GenAI.mil and Three Implications for National Security’s AI Era

Introduction

Roughly six months ago, the Department of War (DoW) launched its internal platform for hosting generative artificial intelligence (AI) models, known as GenAI.mil. The velocity of adoption GenAI.mil catalyzed is unprecedented: It has already scaled to 1.7 million active users across every military service and agency, representing more than 42 percent of the total defense population—a diffusion of innovation curve that would make Geoffrey Moore blush. Within this short window, users have created 100,000 custom AI agents with no-code or low-code development tools and processed upward of 11 million documents. It took only 54 days from the launch order to the platform’s debut—an astonishing sprint, by government or commercial standards. I had a front row seat for the action as director of defense strategy at Google Public Sector, one of the companies playing a key role in the effort.

GenAI.mil portends a significant upheaval for the defense industrial base (DIB). DoW and DIB employees are already reporting that tasks that previously took days or even weeks are being completed in minutes, promising—or threatening—to reshape public and private workforces. Certain analyses or complex processes that may have previously been inconceivable for humans alone to perform are now achievable with the aid of machine intelligence. And perhaps most consequential for this essay series, massive software contracts that might have been awarded to primes and system integrators now may be awarded to software-as-a-service (SaaS) providers or new venture-backed start-ups. Even more surprising, though, is that some of those aspiring disruptors are being disrupted themselves by no-code agents built on the fly using GenAI.mil, raising the question of how the so-called SaaS-pocalypse may or may not affect the DIB. The impact of AI on the acquisition community and industrial base is already noticeable and is likely to grow. Other subtler trends are emerging as industry giants and upstarts alike are exploring AI’s utility for manufacturing in the physical world. In sum, the AI tidal wave has hit the shoreline of the industrial base.

Key Takeaways

The advent of GenAI.mil is a harbinger of some unexpected changes that may reshape the defense industrial base, namely:

• The government may not always be a fast follower; sometimes it may be an early adopter.
• Commercial parity is now attainable and therefore may become expected.
• Commercial infrastructure, not just commercial AI models, may be the next frontier of competition.

The Government Races Ahead

The government may not always be a fast follower; sometimes it may be an early adopter. Conventional post–Cold War wisdom dictates that the government is a slow, unwieldy bureaucracy doomed to trail industry. In this telling, the DoW must content itself with being, at best, a “fast follower” of commercial trends—a notion embedded in the 2022 National Defense Strategy. GenAI.mil upends this narrative. Today, the DoW is actively leading the commercial industry in AI implementation, ahead of the majority of large private enterprises, with usage stats that would be the envy of the top of the Fortune 500. By demonstrating decisive, aggressive leadership, defense officials have set a potent example of how to cut through organizational inertia and suppress institutional naysaying. The DoW isn’t surpassing industry in every dimension of enterprise AI maturity, but by measures of speed and scale, the momentum of the Chief Digital and Artificial Intelligence Office (CDAO), which runs GenAI.mil, is undeniable. Moreover, this bold approach to AI may be transferred to other disciplines and technologies. It seems plausible that the department could achieve a comparable leap ahead in other fields such as digital engineering, modeling and simulation; energy; and autonomy.

The implications of this are significant. For the past five years, venture investors have bet big on the thesis that the DoW would not just dabble in defense start-ups but would start to move multibillion-dollar programs of record into investment categories where digital native companies would have comparative advantages. They lament that this funding shift hasn’t happened yet, at least not at scale. If the trend from fast follower to early adopter continues, it would support the thesis from venture capitalists that venture-scale returns on defense investments are possible. An emboldened DoW acquisition enterprise could accept more risk from newer venture-backed defense companies and may be more open to procurement from a wider set of previously irrelevant commercial technology providers, expanding the definition of “dual use” to goods that previously had not been considered “procurable.” That foreshadows further upheaval in the defense industry.

Commercial Parity Is Now Attainable

For the past decade, commercial parity has largely been an aspirational ideal. The notion is that technology products and services that are simultaneously available for the government to procure are identical or functionally equivalent to what is commercially available, despite the substantial compliance and administrative barriers in the public sector market. President Donald Trump’s administration, sensing correctly that the rapid iteration and intense competition for frontier models was a conducive test bed, set its sights on compressing the latency from commercial release to government use, aiming to reverse the trend of ballooning technical debt and narrow the gap between the AI software that service members can use at work and that which they can access at home.

May 19, 2026, may mark a turning point, with the simultaneous release of Google’s Gemini 3.5 Flash to the global consumer market and the DoW on the same day. This was an unprecedented event. Simultaneous availability for models, features, and tools will soon become a baseline expectation, compared to only a year ago when many vendors planned to charge the government for version upgrades, if they ever planned to provide them at all.

The bar was just raised. This milestone presents both a threat to and an opportunity for conventional defense suppliers and system integrators, prompting them to evolve their offerings and their processes. It also encourages dual-use providers to reconsider forking their code and products between commercial and government customers. If your company isn’t racing toward generative AI as fast as the DoW, you’re already behind.

Commercial Infrastructure, Not Only Commercial Models

Though it is heretical to say in some corners, the department’s future in AI will rely as much or possibly even more on commercial infrastructure, not just commercial AI models. Frontier labs will push for an edge on model performance, but that advantage is likely ephemeral. The enduring contest for this age will be the pipes, not the water. The escalating competition for energy, silicon, critical minerals, network and compute capacity, and even the electromagnetic spectrum shows that though the frontier model competition matters, it’s hardly sufficient for understanding the trajectory of AI adoption. The demand for tokens solely for defense and national security use cases vastly outstrips even optimistic scenarios for future government supply. The only way to scale to meet this demand is a radical idea: Apply the same principles around early adoption and commercial parity that are winning the competition for models and apply them to AI infrastructure.

Historically, the defense establishment has insisted on building custom, physically air-gapped, government-only data centers—or “GovClouds”—isolated from the AI economy. When the center of gravity in geotechnical competition was the network, this made sense from a cybersecurity perspective. When the center of gravity shifted to data, it became unsustainably expensive, ushering in the Joint Warfighting Cloud Capability era. Now that the center of gravity has shifted to compute density and power efficiency, clinging to legacy GovClouds is tantamount to self-sabotage. The next shift in the center of gravity will be to agentic workflows and managing token consumption. To navigate this, the DoW should be guided by two axioms: First, it should take the lowest friction path to the least tokens consumed for the greatest mission outcome. Second, it should pursue an architecture that makes the most efficient use of the nation’s scarcest resources—electricity, chips, data centers, and time—to generate computational power and exchange data. While the premises seem banal, they lead to a startling conclusion: The DoW should move away from the traditional GovCloud paradigm that has reigned since 2011 and pioneer secure ways to use truly commercial infrastructure.

Mirroring its successful approach to AI models, the DoW should prioritize a commercial infrastructure trajectory that utilizes the same supply chains and unit economics powering the free world. Adopting this model yields traditional commercial advantages, such as accelerated deployment speeds and market-driven innovation, while crucially preventing cannibalizing domestic resources. It has the added benefit of being orders of magnitude more resilient than easily targetable GovClouds alone.

Recommendations for Government Agencies and Companies

• Government agencies should replicate the GenAI.mil playbook that is working at the DoW. Moreover, large industries on par with the department—and really any non–digital native institution—would also benefit from this approach or a version of it. Here are the key tenets: First, organizations should create a central, shared portal for model access, data curation, tooling, and agent orchestration. Second, they should provision the platform with scalable commercial cloud resources, and wrap it in a forward-leaning but prudent security layer. Third, they should enrich the platform by gradually expanding the data stores it accesses. Fourth, the department leader or chief executive officer should encourage use of AI by everyone—as these iconic posters illustrate—and make training widely available at scale. Finally, organizations need to continually experiment, adding new features, connectors, and model upgrades each month to keep up with the rapid pace of industry innovation.
• Industry executives should stop assuming that the department’s technical debt will insulate the DIB from the accelerating pace of the commercial world. DoW leaders are becoming more savvy consumers of digital technologies, deliberately pushing public sector industries to behave more like their commercial counterparts. In homage to William Gibson, who said, “The future is here, it’s just not evenly distributed,” commercial parity is already reshaping government procurement patterns, it’s just not evenly distributed, yet. Industry leaders and their government customers alike must now aggressively prepare for a convergence in which the defense industry is expected to keep pace with commercial companies. The new vision for the Defense Innovation Unit is exhibit A.
• Government and industry leaders should pursue AI pathways that offer the most commercial scalability and interoperability, even when it means sacrificing long-held policies or practices. Catalyzed by the shifting industry conditions, leaders need to liberate their organizations from legacy GovClouds, archaic policies, and confining, vendor-locked ontologies or proprietary data schemas that inhibit interoperability. The most elegant and efficient future will be multimodel, multicloud, and multiagent.

Conclusion

The DoW’s swift, decisive, and widespread embrace of AI is impressive—so much so that it challenges conventional dogma about the department’s acquisition practices, and even its culture. The downside is that the DoW is increasingly competing against the domestic economy for the same precious AI resources: data center space, power, chips, etc. Integrating the military into the shared, market-driven AI ecosystem is the only viable scenario for maintaining a permanent technical edge and an affordable AI investment. As the government races to keep pace with the commercial frontier, the traditional DIB will have to run harder to keep up. While these shifts are threatening to some and welcomed by others, it is an opportunity for all to seize a moment defined by the government’s aggressive embrace of frontier models and agents, the diminution of technical debt and the acceleration toward commercial parity, and the application of commercial AI infrastructure to mission in previously unimaginable ways. This paradigm represents more than just new trends in information technology—if it takes hold and is sustained, it represents a systemic shift that improves America’s footing in great power competition. This far tighter coupling between tech industry and defense is America’s democratic, free-market response to the doctrine of military-civil fusion and provides a mechanism for delivering the most capable commercial AI capabilities to the warfighter.

About the Author

Joshua Marcuse is the director of defense strategy at Google Public Sector and the former executive director of the Defense Innovation Board.

About the New American Defense Industrial Base Series

This essay series, The New American Defense Industrial Base, features expert practitioners with experience in government, industry, and finance writing on the most pressing challenges in defense acquisition today. For more in this series, click here. The defense industrial base series is made possible by general support to the Center for a New American Security (CNAS) Defense Program and corporate support for the series.

About the Center for a New American Security

As a research and policy institution committed to the highest standards of organizational, intellectual, and personal integrity, CNAS maintains strict intellectual independence and sole editorial direction and control over its ideas, projects, publications, events, and other research activities. CNAS does not take institutional positions on policy issues, and the content of CNAS publications reflects the views of their authors alone. In keeping with its mission and values, CNAS does not engage in lobbying activity and complies fully with all applicable federal, state, and local laws. CNAS will not engage in any representational activities or advocacy on behalf of any entities or interests and, to the extent that the Center accepts funding from non-U.S. sources, its activities will be limited to bona fide scholastic, academic, and research-related activities, consistent with applicable federal law. The Center publicly acknowledges on its website annually all donors who contribute.