Artificial Intelligence at the Pentagon: Google for Office, Palantir for Operations.

The United States Department of Defense has decided to get serious with generative artificial intelligence and has launched a dedicated platform: GenAI.mil. It is the new digital “condominium” where all frontier AI models made available to the US military bureaucracy coexist.

The first tenant is Google Cloud Gemini for Government, a “hardened” version of the Mountain View model, designed for government use. Through GenAI.mil, access to Gemini reaches the workstations of millions of employees, both military and civilian, with the declared goal of creating an “AI-first” workforce: less time spent on documents, bureaucracy, and research, more time on analysis and decision-making.

For now, the official scope of use is clear: unclassified work, administrative activities, personnel onboarding, document formatting, research support, video and image analysis in low-classification contexts. Everything takes place in a certified environment to handle sensitive but not top-secret data, with the politically crucial promise that Pentagon data will not end up training Google's public models.

In summary, GenAI.mil is born as a single entry point to generative AI for those who live and work in the US military machine, and Gemini is the first model placed on the pedestal.

Google's Return in Uniform (and Why It's Not the AI of Weapons)

The choice to place Google at the center of the new platform is not neutral. It is a grand return after the controversies years ago over Project Maven, where internal employee pressure led the company to withdraw from more explicitly military contracts.

Today, Google re-enters the defense perimeter with a contract worth hundreds of millions of dollars to provide AI capabilities to the Pentagon and, above all, with a different narrative positioning: Gemini is not presented as “the brain of drones,” but as the super-office assistant of the military world.

On GenAI.mil, Gemini for Government is described as a tool capable of speeding up the daily life of defense: advanced research in large volumes of documents, text generation and rewriting, support for planning and analysis of visual material, always within an unclassified and controlled context. The official emphasis is all on productivity and the bureaucratic “battle rhythm,” much more than on operational targeting.

Here we already see the first half of the Google–Palantir duo: Google takes charge of the visible surface of AI, the one that touches every desk, every email, every document.

Where Has Palantir Gone? From NATO Brain to US Backbone

Your question is direct: “But Palantir, which NATO had taken for itself, where has it gone?”

The answer is simple: Palantir has not disappeared, on the contrary. It has become even more “institutional”.

On one hand, NATO has adopted a platform based on Palantir technologies — often referred to as an evolution of the Maven project in the allied context — for military planning, data fusion, and decision support for operational commands. We are talking about a tool that brings together intelligence flows, images, field reports, scenario analysis, and uses machine learning and generative models to reduce decision-making times.

On the other hand, Palantir continues to be deeply rooted in the United States itself, with large contracts with the army — from logistical and operational data analysis to the AI Platform (AIP) that aims to provide a unified view of the battlefield in near real-time. In all these cases, it is not “office AI,” but systems designed for real warfare, where latency, reliability, and integration with sensors and weapon systems are essential.

If we look at the overall picture, Palantir's role is different from that of Google:

• Palantir is the analytical and operational backbone in many NATO and US contexts, close to command centers, intelligence flows, and targeting processes.
• Google is the generalist generative engine that enhances mass productivity within the defense machine.

They are two different layers of the same ecosystem, not a replacement of one with the other.

A Dual Ecosystem: Google on the Surface, Palantir in Depth

To truly understand the Google–Palantir duo, it is worth looking at the architecture, not the individual contracts. Broadly speaking, a fairly clear division of roles is being created.

Here is one of the few distinctions worth maintaining in list form, because it clearly clarifies the map:

• GenAI.mil + Google Gemini
  Plays on the horizontal level: productivity, knowledge management, generative assistance across the entire military bureaucracy. It is the AI that lives in intranets, office automation tools, clarification requests, and textual searches, on millions of desktops.
• Palantir (Gotham, Foundry, AIP, NATO platforms)
  Operates on the vertical and operational level: integration of sensitive data sources, models for pattern recognition in the field, military decision support, conflict planning and simulation. It is the AI that lives in operations rooms, command centers, intelligence flows.

The future could see even more integration. It is plausible that GenAI.mil will remain a multi-model platform, where Gemini is just the first piece and where, over time, we might see:

• other commercial generative models,
• conversational interfaces towards systems like Palantir,
• specialized agents that “talk” with external operational platforms.

At the same time, competition between big tech and “defense-native” companies like Palantir will continue on every new tender: who will control the interface, who will control the data, who will control the models.

In this scenario, the title of the film is truly the Google–Palantir duo:
Google conquers the visible surface of military AI, the one that touches every user;
Palantir fortifies the operational depth, the one that decides times, targets, logistics, and scenarios.

There is no longer a single “war AI,” but a two-faced ecosystem: one of glass (Google), one armored (Palantir).