How Washington took root in French defense

The subject is not new. The digital dependence of the French armies on the United States has been the subject of alerts, reports, and discussions in committees. What changes is the context: a Trump administration that assumes transforming technological dependencies into pressure levers, NATO which has just adopted an American AI platform for its command, and the General Directorate of Internal Security (DGSI) which has renewed its contract with Palantir for three more years. All signals are red.

Palantir, a different provider: It was supposed to be temporary. In 2016, after the November 13, 2015 attacks, the DGSI urgently turned to Palantir. No French tool was available to process the large amounts of data needed to track jihadist networks on such a scale and speed. The American publisher was imposed by default. Nearly ten years later, the contract has just been renewed for an additional three years.

This renewal ended up in the report on the “dependencies of France vis-a-vis foreign countries” that deputies Aurelien Saintoul (LFI) and Francois Cormier-Bouligeon (Ensemble pour la Republique) have just submitted to the Commission of National Defense and Armed Forces and that we were able to consult.

Their verdict is clear. They point out “the danger posed by the renewal of the DGSI’s contract with Palantir” and identify three concrete risks: a possible extraction of French data to the United States, a risk of cut-off if bilateral relations deteriorate, and the gradual stifling of national alternatives, condemned to never dislodge an established holder. What the two deputies add, which is not trivial: dependence on the United States cannot be treated like that which exists with any other partner. It is of a different nature. And of a different scale.

To understand why this contract poses a problem, one must look at what Palantir is. Founded in 2003 in Silicon Valley, the company received financial support from In-Q-Tel, a fund linked to the CIA from the start. It works with most American federal agencies (CIA, FBI, Pentagon), and its co-founder Peter Thiel is a figure in the conservative American right, close to Donald Trump. Its Gotham platform, deployed at the DGSI, allows for the crossing and analysis of heterogeneous data on a large scale. An tool recognized as efficient by the services themselves. However, a tool that France does not control the source code, architecture, or real maintenance conditions in the long term. The rapporteurs are direct: “control of internal situation must remain sovereign”.

A tender was launched in 2022 to find a national successor under the code OTDH (Heterogeneous Data Processing Tool). Three candidates were selected in 2023: Athea (Thales-Atos/Eviden alliance), Blueway, and ChapsVision, often presented as the “French Palantir”. None of them won the contract. The DGSI chose to continue the operational continuity, and the reasoning is understandable: migrating an intelligence platform in activity is not a trivial operation. But the rapporteurs refuse for this constraint to become a permanent excuse. The longer a system is in place, the higher the cost of replacing it becomes. The rapporteurs want to break out of this logic.

ITAR, Cloud Act, minerals: the U.S. arsenal What makes dependence on the United States particularly delicate is that it is not just a matter of market share. It comes with a legal and operational arsenal that Washington uses, without hesitation, to maintain its technological and commercial superiority.

The concrete examples are plentiful in the report. In 2012, the upgrading of French AWACS was blocked by the United States due to the sensitivity of the modules concerned. On the Reaper drones of the Air and Space Army, pilot training had to be exclusively carried out on American soil, and when the needs of the US Air Force increased, the number of available courses for Europeans simply decreased. More recently, the development of the VMAX hypersonic glider by ArianeGroup was delayed by a year: Washington had refused access to the suborbital launchers necessary for testing, using French dependence to hinder a sensitive program. Since then, ArianeGroup has launched its own suborbital launch program to regain its autonomy. On the F-35, purchased by several European partners, doubt persists about the American ability to act remotely on the aircraft’s cloud or on its operational maintenance.

In addition to these operational risks, there are legal constraints that regulate dependencies in a very tangible way. The ITAR (International Traffic in Arms Regulation) and EAR (Export Administration Regulations) regulations give the United States extraterritorial competence over any equipment containing an American component, even a minority. This is not a theoretical possibility. A parliamentary report from 2019 pointed out the use of ITAR as “a weapon in the service of U.S. commercial and strategic interests to the detriment of their European allies,” citing the difficulties in selling the Rafale to Egypt, blocked by the American components integrated into the MBDA missiles of the offer. The alarms were there. No one acted. Today, French BITD companies send nearly a thousand license requests each year to American authorities. And any failure, even unintentional, exposes companies to fines worth several million dollars.

And it is not only in the digital domain that this strategy of coercion is expressed. It now extends to critical raw materials and minerals. Since the first term of Donald Trump, the United States has systematically built a policy of cornering global strategic resources. Successive presidential decrees since 2017 have laid the groundwork for a revival of national mining production.

Trump’s return has moved this strategy into a sharply coercive logic. The annexation aspirations of Canada (which holds 34 critical minerals) and Greenland, whose subsoil contains 25 of the 34 raw materials deemed strategic by the European Union (EU), are not mere rhetorical provocations. The agreement signed on April 30, 2025, with Ukraine illustrates this logic: Washington obtained preferential rights to future Ukrainian mining projects. In exchange for military assistance that, according to the Kiel Institute, had already dropped by 99% compared to previous years.

For France and Europe, the overall picture is worrying. The race for critical minerals is not a game between two but three: Washington seeks to secure its supplies against a China that controls 90% of the global rare earth refining and has started restricting exports since 2025, even to the United States themselves. In this triangular competition, Europe risks being caught between two powers, each of which has already started to lock up resources critical to the manufacture of GPUs, semiconductors, and batteries. The America First Arms Transfer strategy also intends to use arms sales “intentionally as a tool of American foreign policy”. And the Cloud Act, adopted as early as 2018, allows U.S. judicial authorities to access data stored by providers subject to American law, even on European servers.

NATO under American software: If the situation at the DGSI is worrying, the one playing out at NATO is even more so. In March 2025, the Alliance announced the adoption of the Maven Smart System NATO, also developed by Palantir, to equip its allied operations command. Contract signed in six months: a record for an organization whose acquisition procedures are normally an endurance sport.

Maven Smart System is not an administrative management tool. It is a command and control platform (C2) integrating generative artificial intelligence (AI), machine learning models, and real-time data fusion capabilities. It touches the core of how NATO thinks and conducts its operations: intelligence, planning, situational awareness, and rapidity of decision cycles.

For the rapporteurs, the signal is serious. They write bluntly that the choice of Maven represents “a serious risk to French decision sovereignty” and that this system, “once implemented, would completely close the door to European solutions.” And the report goes even further: “France’s decision-making freedom would necessarily be severely constrained, particularly in the face of a loss of access risk.” In other words, if Washington were to cut access, allied staff would work blindfolded.

This is not the only signal. The report also notes more broadly that “the introduction of Palantir and the certification of Apple devices for use by agents working at NATO illustrate the influence of the United States on the institution.” For twenty years, the United States has “elaborated and formalized” the major doctrinal evolutions, directly shaping the format of European armies. Maven is therefore not an anomaly. It is the culmination of a long trend.

The French response exists: Artemis IA, developed by the company ATHEA and financed to the tune of 700 million euros in the military programming law (LPM). But the rapporteurs consider it “far too weak in the face of American technology power.” The numbers are staggering: American digital giants have mobilized over 700 billion euros in artificial intelligence in recent years, equivalent to 1.5% of the EU’s GDP. From the French side, there are signals: Mistral AI became the first unicorn of the French Tech in September 2025 with a raise of 1.7 billion euros, and AMI Labs, Yann LeCun’s startup, raised over one billion dollars since its launch in March 2026. Ambitions are high. But they remain on a different scale. The gap is not a delay of a few quarters.

Under Palantir, the iceberg: Palantir attracts attention, but it is only the visible part of a much larger structural problem. The framework agreement signed with Microsoft in 2008 structures the digital daily life of 220,000 Ministry of Defense employees through INTRADEF and the Active Directory. A real advantage at the time, it allowed the merging of the Ministry’s historical intranets. A real vulnerability today. The rapporteurs don’t evade this: “this situation, previously accepted, is challenged by the evolving behavior of strategic actors, who hesitate no longer to use all levers as pressure means.” And they admit that “a total and immediate exit from Microsoft is not credible without compromising the operational ability of the armed forces.” Sovereign alternatives are emerging (LaSuite, Hexagone) but their adoption remains very marginal.

On hardware, the picture is just as telling: 85% of the Ministry of Defense’s equipment purchases on IT equipment go to ten foreign suppliers. Processors, servers, hardened hardware for the war center: the supply chain structurally escapes any national control in the short term.

The cloud illustrates the paradox of the situation well. The armed forces have sovereign infrastructures for their most sensitive data, which is crucial. But these sovereign infrastructures rely on American components: Intel processors, Nvidia graphics cards, AMD chips. “Having the infrastructure does not necessarily mean being autonomous,” the rapporteurs soberly note. By the end of 2025, Google Cloud and Oracle also secured contracts with the NCIA, NATO’s information and communication agency. The pressure towards American homogenization of the allied cloud is only increasing.

Then there is the question of GPUs. All French AI solutions, including Artemis IA, rely on graphic accelerators produced by Nvidia, which almost has a global monopoly. This is the hardware foundation on which all the software sovereignty that France seeks to build relies. It is safe to say that this sovereignty remains, for now, partially dependent on the goodwill of an American company listed on the stock market.

If Washington turns off the tap: Aurelien Saintoul and Francois Cormier-Bouligeon do not advocate a total technological retreat on all fronts. They know that total digital autonomy is an illusion and say it clearly. What they ask for is more precise and, in a way, more demanding: a real strategy to progressively reduce the most critical dependencies, with a held calendar and adequate means.

The report also highlights a blind spot rarely mentioned in this debate: it is precisely where France has not yet developed its own capabilities that dependence on the United States tends to form. And once the technology is adopted, it becomes very difficult to get rid of it; the cost of a national alternative seems disproportionate, and the historical alliance with Washington mitigates the sense of urgency. This mechanism is not new. In 2019, deputy Francoise Dumas noted that several EU member states displayed their support for European strategic autonomy while systematically buying American. Six years later, the observation has worsened, with digital now being included. It is this mechanism, repeated on dozens of segments for thirty years, that has built the architecture of dependencies described in today’s report.

For years, dependence on Palantir has been presented as temporary. French alternatives have been announced as imminent. Contracts have been renewed. If tomorrow Washington decided, for any reason, to turn off the tap; by diplomatic pressure, commercial arbitration, or simply because American priorities have changed, the question would not be whether France has an alternative. It would be whether it would have time to find one.