#The Pentagon Just Signed AI Deals With 8 Big Tech Companies. Here's the Full List — and Who Said No

The short version

The U.S. Department of Defense has formalized AI capability contracts with a group of major technology companies, a move that represents one of the more significant expansions of the military-tech relationship in recent memory. The companies that signed include some of the most recognizable names in American tech. The ones who declined, or who are conspicuously absent from the list, are equally revealing. This is not simply a procurement story. It is a window into how the tech industry is navigating the hardest version of the question that has been building for years: what are you willing to build, and for whom?

#Why this matters right now

The relationship between Silicon Valley and the Pentagon has never been simple, and the past decade has made it considerably more complicated. Google's internal revolt over Project Maven in 2018, when thousands of employees signed a letter protesting the company's involvement in a drone targeting AI contract, set a visible precedent. Google eventually did not renew that contract. The episode became a reference point for every subsequent conversation about tech companies and defense work.

What has shifted since then is the broader context. The war in Ukraine made AI-assisted military capability feel less abstract to people who had previously thought of it in theoretical terms. Drone warfare, satellite intelligence, battlefield communication systems, logistics optimization: these are now live, documented applications of technology that looks a lot like what tech companies build commercially. The gap between "AI for enterprise productivity" and "AI for military operations" has narrowed in ways that are hard to unsee.

At the same time, the political environment in the U.S. has made defense contracts more attractive to tech companies than they were five or six years ago. The current administration has been vocal about expecting American companies to contribute to national security priorities, and the implicit pressure to cooperate has real weight when you are a company that depends on favorable regulatory treatment, government cloud procurement, and access to federal data infrastructure.

So the announcement of these contracts landed in a moment when the industry's positions on defense work are more varied and more honestly stated than they were in the Maven era.

#Who signed and what they are building

The eight companies named in the Pentagon's contracting announcement span a range of AI capability types, and the work they are being contracted for reflects the DoD's current priorities: logistics and supply chain optimization, intelligence analysis and data processing, cybersecurity and threat detection, and decision support systems for operational planning.

Microsoft is the least surprising name on the list. The company has had deep federal and defense contracts for years, and its acquisition of Activision aside, its identity as an enterprise and government infrastructure provider has never been in serious tension with defense work. Azure is already the backbone of classified cloud systems through its government cloud offerings. Adding AI capability contracts is an extension of existing relationships, not a new strategic direction.

Google is back, which is the more notable data point. After the Maven episode and the employee backlash that followed, Google rebuilt its approach to defense work with a set of AI principles that were meant to define what it would and would not do. The company has since taken on defense-adjacent contracts, and its presence on this list suggests the internal calculus has shifted. Whether that reflects changed employee sentiment, changed leadership priorities, or simply the passage of time and personnel turnover is not entirely clear from the outside.

Amazon, through AWS, is another expected name. AWS GovCloud and the classified regions Amazon operates have made it a standard infrastructure partner for federal agencies for years. The AI capability work being contracted here likely involves bedrock model access and inference infrastructure more than anything that requires Amazon to build new defense-specific systems from scratch.

Palantir is on the list and has been oriented toward government and defense work since its founding. This is not a tension point for Palantir; it is the core business. The company built its identity around working with intelligence agencies and the military, and its presence in a DoD AI contract is about as surprising as a defense contractor winning a defense contract.

Anthropic and OpenAI are both named, which is the part of the list that will generate the most discussion. Both companies have framed themselves, to varying degrees, around the responsible and beneficial development of AI. Both have published safety commitments and principles. Taking DoD contracts does not necessarily violate those principles, depending on what exactly the contracts involve, but it complicates the narrative each company has built about why it exists and what it is optimizing for. The specific scope of their involvement matters a great deal, and that detail is not fully public.

Oracle and a defense-oriented AI startup round out the list. Oracle's presence is consistent with its aggressive push into government cloud infrastructure. The startup's presence suggests the DoD is deliberately trying to build relationships with emerging AI companies, not just consolidating with established players.

#Who said no, and why it matters

The notable absences are where the story gets more interesting.

Apple is not on the list. Apple has never been a significant defense contractor and has not shown interest in becoming one. Its business model does not depend on government procurement in the way that Microsoft's or Amazon's does, and its brand is built heavily on consumer trust and privacy positioning. Taking visible defense AI contracts would create real tension with that identity. The absence feels like a deliberate choice rather than an oversight.

Meta is also absent. Mark Zuckerberg has made a point of positioning Meta as a more defense-friendly company than it was during the content moderation controversies of the early 2020s, but Meta's core business is advertising and consumer social platforms. The Pentagon does not need Instagram's recommendation algorithms, and the kinds of AI capability the DoD is buying do not map cleanly to what Meta actually builds.

The more pointed absence is among smaller AI companies and research labs that have explicitly declined or remained silent. Some of the most technically capable AI teams in the U.S. are at organizations that have, for now, stayed out of overt defense contracting. Whether that position holds as DoD money and pressure increase is one of the more consequential open questions in the industry.

The employee dimension is also real in a way that corporate statements do not always capture. Several of the companies that signed these contracts have employees who will disagree, sometimes publicly. The Maven-era precedent showed that internal dissent can actually move corporate decisions. Whether the current generation of AI researchers and engineers at these companies will mount similar pressure campaigns, or whether the political and cultural moment has shifted enough that it will not happen, is genuinely uncertain.

#The harder questions underneath the contract list

Defense AI contracts are not all the same thing, and the tendency to discuss them as if they were obscures the real ethical distinctions worth making.

There is a meaningful difference between AI that optimizes military logistics and AI that assists in targeting decisions. There is a difference between cybersecurity systems that defend networks and offensive systems designed to compromise adversaries'. There is a difference between intelligence analysis tools that help humans understand large datasets and autonomous systems that make or recommend lethal decisions with minimal human oversight.

The DoD's stated position is that its AI use will maintain meaningful human control over consequential decisions, particularly in lethal applications. That commitment, written into policy, is worth something. It is not a blank check, and it requires ongoing enforcement and transparency to mean anything in practice. The history of military procurement suggests that stated constraints do not always survive contact with operational pressure.

What the companies that signed have agreed to build, specifically, would tell you a great deal about whether the concerns raised by critics are proportionate or overblown. The fact that much of that detail is classified or not publicly disclosed makes it difficult to evaluate from outside. Companies that want credit for signing responsibly need to be willing to describe, at some level of specificity, what responsible looks like in practice.

#What this means for you

If you work at one of these companies, particularly in an AI-focused role, this announcement is relevant to your professional life in a concrete way. It is worth understanding what your company has actually committed to, what the work involves, and whether that sits inside your own sense of what you are willing to contribute to. The Maven precedent showed that individual researchers and engineers are not powerless, but it also showed that collective action requires organization and that corporate responses to that pressure are not always what employees hope for.

If you follow AI development as a policy or technology story, the more important question than "who signed" is what oversight mechanisms exist for how this AI gets deployed. The contracts themselves are the beginning of a process, not a conclusion. The DoD's internal AI governance frameworks, the role of Congress in oversight, and the degree to which contracted companies retain any visibility into how their systems are used all matter more than the signing ceremony.

If you are an investor in any of the companies on this list, the defense AI market is genuinely large and growing, and early relationships tend to compound into larger ones over time. The risk side is reputational and involves talent: if signing these contracts makes it materially harder to recruit the best AI researchers, that is a real cost that does not show up immediately in the numbers.

#A few questions worth asking

Does taking a DoD AI contract mean a company is building weapons?

Not necessarily, and this distinction matters. Most of what the Pentagon needs AI for is not weapons: it is data analysis, logistics, cybersecurity, predictive maintenance, administrative automation. The same AI that optimizes supply chains for a retailer can optimize them for a military base. The more sensitive applications, targeting assistance, autonomous systems, offensive cyber, are a subset of defense AI work and presumably involve more scrutiny and different contractual structures. The conflation of all defense AI with weapons development is not accurate, though it is understandable given how little public detail exists.

What happened to the employee activism that killed Project Maven?

It has not disappeared, but it has become less organizationally coherent. Google's subsequent decisions around defense work happened with less visible internal revolt than Maven, which could mean the company handled the communication better, or that the employee base has changed, or that the broader political climate has made public dissent feel less safe. Probably some combination. The researchers most likely to lead organized opposition at any given company are also the ones with the most employment options, which creates its own dynamic.

Is there a meaningful difference between Anthropic or OpenAI doing this work versus a traditional defense contractor?

Yes, and the difference runs in both directions. Companies that have built identities around AI safety and beneficial development carry a different kind of credibility when they say a system is safe to deploy. That credibility is an asset to the DoD if genuine. It is also a thing that can be diluted if the companies take contracts that are hard to reconcile with their stated principles. A traditional defense contractor does not have that credibility to protect or lose. Whether the DoD values the safety credibility or just the technical capability is a question worth asking.

How do allied countries think about this?

The Pentagon contracting with U.S.-headquartered tech companies on AI capability has implications for how allied militaries access similar tools, and for how adversaries assess U.S. AI military capability. NATO allies are watching this closely, and several are developing their own frameworks for defense AI acquisition that will need to interoperate with U.S. systems. The global dimension of military AI development is significantly more complicated than any single contracting announcement suggests.

Will any of these companies eventually pull back the way Google did with Maven?

Some of them might, under the right conditions. The conditions that produced the Maven reversal involved a specific combination of public visibility, organized internal dissent, and a contract scope that was easy to characterize as directly enabling lethal targeting. If any of the current contracts produce a similar combination, the pressure dynamics could repeat. The companies that signed this time are probably better prepared for that scenario than Google was in 2018, which means more careful internal communication and more prepared public responses, not necessarily different decisions.