The U.S. Army and a Second Manhattan Project for AI

In the near future, the U.S. Army can once again become the nation’s prime integrator of science and defense innovation—just as it did in World War II with the original Manhattan Project. While the scientists provided the breakthroughs, it was the Army—through the Corps of Engineers, military police, signal corps, logistics, and counterintelligence units—that built the secret cities, secured the facilities, and set the conditions for those breakthroughs to flourish. In short, the Army played a key enabling role that created and secured the space in which scientific talent could build the atomic bomb. Land power is more than just controlling terrain—it is about setting conditions for follow-on actions, from scientific breakthroughs to major joint offensive campaigns, that change the world.

What Would a Second Manhattan Project Look Like?

Today, a growing chorus of voices in government and industry is calling for a “second Manhattan Project”—this time focused on ensuring U.S. leadership in artificial intelligence. In November 2024, the U.S.-China Economic and Security Review Commission recommended a Manhattan Project–style initiative to win the long-term technological competition with China, emphasizing a centralized, state-led effort that coordinates private sector initiatives. The idea is to mobilize public and private talent around AI infrastructure, just as the original Manhattan Project unified the atomic effort. These calls were echoed by Department of Energy (DOE) Secretary Chris Wright in July 2025 to include linking the need for a nuclear renaissance to power the initiative.

Meanwhile, industry is sprinting. xAI, for example, is building out gigawatt-scale power and massive datacenters around Memphis to power training clusters measured in the hundreds of thousands of graphics processing units (GPUs)—an emblem of how control of compute and energy is becoming a strategic terrain in its own right. Multiple agencies—along with private sector bodies and companies—see the need for a concerted effort that links the science of AI to infrastructure investments and multiyear contracts for leading AI companies backed by the U.S. government. Both proponents and critics see the race for AI leadership as the defining battle of the twenty-first century.

There are also innovative calls to use private equity–style vehicles to finance modernizing Army capabilities and infrastructure. During the recent Association of the United States Army Annual Meeting and Exposition, the Secretary of the Army announced Project FUZE, a new initiative to accelerate Army modernization and search beyond traditional defense acquisition routes. Additional modernization proposals include transforming bases to support public-private datacenters and their supporting energy infrastructure—required to power an AI future—and critical mineral refineries. The deals would turn otherwise idle land into productive capital while building the networks required to generate algorithmic combat power.

The Army and the First Manhattan Project

Under General Leslie Groves, the Manhattan Engineer District administered the design, construction, and operation of the projects infrastructure—effectively standing up entire cities at Oak Ridge, Hanford, and Los Alamos in under two years. The Corps of Engineers synchronized land acquisition, industrial contracting, power and water, and compartmented facilities under tight secrecy. These efforts were not limited to construction. They also included concerted efforts to build new networks connecting the sites including communications relays and roads that supported an army of scientists.

These activities—often called combat support and combat service support—illustrate how to think about land power. More than pitched infantry battles, land power is the process of changing the environment and setting conditions for moving forces and generating effects. In the case of the Manhattan Project, the “effect” was building the infrastructure and support system required for thousands of scientists to focus on solving basic and applied research programs.

The paramount need for secrecy surrounded these efforts. The Army G-2 assigned Major John Lansdale Jr. as the project’s counterintelligence lead. Landsdale built an extensive array of overt physical security and quasi-clandestine countersurveillance networks that monitored each site and cleared workers. These efforts were run out of the Protective Security Section that coordinated with the FBI and focused on personnel, plant, and information security. These activities illustrate the often overlooked role of defensive intelligence operations. The Army doesn’t just secure key terrain physically, it denies further infiltration from a mix of human and technical intelligence.

The Army and a Second Manhattan Project

Just like in the original Manhattan project, there is a massive infrastructure component to the AI race. Unlike the secluded labs of the World War II, the current technology race relies on known math and model architectures, with innovation in refinement (i.e., data, weights, optimization), access to chips, scalability, and fit-for-purpose model applications. Even if a new approach to generative AI emerges, these characteristics are unlikely to change and require a larger network of firms more than they do secluded cloisters of scientists.

The terrain is different. As a result, securing that terrain and setting conditions for ushering in rapid increases in AI performance look different.

First, there will be a need for building and protecting critical infrastructure. While much of this will be commercial, the Army can support these effort through accelerating contracting and even mobilizing Army Reserve and National Guard units for short surges. This surge capability could be used to augment the construction of datacenters and energy projects that support national AI capacity. In fact, some of the sites—including Oak Ridge—that DOE sees as important to new AI datacenters were key hubs in the original Manhattan Project.

Second, this effort would also have to focus on site security, but with a twist. Unlike Major Lansdale’s Protective Security Service, the Army would need to link together new teams that combine counterintelligence agents and analysts, signals intelligence analysts, information warfare specialists, data scientists (i.e., operations researchers), and cyber protection teams. These teams would provide the type of cross-functional expertise required to help industry understand how modern adversaries seek to infiltrate and exploit critical infrastructure. Additionally, the cross-functional teams would maintain the precarious balance between upholding operations security while managing adversary perceptions to avoid accelerating an AI arms race. That would require working through entities like Cybersecurity and Infrastructure Security Agency (CISA) in the Department of Homeland Security (DHS) alongside the FBI and key agencies like the Departments of Commerce and Energy as well as the Intelligence community while preserving key protections granted to U.S. persons under Executive Order 12333 and Department of Defense Directive 5240.01.

Third, the Army has an opportunity to accelerate commercial adoption by both supporting the construction and security of infrastructure, and training the next generation workforce. There is a deep history of the Army building critical transportation infrastructure and later having soldiers lead businesses linked to it, expanding economic opportunity. The Army Corps of Engineers took a leading role designing and building major early U.S. transportation networks including the Chesapeake Bay, Delaware Canal, and C&O Canal, and later critical infrastructure that changed global trade, including the Panama Canal. For much of the nineteenth century, West Point graduates were critical to the development of U.S railroads. The Army can revive this history by accelerating AI training and certification in its ranks while also building the critical skills required to build datacenters and energy plants. The AI economy will need people who can adapt AI models to reimagine work as well as operate heavy equipment, weld, and wire buildings.

In parallel, the Army should become the first joint force organization to field AI at scale, visibly. This effort should include institutionalizing agentic workflows in fires, sustainment, protection, and even military planning. This effort will require the Army to invest in benchmarking and uplift modeling. Benchmarking involves identifying general model tendency and bias to support refining them for highly contextual activities like warfighting. Uplift modeling compares how users complete tasks with AI, helping identify better optimization. Both are required to achieve better performance. The Army cannot just buy its way into the AI future without making significant investments in improving a legacy model of training and professional military education.

Conclusion: Start Now

The time to change is now, and it must involve a mix of innovation at the edge alongside large investments. It must build on existing efforts, like a recent AI tabletop exercise with industry leaders attended by one of the authors. A Second Manhattan Project for AI will not hinge on a single breakthrough or a solitary lab. It will be decided by whether the United States can build and secure the enabling conditions—compute, energy, networks, data integrity, and a trained workforce—and translate them into repeatable operational advantage. That is a quintessentially Army problem to solve: Just as the service once turned dispersed scientific effort into a disciplined enterprise, it can now integrate a diverse AI ecosystem into doctrine, training, and combat power. Nationally, that means helping accelerate the build-out of datacenters and resilient energy, expanding Guard and Reserve engineering and cyber protection capacity, and partnering with the DOE, DHS and CISA, FBI, and industry to defend the AI supply chain. Internally, it means fielding agentic workflows at echelon, institutionalizing benchmarking and uplift modeling, weaving AI literacy through professional military education and unit training, and also growing innovation centers on installations to help soldiers upskill in AI so gains are measurable, repeatable, and accountable.

If the last Manhattan Project proved that land power can create the physical and security conditions for science to change the war, the next one must show that land power can create the computational and organizational conditions for AI to change the force—ethically, securely, and at speed. Do that, and the Army won’t just keep pace with the AI age; it will set the standard for how a democracy mobilizes technology into credible deterrence and decisive advantage.

Major General Jake S. Kwon is the director of strategic operations in the Headquarters, Department of the Army G-3/5/7. Benjamin Jensen is the director of the Futures Lab and a senior fellow for the Defense and Security Department at the Center for Strategic and International Studies (CSIS) in Washington, D.C.