When will artificial intelligence surpass the cognitive capabilities of humans? The exact timeline for achieving this milestone—known as artificial general intelligence (AGI)—remains uncertain, yet recent advances indicate that transformative leaps may be rapidly approaching—clearly with significant defense implications. In a 2022 survey of artificial intelligence experts, 20 percent believed that human-level AI would be developed by 2032. Major leaders in the space are now talking about its creation within two or three years. One of the most evocative descriptions of what this capability could look like is “a nation of geniuses in a data center.” These geniuses would be able to do anything a human can do today with a laptop and internet connection.

If these timelines are at all reasonable, ensuring the US military is AGI ready is of paramount importance. With AGI, the tempo of military innovation could shift from decades to mere weeks, upending the established processes for fielding new capabilities. Tremendous national effort is going toward making sure America has the best AI in the world. Placing too much faith in always having the best AI carries serious risks, though, especially when computing resources and open-source breakthroughs are global and diffuse quickly. A Chinese-built model, Deepseek R1, recently drove home the point that the race for AGI can turn on a dime. Deepseek R1 uses a reasoning technique recently pioneered by OpenAI, yet comes close to matching the performance of OpenAI’s flagship model, o1. While semiconductor export controls heavily handicap the race in America’s favor, rapid algorithmic improvements could change the landscape overnight.

A wiser approach would combine the pursuit of advanced AI with the removal of bottlenecks that might hinder its real-world impact. Defense technologies typically follow a well-worn cycle: battlefield observations spur ideas, fundamental research proves underlying science, engineers design and prototype systems, rigorous testing confirms performance, and then factories scale up production for wide deployment. Today, finely honed human intelligence is the hardest to find input in every step of this process, but it might not be in a world with AGI.

That which is scarce is valuable. Once AGI is achieved and intelligence becomes relatively abundant, physical labs, manufacturing lines, test ranges, and global networks for collecting combat data will form the indispensable bridges between a clever concept and a fielded capability. A fully realized AGI could draft elaborate plans and run simulations at lightning speed, yet it will not be able to assemble physical components or measure real-world conditions by itself. Nations that invest in robust infrastructure, flexible industrial capacity, and reliable feedback mechanisms will be best prepared to capitalize on AGI’s promises, while those that do not may find themselves unable to keep pace when intelligence alone is no longer a meaningful limiter.

Moravec’s Paradox and Physical Production

Moravec’s paradox highlights why physical tasks remain the real bottleneck, even under an AI revolution. While AI can win sophisticated games and write complex code, it struggles maneuvering in the real world. In the defense sector, this could become painfully clear when a brilliant new aircraft or missile design leaps from an AI’s mind onto a screen, but then faces a slow and expensive manufacturing process. Reconfiguring assembly lines, sourcing raw materials, and hiring skilled tradespeople are not trivial endeavors, and general-purpose robots are still clumsy at many of the tasks required to build high-grade military hardware.

If AGI is ubiquitous, the difference between victory and defeat may hinge on who can bring digital insights to life the fastest. Militaries that assume AI alone will solve every problem risk being underprepared for the physical realities of technology development. Investing in flexible manufacturing plants today that are capable of switching rapidly from one design to another is imperative.

Industry can be incentivized to build this capacity by contracting consistent production of significant quantities of short-lifespan weapons and platforms. Building cheaper equipment that does not last as long works better in a world where technology advances quickly. Program offices can reallocate resources from sustainment and maintenance activities toward production, which will result in more up-to-date capability and greater capacity to rapidly grow a force.

Even if current autonomy algorithms for unmanned systems remain imperfect, building out capacity to produce them now ensures that hardware will be ready the moment software catches up. A defense force with a mature pipeline of aerial drones, robotic ground vehicles, and unmanned naval platforms is poised to integrate sudden AGI-driven design leaps far more quickly than one starting from scratch.

America should also emphasize manufacturing research and development efforts as a critical national security priority. Research into advanced robotics, 3D printing, and other cutting-edge production methods can compress the cycle time between concept and fielded system. With AI’s ability to spin out new designs at lightning speed, hardware must keep pace. Otherwise, the most brilliant concept remains stuck in a rendering on someone’s screen, waiting for a factory line that can finally bring it to life.

Fundamental Science and Physical Data

After a new idea emerges in response to battlefield observations, the next immediate hurdle is often fundamental science. Even an AGI that claims near omniscience relies on physical data to undergird and refine its predictions. AI might propose a revolutionary material or propulsion system, but actual labs must synthesize and test it in real-world conditions. Without empirical measurements, the most skillful reasoning models will be untethered from reality, having no basis from which to reason.

Once materials are synthesized and science confirmed, at the other end of the acquisitions cycle the military testing community must take initial prototypes and put them through their paces to confirm they work as intended. Without sufficient testing capacity, the speed at which America can develop capability has an inherent limit.

This need for physical validation makes research labs and testing infrastructure indispensable. AI can accelerate ideas, but only wind tunnels, test ranges, and high-temperature chambers can confirm which ones truly work. Investing in these environments ensures discoveries sparked by AI are refined and validated quickly.

Nations that neglect physical testing capacity risk lagging precisely when AI speeds up conceptual research. Defense planners should invest in expanding physical test beds and research labs as quickly as possible.

Warfighting Context and Real-World Decision-Making

During World War II, the P-51 Mustang’s ultimate success depended on early frontline feedback about initially poor range and altitude performance. Pilots reported real-world deficiencies, and engineers responded by installing the Merlin engine, dramatically altering the air war. In the future, iterative feedback will become even more critical. War is a moving target: enemy tactics change, meaning AI-driven designs must adapt quickly.

Advanced AI is useless if it cannot receive timely updates from the field, because it will be building solutions to the wrong problems. This is why robust connectivity—through low earth orbit constellations like Starlink—should be a top investment priority. Without high-bandwidth network infrastructure, real-time sensor readings and mission logs never reach data centers for analysis. This deficit will reduce even the smartest AI to guesswork.

An AI without modern networking is like a brilliant but blind strategist, stuck refining solutions based on outdated assumptions. By contrast, militaries that equip forces with high-bandwidth satellite communications will be able to feed AGI the fuel it needs to think. If America waits until AGI is here to start proliferating this architecture, there will be a critical period of time where a competitor that does make those investments has a key advantage.

The United States should embark on an effort to connect all its military platforms with low earth orbit constellations that can move data around the planet quickly. It should be possible to take every bit of data collected from a two-hour flight, twelve-hour infantry patrol, or two-week submarine mission, anywhere in the world, and transmit it to a data center in America in seconds. Any part of the force that lacks this capability will not benefit directly from AGI.

If AGI is like a nation of geniuses in a data center, the real question is: Can we build the bridges to bring their ideas to life? The capability development cycle—observations, ideas, fundamental science, engineering design, testing, manufacturing, and field deployment—will not vanish with the advent of AGI. Instead, it will accelerate. AGI will be able to drive concept creation and streamline engineering tasks at an unimaginable pace, but everything else will still require tangible infrastructure and physical effort.

Countries that believe they can rely on AI to solve all their problems overnight will be caught off guard when their conceptual designs remain stuck in digital limbo. Meanwhile, those that master rapid prototyping, industrial retooling, and the systematic gathering of real-world context will be positioned to ride the exponential wave of AI progress rather than be engulfed by it.

In the end, the defense community that keeps pace with AI’s intellectual leaps while remaining deeply anchored in fundamental science and physical production will hold a decisive edge. Nothing about superior AI negates the need for steel, sweat, and data. The cycle endures, and the winners will be those who learn to close the loop faster than anyone else.

While the White House and companies like OpenAI and Anthropic work to build the best AI in the world, the Department of Defense should skate to where the puck is going. The time is now to build the infrastructure that will be the limiting factor in an age of unconstrained and abundant intelligence.

Sean Lavelle is a Navy aerospace engineering duty officer with master’s degrees in finance and machine learning from Johns Hopkins and Georgia Tech. He is the founder of the first all active-duty software development team in the Navy and has built and deployed more than sixty software applications to units across the Navy. He has previously been published in War on the Rocks, USNI Proceedings, the Strategy Bridge, RealClearDefense, RealClearWorld, the National Interest, and the Defense Acquisitions Research Journal.

The views expressed are those of the author and do not reflect the official position of the United States Military Academy, Department of the Army, or Department of Defense.

Image credit: Mass Communication Specialist 2nd Class Evan Diaz, US Navy