The New Military Internet: How AI Firms Became Defense Contractors

The New Military Internet: How AI Firms Became Defense Contractors

By K. Neelu | ICTpost Global AI Intelligence

OpenAI’s NATO deal is the headline. The money is landing one layer down, with Microsoft, Palantir, Anduril and the cybersecurity firms that guard the network.


On November 17, in a conference hall in Izmir, Turkey, NATO opened its annual technology showcase, and the biggest announcement did not come from a missile maker. OpenAI said it was giving the alliance’s core institutions access to its Daybreak cyber-defense models, and disclosed one number that mattered more than the rest of the statement: NATO had been consuming up to 300 billion tokens a month (Quartz).

A token is the unit by which a language model meters its work, roughly three-quarters of a word. Three hundred billion a month is not a pilot. It is a utility bill. The largest military alliance in history has begun buying intelligence the way it buys electricity: continuously, by the unit, from a private vendor.

Here is the conclusion investors should start with, not end with. OpenAI is the name on the release, but in a metered market the best business belongs to whoever owns the meter, the data platform and the locks on the doors. The companies positioned to collect most of the defense-AI economics over the next decade are Microsoft, Palantir, Anduril and the cybersecurity vendors, in roughly that order of certainty. OpenAI may end up the most famous participant and not the most profitable one.

The stack, and who gets paid at each layer

What is being built is not a weapon. It is a network, and the money flows through it in layers.

Layer Company How it gets paid
Model OpenAI Tokens, metered monthly
Cloud Microsoft Compute, plus 27% of OpenAI’s equity
Operating layer Palantir Platform fees under decade-long enterprise agreements
Battle network Anduril Command-and-control software bundled with hardware
Cybersecurity CrowdStrike, Palo Alto Networks Security licenses across every federal network
Hardware Lockheed Martin Interceptors, missiles and the sensors that feed the stack

The demand side is already enormous. The Pentagon’s $1.5 trillion request for fiscal 2027, a 42% increase and the largest year-over-year jump since World War II, carves out $58.5 billion for AI and joint command-and-control, including a $46 billion multi-year “sovereign AI arsenal,” alongside $17.5 billion for the Golden Dome missile shield and $11.3 billion for cyber (The Hill; Greenberg Traurig). The Pentagon’s GenAI.mil portal went from 80,000 users at its December launch to 1.5 million daily users by June (DefenseScoop). NATO allies have pledged 5% of GDP on defense by 2035, with 1.5 points reserved for cyber, resilience and innovation, which is, functionally, a software budget (NATO).

The model layer: fame, and the price of saying no

OpenAI’s path into the Pentagon ran through a rival’s exit. In February, after Anthropic refused terms allowing its models to be used for “any lawful purpose,” the Defense Department designated it a supply-chain risk, a label normally reserved for foreign adversaries. OpenAI announced a classified-network deal within hours; Sam Altman later conceded it had looked “opportunistic and sloppy” and amended it (NPR). By May, eight vendors including Google, Nvidia, Amazon and xAI had signed similar agreements (The Hill).

That is the model layer’s structural weakness. It is crowded, the customer can blacklist a supplier by decree, and models are increasingly interchangeable. The attention goes here. The margin does not.

The cloud: paid twice on every transaction

Every token NATO burns runs on someone’s servers, and for OpenAI that is overwhelmingly Microsoft. Under their restructured partnership, Microsoft holds about 27% of OpenAI, a stake it valued at roughly $135 billion, and OpenAI committed to buy an additional $250 billion of Azure services, with API products Azure-exclusive (Business Standard). Microsoft collects as landlord and as shareholder on the same sale.

It has also spent years building the classified real estate. In January it extended its AI services into the Secret and Top Secret Azure regions (C# Corner). When the Pentagon says it wants models at every classification level, the cloud already living at those levels gets paid whichever model wins.

The operating layer: the only audited defense-AI revenue that matters

Palantir and Anduril are the two companies in this story whose defense revenue is already large and growing at startup speed.

Palantir’s third quarter was the first in which revenue cleared $2 billion; U.S. government revenue was $953 million, up 96% (Palantir 8-K, SEC). Behind it sits a $10 billion Army enterprise agreement and Maven, the targeting-and-intelligence platform the Pentagon is formalizing as a program of record. NATO bought its own version, Maven Smart System NATO, in 2025 (NATO Annual Report).

Anduril got the bigger ceiling: a ten-year Army contract worth up to $20 billion, built around Lattice, its software for fusing sensor data into one battlefield picture (U.S. Army). It doubled revenue in 2025 to $2.2 billion, raised $5 billion at a $61 billion valuation in May, and an October employee tender priced it near $90 billion (CNBC).

Both agreements share a design: pre-negotiated pricing, and the right for any part of the Defense Department to order off the same menu. The Army calls it procurement reform. It is also the creation of two platforms through which a growing share of military software spend must pass. Switching costs in defense software were high. These make them structural.

The guards: selling to both sides of the AI feud

OpenAI’s NATO deal is, at its core, a cybersecurity deal. That is a market CrowdStrike and Palo Alto Networks already own inside government. CrowdStrike ended its fiscal year at $5.25 billion in recurring revenue and launched FedRAMP High-authorized Charlotte AI for Gov in March (CrowdStrike, SEC); its Project QuiltWorks coalition counts both OpenAI and Anthropic as members (CrowdStrike, SEC). The model makers fight; the security vendor sells to both. Palo Alto Networks signed a government-wide GSA agreement in December and now lobbies the defense budget directly (GSA; Legis1).

The primes: not losing, buying in

Lockheed Martin reported a record $292 billion backlog last month, up 27% in one quarter on a $52.5 billion interceptor award (Lockheed Martin via Finviz). Golden Dome needs interceptors; interceptors need an AI command layer. Lockheed sells the former and has bought into the latter, growing its venture fund to $1 billion as defense primes collectively put a record $4.1 billion into startup rounds this year (Konvoy).

What investors get wrong

Ceilings are not revenue: the $20 billion and $10 billion Army figures are maximums over ten years, and the $46 billion AI arsenal has not passed Congress. The base is still small: defense-tech startups receive under 1% of total defense contracts, per Wall Street Journal reporting (Startup Fortune). And the money does not stop at the headline. When NATO pays OpenAI, Microsoft collects twice. When the Army buys Lattice, legacy hardware feeds it. The stack pays out bottom-up as reliably as top-down.

The next five years

Four futures are plausible, and the stack decides which.

Scenario A: OpenAI becomes the Microsoft of defense AI. Daybreak and the classified deployments become standard issue across NATO’s 32 members, and the model layer consolidates around one vendor. Requires models to stop commoditizing. Least likely.

Scenario B: Palantir and Anduril become the new Lockheed. Enterprise agreements harden into programs of record, Maven and Lattice become the Pentagon’s operating systems, and the two companies absorb a prime-sized share of software spend. Most consistent with the contracts already signed.

Scenario C: Microsoft quietly captures most of the economics. Whichever model wins, it runs on Azure at every classification level, and Microsoft books compute and equity gains on the same growth. The base case for a patient investor.

Scenario D: NATO builds sovereign European models. Europe’s defense-tech funding nearly tripled to $7.4 billion this year (Tech.eu), and capitals paying tokens to an American firm under rules an American president can rewrite have every incentive to build their own. The 5% pledge funds it. The wildcard.

For most of the past century, governments bought tanks, ships and missiles from defense contractors. Today they are beginning to buy intelligence itself, measured by the token and delivered through the cloud. The defense contractor of the AI age may not manufacture weapons at all. It may own the model, the data platform, the cyber layer or the servers underneath. The battlefields are changing. So are the companies collecting the checks. editor@ictpost.com

Did you like this? Share it:

Leave a Reply

Your email address will not be published.

77  +    =  83