Join War on the Rocks and gain access to content trusted by policymakers, military leaders, and strategic thinkers worldwide.
For decades, the United States’ nuclear weapons enterprise has wrestled with an institutional paradox. On one hand is the imperative to make the complex more “responsive” to global tumult and technical surprise, able to field new weapons quickly in reaction to unanticipated threats. On the other are the unforgiving engineering requirements of thermonuclear warheads, devices whose development is measured in years. No silver bullet solution to this unstoppable-force-meets-immovable-object conundrum will likely present itself. But the contours of a new model in systems development are taking shape that may bring the long-sought responsiveness of the complex far closer to reality.
Under the Genesis Mission, a White House initiative led by the Department of Energy, America’s weapons laboratories are establishing an interconnected web of supercomputers empowered by AI. Using the U.S. government’s unique classified datasets, Genesis is intended to spur a wide range of scientific advances, including in national security. Among its priorities is the U.S. nuclear weapons stockpile, where the platform will be used to improve warhead designs, discover new strategic materials, and accelerate production timelines.
As the administrator of the National Nuclear Security Administration, my foremost responsibility is to accelerate the modernization of the nuclear stockpile as well as the scientific and production infrastructure that undergirds it. I and other leaders of the agency have impressed a sense of urgency, not only to speed planned warhead upgrades but also to devise prompt, precise, and proportional weapons beyond the program of record, all in a timeframe that will influence adversary decision-making.
Naturally, the prospect of tailored nuclear weapons raises timely questions of arms race dynamics. But the specter of a return to the tit-for-tat nuclear buildups of the Cold War is inapt. The United States is not in an arms race with Russia or China, although it is in a capabilities race, at the center of which is advanced science and manufacturing. Possessing the means to quickly field bespoke nuclear weapons disincentivizes America’s rivals from attempting to gain a strategic advantage by introducing novel nuclear capabilities.
Achieving greater responsiveness will hinge to a large extent on the adoption of game-changing technologies. Chief among these is AI. As a proof of concept to demonstrate how AI can develop systems faster and at lower cost, a team of imaginative scientists conceived Aires Tide. The system itself is a flight test vehicle designed to measure the environmental conditions that a weapon would encounter on the path to its target. But its true significance lies in what the design and production method promises for future systems.
Aires Tide combined the generative power of AI with additive manufacturing, or “3D printing,” a process increasingly used to produce components for the nuclear stockpile. As a paradigm for responding to new deterrence requirements, our scientists and engineers are showing that this combination can develop solutions at a fraction of the time and expense of traditional programs.
Any mention of AI in the same breath as nuclear weapons naturally makes people skittish. According to recent reporting, several AI models have shown an unsettling penchant for launching these weapons in simulated war games. In considering AI’s role in nuclear deterrence, it is crucial to distinguish between functions such as designing and building warheads, where the technology can be used virtuously, and decisions surrounding the actual use of these weapons, which are reserved for the President alone. The U.S. nuclear weapons enterprise has deeply internalized the philosophy that AI can support, but never replace, the expertise and accountability of human decision-makers.
Nor can just anyone simply command AI to design a nuclear weapon or delivery vehicle and expect to get anything useful out of it. The program used to design Aires Tide, a frontier commercial AI model, was licensed for government laboratories to operate with the built-in “guardrails” specially configured for national security use cases. But when loaded onto the laboratories’ air-gapped, classified computers, the model supplies outputs that are available only to the U.S. government.
The computers used to design Aires Tide are among the most powerful ever built: Lawrence Livermore National Laboratory’s El Capitan, one of the world’s fastest supercomputers, and Los Alamos National Laboratory’s Venado, an AI-optimized system developed through a joint venture with Hewlett Packard and NVIDIA. An industry team spent the better part of a year at Los Alamos loading its AI model onto Venado to ensure it could successfully be used in a secure environment.
The Aires Tide design process involved laboratory engineers feeding specifications for the vehicle and its payloads into the computer, from which thousands of initial options emerged. Simulations on El Capitan performed a down-select process. The engineers and the AI model then conducted successive iterations, rapidly and continuously refining the vehicle design. Humans made the final selection, validating the design’s faithfulness to the original requirements and also ensuring that the outputs of AI-driven processes meet the exceptional demands of weapon systems. This method offers a glimpse into how the enterprise can harness AI to achieve greater efficiency without relinquishing the critical human hand on the tiller that one would expect in any activity with the word “nuclear” in it.
Using AI and additive manufacturing to develop new systems sidesteps two constraints — sparse engineering talent and high production costs — that have historically bedeviled the manufacture of advanced weapons. Consider the “Tallboy” bomb, developed by British engineer Barnes Wallis during World War II to burrow under and destroy heavily fortified German targets. Tallboy’s aerodynamic shape and gyroscopic spin gave it a supersonic terminal velocity, and a high-tensile steel casing allowed it to penetrate to a depth of 20 meters. But the bomb was labor-intensive to build, requiring precise casting and machining, and the materials were expensive. Moreover, its design depended on Wallis’ singular engineering genius.
Aires Tide had no such requirements. Its principal designer was another machine, and the constituent materials were relatively cheap. Of course, Aires Tide is not a weapon. Yet, as an archetype for quickly designing and building complex systems, the model holds immense possibility.
Here a brief primer on the division of labor between the National Nuclear Security Administration and the Pentagon in fielding nuclear weapons is in order. The former is responsible for the “nuclear explosive package” — the plutonium core, high explosives, and initiators — that drives the weapon’s detonation. The latter develops the delivery means — ballistic and cruise missiles, submarines, and bombers — that carry the warheads to their targets. Both the warheads and their conveyances are fashioned in tandem, each informing the other’s design.
The final step is “flight testing,” where the laboratories and the military collaborate on high-fidelity facsimiles of a weapon, minus the fissile material, which are launched from a missile or aircraft to evaluate their performance in flight. Test vehicles are expensive, on the order of $1–2 million per unit, and the laboratories are at the mercy of a packed schedule of tests conducted by the armed services. Developing a vehicle in-house not only gives the labs more scheduling flexibility but also saves enormous amounts of money, perhaps a tenfold reduction in cost per flight test. Aires Tide was thus both a demonstration of Genesis’ extraordinary generative power and a practical solution to a process bottleneck.
Once the AI model had generated a design for Aires Tide that exceeded flight requirements, printing the vehicle began quickly at the Kansas City National Security Campus’ satellite facility in Albuquerque. In imagining this process, one must put aside the image of desktop 3D printers churning out little knickknacks from spools of filament. The machine that birthed Aires Tide, the Velo3D Sapphire XC, is the same system that SpaceX uses to print Raptor 3 engines. Monstrous in size, the device weighs more than eight tons, stretches nearly 30 feet, and features its own staircase.
The manufacturing process, known as Laser Powder Bed Fusion, begins by depositing a thin layer of powdered Inconel, a nickel-chromium superalloy renowned for its strength and heat resistance, onto the printer’s baseplate. Eight 1-kilowatt lasers simultaneously bombard the powder, congealing it into solid metal, before the next layer is added.
The individual pieces of Aires Tide’s long, cone-shaped fuselage were printed simultaneously, with smaller-circumference bits nesting within progressively larger ones like a Russian doll. Thus, the initial print was a compact mass, a mere 18 inches high, and only reached its full 11 feet when taken apart and assembled. The ability to print the entire body all at once and avoid the laborious, piece-by-piece process of traditional manufacturing is an enormous time-saver, one of the chief advantages of 3D printing. But there are many others.
The conventional approach to making engineered objects involves “subtractive” manufacturing, where a machinist starts with a large mass of forged material and grinds or mills it down to the desired shape. This technique is wasteful, as the shavings are part of the price of the material but cannot be used in the manufactured product. With additive manufacturing, virtually none of the feedstock is wasted, so the “buy-to-fly” ratio — how much raw material you purchase versus how much you actually use — is nearly one-to-one.
Finally, 3D printing mitigates supply chain risks that are intrinsic to conventional techniques. There are swings in the availability of foreign-sourced solid metals, whereas feedstock in powdered form is much more domestically accessible, reducing vulnerability to the loss of external sources.
For all the technical sophistication of Aires Tide, its most remarkable attribute is the time it took to design and build functional prototypes, the central factor in stockpile responsiveness. The engineering team began work only last November, and even with the government shutdown, the first fuselage was printed by mid-January. Furthermore, the program ultimately yielded not one but a small family of test vehicles.
A miniature version dubbed Aires Cub was printed for initial testing in a wind tunnel. In May, a half-scale version named Aires Fox was dropped from 32,000 feet at the Dugway Proving Ground, an Army testing range in Utah. The test evaluated everything from the sensors’ vibration resistance to the performance of the actuators, small fins that maneuver the vehicle during its descent. A trove of data from the flight tests will be fed into the El Capitan supercomputer, generating insights to optimize future systems developed using the same method.
What forms those systems take seems more likely to be limited by the engineers’ imaginations than by any technical constraints. Additive manufacturing is already being used to make an array of components for U.S. nuclear warheads. Printing a vehicle that could accommodate a nuclear explosive package may not be simple, but it is a natural next step. Could Genesis and additive manufacturing help develop proper reentry vehicles, engineered to withstand the extraordinary heat and pressure of reentering the Earth’s atmosphere and hardened against enemy defenses? I see no reason why not.
Of course, eagerness to conceive new systems cannot — and will not — force a choice between accepting underperforming or unsafe weapons and departing from proven approaches to certifying America’s deterrent. To begin with, no imagined system would incorporate a warhead that departs substantially from the historical nuclear explosive testing basis. For the nuclear explosive package and other components and materials that comprise a weapon, the same computer modeling, simulation, experimentation, and testing tools used to assess warhead performance today can readily supply confidence that such new systems are safe and reliable.
Even now, this premise is being demonstrated under a new program to accelerate weapons development, which merges traditional design by human experts and manufacture by casting and machining with advanced design and fabrication techniques. Last summer, the National Nuclear Security Administration chartered a Nuclear Deterrence Rapid Capabilities Team to compress the timeline for developing new weapons from 10 years to within three to five. Under this model, scientists and engineers are identifying concepts with an eye toward rapid production, allowing innovations to transition quickly from prototypes to deployable assets. Yet, these systems would be certified with the same suite of scientific tools we rely on to sustain and modernize legacy warheads today. Some fruits of this program will likely become mainstays of the future stockpile. Others, we hope circumstances never force us to reveal.
The preservation of stability is a fundamental motivator of these efforts. The United States has pointedly criticized Moscow and Beijing for developing exotic nuclear systems, among them Russia’s nuclear-powered Burevestnik cruise missile and China’s fractional orbital bombardment system. Far from pursuing destabilizing weapons like these, however, the U.S. nuclear enterprise is positioning itself to maintain stability, particularly at the regional level, by demonstrating the ability to counter adversaries’ provocative posture shifts.
A state that lacks the technical and industrial wherewithal to answer its enemies’ disruptive moves merely invites its competitors to make them. Aires Tide and initiatives to prototype new weapons signal the United States’ capacity to respond to such actions even if it has no wish for an arms race. Although critics might argue that these investments are ripe for misinterpretation, to be genuinely threatening the programs would have to be accompanied by other signs of an appetite for arms racing that simply do not exist. Rather, our focus is to ensure the superiority of America’s weapons science and manufacturing competencies, and in this we are moving with haste.
Today, acceleration is more than a catchphrase within America’s nuclear weapons complex. It is the lodestar guiding everything from weapon design to materials production. For the nuclear enterprise, the holy grail is the means to field systems to fit niche deterrence requirements almost as soon as they are identified. As American scientists and engineers search for transformational solutions, they now do so armed with cutting-edge manufacturing tools and, even more consequentially, the most formidable computing platforms ever devised.
Brandon M. Williams serves as the Department of Energy’s under secretary for nuclear security and administrator of the National Nuclear Security Administration.
Image: Craig Fritz via Sandia National Laboratories.
