A growing push to regulate artificial intelligence in military applications is running into a fundamental problem: no one can agree on who exactly represents the “AI industry.” A new analysis by the Stockholm International Peace Research Institute argues that without a clearer understanding of the sprawling ecosystem behind military AI, global governance efforts risk being ineffective—or even skewed by commercial interests.
Calls to involve private companies in shaping rules for military AI have become increasingly prominent. António Guterres has emphasized the importance of including industry alongside governments, scientists, and civil society in discussions about responsible AI use. Initiatives such as the Summit on Responsible AI in the Military Domain reflect this growing consensus. Yet the question of which companies should be involved remains largely unresolved.
Governments cite two main reasons for bringing industry into the conversation. First, companies possess the technical expertise needed to understand rapidly evolving AI systems. Second, because most AI technologies are designed and built by private firms, their decisions during development directly affect whether these systems can comply with legal and ethical standards in warfare.
But as SIPRI highlights, the “AI industry” is not a single, coherent sector. Instead, it is a layered and interconnected ecosystem often described as a “stack.” At the base lies the hardware and infrastructure layer—data centers, cloud computing systems, and physical networks. Tech giants such as Amazon Web Services and Microsoft Azure dominate this level, providing the computational backbone for AI development.
Above that sits the foundational software layer, which includes large-scale AI models and data-processing frameworks. Companies like OpenAI and Anthropic operate here, building systems that can be adapted for a wide range of uses—including military ones. These models are particularly influential because vulnerabilities at this level can cascade into downstream applications, potentially exposing military systems to manipulation or error.
At the top is the application layer, where AI tools are tailored for specific military functions. This includes everything from battlefield decision-support systems to logistics optimization and maintenance planning. It is at this level that the risks and ethical dilemmas of military AI become most visible, as systems directly interact with human operators and, potentially, targets.
One of the report’s key insights is that most military AI applications rely heavily on technologies originally developed for civilian use. Rather than building entirely separate systems, militaries typically adapt components from the commercial AI stack. For example, customized versions of commercial AI models may be deployed in classified environments, blending public-sector needs with private-sector innovation.
The companies involved in this ecosystem fall into four broad categories. Traditional defense contractors—often referred to as “defense primes”—remain central players. Firms like RTX Corporation have decades of experience integrating automation and AI-like systems into weapons and support platforms. Their long-standing relationships with governments give them a strong foothold in military procurement.
Alongside them are newer entrants known as “neoprimes” and defense start-ups. Companies such as Palantir Technologies and Anduril Industries specialize in software-driven solutions, including data analysis and battlefield management systems. Unlike traditional contractors, these firms often develop products independently and then market them to militaries, accelerating innovation but also reshaping procurement dynamics.
A third group consists of major technology corporations—often labeled “big tech.” Companies like Google and Meta operate across multiple layers of the AI stack, from infrastructure to applications. Their vast resources and global reach make them indispensable, but also raise concerns about concentration of power and influence.
Finally, there are the foundation model providers, a relatively new but rapidly growing category. These firms develop large AI systems capable of powering diverse applications. Notably, several have shifted their stance on military use. In 2024, OpenAI reversed earlier restrictions on military applications of its technology, while Anthropic entered into partnerships to supply AI tools for defense purposes. Access to defense funding is seen as a key driver behind this shift, as companies seek to offset the enormous costs of training advanced AI models.
This evolving landscape presents a challenge for policymakers. Engaging “industry” in AI governance is not simply a matter of inviting a few representatives to the table. Different companies influence military AI in fundamentally different ways, depending on their position within the stack. A cloud provider, for example, plays a very different role from a company designing targeting algorithms.
SIPRI argues that more precise criteria are needed to determine which actors should be involved in specific policy discussions. Factors such as a company’s proximity to use-of-force decisions, its influence over system behavior, and its role in connecting different parts of the supply chain could help guide these decisions.
There is also a deeper concern about the motivations of private firms. While their expertise is invaluable, companies are ultimately driven by commercial interests. Participation in regulatory processes can give them an opportunity to shape standards in ways that favor their business models. Without careful oversight, this could distort efforts to establish meaningful safeguards.
The stakes are high. As AI becomes increasingly embedded in military systems, decisions made today about governance and accountability will shape the future of warfare. Ensuring that the right voices are included—and that their influence is balanced—will be critical to building effective and credible frameworks.
SIPRI’s analysis suggests that the next phase of global debate must move beyond broad calls for inclusion and grapple with the complexities of the industry itself. Understanding who builds military AI, how they operate, and what incentives drive them is no longer optional. It is essential to preventing a future where the rules of war are written not just by states, but by the companies that supply their most powerful technologies.
