vertically, weigh a few kilograms, and dive at 85 degrees onto their target. Unit cost: a few thousand dollars, against hundreds of thousands for traditional missiles.
• Altius 600M and 700M: loitering munitions (drones able to loiter over an area before detecting and striking a target). The British government has bought some for Ukraine.
• Ghost: a heavier reconnaissance and attack drone.
• Fury: an autonomous combat aircraft designed as a "wingman" for human pilots.
• Roadrunner-M: an anti-drone defence system, vertically launched, able to return to base if the mission fails.
• Lattice: the software platform that orchestrates all these systems, allowing a single operator to manage several dozen drones simultaneously.

Anduril is not yet profitable. The company expects around $1 billion in losses in 2026, and does not expect profitability before 2030. But the ambition is clear: to become the Lockheed Martin of AI defence, replacing classic industrial processes with fast software development cycles.

Palantir: the platform that knows everything

The other major player is Palantir Technologies. Co-founded in 2003 by Peter Thiel (entrepreneur and investor, co-founder of PayPal, one of Facebook's first investors) and funded in its early days by In-Q-Tel (the investment fund linked to the CIA), Palantir has been listed on the stock exchange since 2020 and was valued at more than $280 billion in 2026.

Palantir's flagship product is called Gotham, a platform that analyses and correlates enormous quantities of data from heterogeneous sources. The aim is to eliminate "data silos" (the technical barriers that keep information collected in different contexts separate). In practice, Gotham can cross-reference satellite images, intercepted communications, troop movements, combatant identities and financial data to produce a real-time map of a theatre of war.

The other major product is Maven Smart System, the Pentagon's AI system for satellite and reconnaissance image analysis. The Maven contract was extended in 2025 and is worth several hundred million dollars a year. It was Maven that caused the scandal at Google in 2018, when 4,000 employees signed a letter calling for the project to be dropped. Google pulled out. Palantir took over the contract and grew it.

Palantir sells its tools to several governments: the United States of course, but also Israel, the United Kingdom, Ukraine, Poland and the Emirates. The company is also used by police forces (NYPD, LAPD) and civilian intelligence services. As we documented in our article on the US-China AI war, Palantir is one of the central players in the global surveillance infrastructure.

Ukraine: the world's laboratory for autonomous weapons

Here is the angle that absolutely must be understood. The war in Ukraine has become, despite the tragedies it entails, the world's largest laboratory for autonomous weapons. There are several reasons for this.

First, Ukraine has become a state that massively adopts Western and Israeli military AI technologies. The Delta network (a Ukrainian real-time battlefield management system) integrates data from drones, sensors and satellites, and allows commanders to make decisions within seconds. On 15 May 2026, Kyiv announced it would share data collected during the war with its allies and certain private companies to train military AI systems. This is probably the largest dataset in the world on modern warfare.

Second, Ukraine itself produces massive technological innovations. Several Ukrainian startups have become global benchmarks within a few years. OSIRIS AI, which presented its UEB-1 interceptor drone at the Xponential Europe trade fair in March 2026 in Düsseldorf, is now valued at several hundred million dollars. The Granta GA-10FPV-AI, a Lithuanian first-person-view drone with autonomous AI, is being used en masse on the front line. Lithuania, Estonia and Poland have become drone production hubs, sometimes at unit costs below $1,000.

Third, the Russian-Ukrainian asymmetry forced innovation. Ukraine could not match Russia in numbers of tanks, artillery or pilots. It bet on cheap drones and algorithms. The result: according to several military estimates, more than 80% of the destruction of Russian equipment in 2025 was caused by drones. Conventional tanks have become almost obsolete in the face of a swarm of FPV drones (First Person View, racing drones piloted with a headset that transmits the view from the onboard camera).

The other players that matter

The military AI market is not dominated by Anduril and Palantir alone. Several other major players shape the industry.

Shield AI. Founded in 2015, based in San Diego. Its flagship product is the V-BAT, a vertical take-off and landing reconnaissance drone, used extensively in Ukraine. Valued at more than $5 billion in 2025.

Helsing. German startup based in Munich, founded in 2021. Specialised in AI for European defence. Produces the HX-2 attack drone on the production line. Probably the only true European champion in the sector. Valued at €5.4 billion in 2025.

Saronic, Saildrone, HavocAI. Three American startups specialising in autonomous maritime drones. With growing tension in the South China Sea (an area disputed between China and several neighbouring countries such as the Philippines, Vietnam and Taiwan), demand is exploding.

Skydio. Reconnaissance drones for the military and police market. Founded in 2014, valued at $2.2 billion.

Vannevar Labs, Distyl AI. Direct competitors to Palantir in defence software.

Mach Industries. Younger (founded in 2023), specialising in missiles produced quickly and at low cost. Raised $335 million in 2024.

Hadrian. Manufacturing of precision metal parts for defence, with factories highly automated by AI.

In total, the American defence tech sector raised more than $15 billion in 2025, according to compilations by PitchBook (a financial data platform on venture capital). This concentration of capital on military AI is unprecedented in the sector's recent history.

The ethical debate that must be named

Beyond the figures, the emergence of autonomous weapons raises fundamental ethical questions that the international community cannot resolve.

Question 1: should the decision to kill remain human? This is the central debate. "LAWS" (Lethal Autonomous Weapons Systems, the official UN term) refers to weapons that can identify, track and strike a target without final human validation. Almost all current AI weapons still include a human in the loop, but that barrier is eroding. Several states (Russia, China) are reportedly ready to deploy fully autonomous LAWS. Western democracies are hesitating, but are gradually pushing back their limits.

Question 2: liability in the event of error. If an autonomous drone strikes the wrong target (a school instead of a military HQ), who is responsible? The commander who launched the mission? The company that sold the drone? The developer of the algorithm? International humanitarian law (the Geneva Conventions) has no framework for these situations.

Question 3: proliferation. FPV drones costing 1,000 dollars have become accessible to any actor, state or otherwise. How do you prevent Daesh, Hezbollah, Mexican cartels, or simply organised criminals from using these technologies? Several cases of attacks by improvised drones are already documented. The technical barrier is collapsing.

Question 4: the drift towards permanent war. When war costs less (cheap drones) and risks fewer politically costly lives (no soldiers on the front line), governments may be tempted to use it more readily. This logic was visible in the use of drones by the Obama and then Trump administrations in Yemen, Pakistan and Somalia. With autonomous drones, the threshold becomes even lower.

The UN has been trying since 2014 to negotiate a treaty banning lethal autonomous weapons. More than 30 states support such a ban. But the United States, Russia, Israel and the United Kingdom oppose it. No agreement is in sight.

The big absentee: Europe

As in many other AI domains documented in our article on the French lag and our analysis of the European position, Europe is massively behind on military AI.

France has a few players: Thales (French defence electronics group, one of the world's leaders in the sector), Safran (aeronautical engine maker, which also builds drones), Dassault Aviation (manufacturer of the Rafale and of drones). But none of them is in the race among the new AI-native startups. The European FCAS programme (Future Combat Air System, a Franco-German-Spanish project for a sixth-generation fighter jet) costs tens of billions and is progressing very slowly.

Germany has Helsing as its champion, as mentioned above, but it is Europe's sole exception. The United Kingdom, despite its military history, has no native equivalent of Anduril or Palantir. Italy, Spain, the Nordic countries: no leading player.

Consequence: if Europe enters a major conflict, it will depend largely on American and Israeli technologies. This strategic dependency is one of the major political challenges for the next 5 to 10 years. The debate on "European strategic autonomy" championed by France is precisely about this subject, but without concrete results for now.

Anthropic and the ethical limit of AI labs

Here is a little-known but important angle. Several commercial AI labs have put contractual limits on the military use of their models. Anthropic, the creator of Claude that we covered extensively in our article on Opus 4.7, long refused to sign contracts with the Pentagon, unlike OpenAI.

That position shifted in 2024-2025. Anthropic eventually accepted limited contracts with AWS Bedrock to serve US government clients, but with clauses prohibiting the use of Claude for "lethal decision-making applications". OpenAI, for its part, removed in January 2024 the clause in its terms of use that specifically banned military use. Google had done the same in 2023.

Several employees at Google, OpenAI, Microsoft and others publicly protested against these reversals. In 2024 and 2025, several open letters gathered thousands of signatures, to little effect. The military market is too lucrative for companies to forgo it, especially against a backdrop of growing geopolitical tension.

What it changes in practice

Beyond the broader ethical debate, several very concrete implications are emerging for citizens, investors and policymakers.

For citizens. Future conflicts will involve autonomous weapons on a massive scale. Protection against these weapons (anti-drone jammers, detection systems) will become a matter of civil defence. "Remote" wars are politically easier to start, which could paradoxically increase their frequency.

For investors. The AI defence sector is in a full-blown bubble, but with real fundamentals. Anduril and Palantir are profitable or on their way to being so, unlike many consumer AI startups we discussed in our article on the AI bubble. Global military budgets are rising. The sector has longer contractual visibility than the tech average.

For policymakers. The window to regulate autonomous weapons is closing fast. Once these technologies are deployed on a massive scale by every major power, international regulations will be practically impossible to enforce. As we explored in our article on the limits of AI regulation, the time to act is now or never.

For non-defence tech companies. AI talent is precious and limited. The salaries offered by Anduril, Palantir and Helsing (often $200,000 to $500,000 a year for senior engineers) draw in some of the best talent that would otherwise have worked at OpenAI or Google. This competition for talent will probably intensify.

The legacy being built silently

Even more than the consumer AI bubble, the emergence of the military-AI complex is probably the most lasting transformation currently under way in the sector. Why? Because military budgets are shielded from economic cycles, because the players are on multi-year contracts, because the structural geopolitical competition between the US, China and Russia guarantees demand for decades.

When we look back on 2026 in 20 years, we will probably say two things. First, it was the year defence became software. More than hardware, code and algorithms became the main determinant of military power. Second, it was the year Europe definitively positioned itself as a customer rather than a producer of this technology.

It remains to be seen whether this transformation will produce more stability (precision weapons cause fewer civilian casualties than mass bombardment) or more conflict (war becomes more accessible). Both theses have serious defenders. The answer will depend on political choices over the next 5 to 10 years.

To go further, you can consult our article on the US-China AI war, which places these issues within the great geopolitical rivalry of the 21st century, and our analysis of AI ghost workers, which reveals another hidden side of the AI industries. The subject of defence is probably one of the most important to follow in the years ahead.