In 2017, completing one loop of "find target → strike" took a US command post 743 minutes - nearly 12 and a half hours. In 2026, that number is under 1 minute. Not because missiles got faster, but because every satellite, drone, submarine, stealth aircraft, and 20,000 analysts are all looking at the same picture, through the same software. That software is called Palantir Maven Smart System. This piece is the story of how it was built - and why it, not the F-35, is now America's most powerful weapon.
Scope: This piece dives deep into US military power in the AI era - specifically Palantir Maven and the doctrine of network-centric warfare. We deliberately leave out any comparison with rivals - not because it doesn't matter, but because the picture on the US side alone is complex and fresh enough to fill an entire post.
Sources: Public reporting from the NGA, CDAO, DoD; Wikipedia's Project Maven entry (compiled 2024–2026); IDIQ contract filings on SAM.gov; public statements from Palantir and Anthropic leadership. Figures are order-of-magnitude - defense budgets always have gray zones.
Four Numbers To Start With
These four numbers say one thing: Maven is no longer an experiment. It has become the default operating system for the US military. All 11 of 11 combatant commands - from INDOPACOM (Indo- Pacific) to NORTHCOM (North America) - use it. And it gains new users every day. May 2025: 20,000. March 2026: nearly 25,000. A growth rate no internal Pentagon software program has ever matched.
First - How Is "Old" Warfare Different From "New"?
To understand why Maven matters, you have to understand what it replaces. 20th-century industrial warfare ran on a sequential pipeline - information flowed one way, from sensor, through analysis, up to command, and only then back down as an order. Every link was a bottleneck. Every bottleneck ate up hours.
A U-2 flies over the target, shoots film. The film is developed at base. An analyst pores over it with a magnifying glass, types up a report, sends it up to the command post. A general reads the report, gives the order. The order travels by radio. The pilot takes off, finds the target with the naked eye, drops "dumb" (unguided) bombs. Loop time: days to weeks. Every sensor, every weapon, every service branch - siloed.
A SAR satellite spots a tank at 3 a.m. Within 15 seconds, Maven has fused the satellite imagery + intercepted radio signals + OSINT reporting into a single target package. An analyst reviews it on screen. Coordinates are transmitted over Link 16 to an F-35 400 km away. The pilot presses the button. A JASSM-ER flies 925 km to the target, accurate to within 3 meters. Loop time: minutes.
The standard framework for this loop is called F2T2EA - Find, Fix, Track, Target, Engage, Assess. This is the "recipe" of modern warfare, and AI is compressing every step:
According to a CDAO (Chief Digital & AI Office, Pentagon) report, the traditional F2T2EA cycle for an average low-priority target used to take 743 minutes. With Maven integrating LLMs, that number has dropped to under 60 seconds in real-world operations. What matters here: this isn't a 10–20% optimization - it's a compression of two significant orders of magnitude.
Maven Smart System - The US Military's "AI Brain"
Maven is not an algorithm. Nor is it a "robot." The most accurate description: Maven is an operating system for military decisions. Instead of intelligence officers having to open 9 different systems (one for each data type - satellite imagery, drone video, signals, cables, maps, flight schedules, target lists, and so on), Maven merges everything into a single interface. Click a coordinate on the map - and every relevant piece of data appears instantly: the latest imagery, activity history, linked entities, casualty risk. AI does the correlation automatically.
Technically, Maven is three stacked layers. The software runs on AWS cloud (IL6 tier - the highest security level for classified information), uses Palantir's ontology platform as the "data backbone," and integrates LLMs (Anthropic's Claude was the workhorse in 2024–2025) as the language-reasoning layer:
- Drone video (FMV) - MQ-9 Reaper, MQ-4 Triton, tactical drones
- Satellite imagery - NRO (military) + Maxar, Planet Labs, BlackSky (commercial)
- SAR radar - ground imaging through cloud cover, day and night
- SIGINT - intercepted electromagnetic signals, radio communications
- HUMINT / OSINT - human intelligence reports + open sources, social media
- AIS, ADS-B signals - commercial ships, civilian aircraft
- Ontology layer - turns raw data into a "digital map" of the battlefield; tanks, buildings, and convoys are all objects with relationships to one another
- Computer vision - identifies vehicles, launchers, aircraft in imagery/video
- Pattern detection - flags unusual movement, deployment behavior
- Claude (Anthropic) - reads intelligence reports in English/Russian/Arabic, answers questions, generates course-of-action recommendations
- Target package - complete target dossier, cross-referenced across sources
- Course-of-action - 2–3 options, with risk/benefit tradeoffs
- Collateral Damage Estimate - predicted civilian casualties before a strike
- Common Operating Picture - a shared battlefield picture, identical across every service branch
- Human-in-the-loop - a human must approve before any action
International Humanitarian Law (IHL) and DoD Directive 3000.09 require every lethal use-of-force decision to have human approval. Maven is designed for a decision support role - not autonomous kill authority. In practice, this is tested continuously: when the system generates 5,000 target packages a day, the human "approval" window compresses to just seconds per package. The question isn't "is there a human in the loop" but "is the human still actually reviewing it?"
Timeline - From Failed Project To Military Operating System
Maven was born out of failure. In 2017, the Pentagon awarded its first contract to Google. A year later, 4,000 Google employees signed a letter of protest. Google withdrew. The project looked dead. Palantir stepped in, and five years later became DoD's largest contractor for military AI:
In March 2026, the Pentagon declared Anthropic a "supply chain risk" and demanded it be replaced within 6 months. The stated reason: a dispute over AI safety guardrails - Anthropic refused to strip out Claude's safety filters for certain categories of military commands. This is the first public collision between Silicon Valley AI ethics and DoD operational requirements. The candidates to replace it: an in-house military model, or a "less constrained" LLM.
Palantir's Big Win - Mapping The Contracts
| Contract | Value | Timeframe | Scope |
|---|---|---|---|
| Original Maven IDIQ (NGA) | $480M | 2024–2029 | 5 Combatant Commands |
| Maven ceiling expansion | +$795M → $1.3B | Through 2029 | All 11 Combatant Commands + NATO |
| NGA access expansion | $28M | 2025 | Additional accounts, features |
| Army TITAN + Vantage | ~$10B (10 years) | 2024–2034 | Army ground intelligence system |
| USMC - Marine Corps | Separate (undisclosed) | 2025 | MSS expeditionary variant |
| MSS NATO | Undisclosed | 2025+ | Allied Command Operations |
Add it up: Palantir is the fastest-growing defense software contractor in modern history. For comparison: Lockheed Martin took 25 years and roughly $400 billion to build the F-35; Palantir went from "controversial startup" to "NATO's C2 infrastructure" in under 15 years, with an engineering team that never numbered more than a few thousand people. This is an important power transfer: America's most expensive weapon of the 21st century isn't hardware - it's software and software contracts.
JADC2 - A Network Covering Every Domain Of Battle
Maven is software. But software is only useful when there's a pipe for data to flow through. That pipe is called JADC2 - Joint All-Domain Command & Control. The Pentagon's vision: connect every sensor to every weapon across every domain of battle - land, sea, air, space, cyber. Each service branch fields its own "version" of JADC2:
Integrates ground tactical networks with UK and Australian partners. Demonstrates AI can compress sensor-to-shooter time down to seconds. Tested at Yuma Proving Ground (Arizona).
A network of warships - submarines - naval aircraft sharing data in real time. Already installed on 3 aircraft carriers. A submarine detects a target → a destroyer 300 km away fires a Tomahawk, no direct communication needed.
Advanced Battle Management System - command and control for air combat. Links F-35s, F-22s, B-21s, airborne early-warning aircraft, and ground-based air defense into a unified network.
Proliferated Warfighter Space Architecture: 126 Link 16 satellites already in orbit as of late 2025; target of 1,000+ LEO satellites for data relay, missile warning, and global target tracking. Tranche 3: $3.5 billion.
And the common language that lets everything talk to everything else is Link 16 - the tactical data-link protocol born in the 1990s, now upgraded and pushed onto satellites. Link 16 operates in the 960–1,215 MHz band, frequency-hops to resist jamming, and is hard-encrypted. Any NATO platform - F-35, AEGIS, Patriot, THAAD, dismounted soldiers with special-ops radios - "speaks" Link 16. That's why 126 Link 16 satellites in orbit (2025) is a pivotal milestone: the Link 16 network now has global coverage, no longer dependent on line-of-sight radio between aircraft.
Network-Centric Warfare - A Concrete Example
Theory is one thing. Let's walk through a realistic scenario to see what the network actually does - a classic INDOPACOM training scenario:
A Virginia-class submarine is submerged at 200m in contested waters. Passive sonar picks up the propeller signature of an enemy destroyer 50 nautical miles away. It raises a comms mast for 10 seconds, transmits coordinates via satellite.
Maven receives the coordinates, automatically requests a commercial SAR satellite (Capella or ICEYE) to image the area on its next pass - 7 minutes away. Radar imagery cuts through cloud cover, and AI confirms it's indeed a destroyer, not a commercial vessel.
An F-35B on patrol near Guam receives the coordinates over Link 16. The pilot doesn't need to activate radar (which would reveal position) - the sensor fusion system merges the satellite and submarine data into a single "track" on the display. The F-35 maneuvers into firing position.
The AGM-158C LRASM long-range anti-ship missile is launched. It flies autonomously, no need to stay in contact with the F-35 (which may have already returned to base). In the terminal phase, LRASM classifies the target itself using passive sensors, avoiding civilian vessels. Impact.
Four different services (undersea Navy, Space Force, Air Force, surface Navy), three domains of battle (undersea, space, air), and nobody had to pick up a phone and call anyone. Maven fuses, Link 16 transmits, sensor fusion displays. Humans do exactly two things: detect and approve. Total elapsed time: under 15 minutes. In a scenario without the network, this could easily take a full day - and the ship might already be gone.
Why This Is A Foundational Shift
It's tempting to read these numbers as "a new technology doing the old job faster." But in reality, it's upending three foundational assumptions of modern warfare:
Previously, the military's bottleneck was people: analysts, commanders, pilots. America was strong because it had more well-trained people. With Maven, the bottleneck shifts to data quality and software speed. One analyst with Maven outperforms 100 analysts without it. This shifts strategic advantage from "headcount" to "data volume."
In the 20th century, the F-15 was strong because it was fast and maneuverable. The F-35 is strong because it is a network node. The F-35 carries 5 sensor systems + Link 16; sensor fusion merges it all into one picture; the pilot shares that picture with every F-35, F-22, AEGIS ship, and B-21 within range. The F-35's value isn't the aircraft itself - it's the network it activates. A single F-35 is just a plane; 10 F-35s in the network are a reconnaissance ecosystem.
Within the $886 billion 2026 US defense budget, spending on software and AI is growing fastest. Contracts for Palantir + Anduril + new software contractors ($10B Army, $1.3B Maven, $3.5B PWSA, and more) are forcing generals to ask a hard question: should we cut one aircraft carrier to fund 1,000 more satellites? This is an internal budget fight that could shape the next 20 years.
If Maven proposes 5,000 targets a day, and each analyst can "review" only 500 - humans are no longer really deciding, just rejecting exceptions. This is the biggest ethical question without an answer yet: at what point does "human-in-the-loop" become a ritual? Anthropic getting "shown the door" in March 2026 is the first symptom of this tension.
"In network-centric warfare, the basic unit of combat is no longer the platoon, no longer the squadron - it's sensor plus decision plus weapon. Whichever nation fuses those three fastest, wins."
Bottom Line - 5 Things To Remember
So What About China & Russia - Do They Have This Yet?
This is the obvious next question. Short answer: no other nation has anything comparable to Maven that's been battle-tested. China is investing very heavily but is still at the doctrine and enterprise stage; Russia is even further behind - and the war in Ukraine has exposed that gap mercilessly.
The 智能化战争 (intelligentized warfare) doctrine targets 2027. The JADC2 equivalent is MDPW - Multi-Domain Precision Warfare. The lead integrator is CETC (China Electronics Technology Group) - a state-owned conglomerate that has sold C4ISR systems to 100+ countries. The domestic LLM DeepSeek R1 is reportedly used to generate military scenarios at scale. In April 2024, the PLA dissolved its Strategic Support Force (SSF) and stood up a new Information Support Force reporting directly to the Central Military Commission - a sign the structure is still being sorted out. ISR satellites: ~178 Yaogan plus hundreds of Jilin-1 satellites. Missing: public evidence of an integrated multi-source system running at Maven's kill-chain speed; no large-scale conflict since 1979 to test any of it against.
Russia has ambition: ERA Technopolis (a military-tech city in Anapa, opened 2018) and automated C2 systems like Andromeda-D, Akatsiya-M, and Strelets. The reality in Ukraine (2022–2026): C2 gaps have been exposed publicly - generals forced to use civilian radios, targets losing track after satellite imaging, sensor-to-shooter still measured in hours rather than minutes. Russia is strong in tactical electronic warfare and loitering munitions (Lancet), but at the level of multi-source AI fusion there's no evidence of anything comparable to Maven. Corruption and a rigid command structure - well documented by NATO - are a bigger drag than the technology itself.
The point to remember: the gap between the US and everyone else isn't a hardware gap (satellites, aircraft, weapons - China has closed that distance considerably). The gap is in the software and data ecosystem battle-tested over years: 25,000 real users, 179+ data sources already fused, 9 years of stumbling from Google's withdrawal to MSS NATO. That's not something you can copy quickly - and it's also the thing most at risk if the software supply chain (Palantir, Anthropic, AWS) gets disrupted by internal politics.
03 Discussion
Leave a note
A considered space for questions, counterpoints, and useful additions. Civil, on-topic, signed.
Reader notes
...Loading notes...