For years, the U.S. national security establishment has pushed to integrate artificial intelligence (AI) into every element of its operations. But all of that computing power could come at a significant cost in energy and greenhouse gas emissions.
The Pentagon is already among the top users of computing power within the federal government, and researchers estimate intelligence agencies could soon use as much data as tech giants such as Amazon and Meta.
However, the U.S. national security community has the opportunity to bend the emissions growth curve. To do so, the U.S. government must make sustainability in computing a policy priority by funding research, using the power of its purse to encourage responsible change in the private sector, and encouraging a robust public debate about the trade-offs between achieving AI supremacy and curbing global warming.
Most importantly, it should look ahead to the promise that, with fundamental integration into warfare and intelligence gathering, AI could help military planners save fuel, conserve munitions and other supplies, and reduce collateral damage. And by investing in developing technologies that are not yet commercially viable but could be available soon, the Pentagon has an opportunity to fund advances in sustainability that could benefit humanity.
What Drives the U.S. AI Push?
The national security establishment’s all-encompassing adoption of AI is part of the United States’ strategic competition with China and other peer adversaries that have “widely communicated their intentions to field AI for military advantage,” the Pentagon said in its latest AI strategy paper, published in June 2023.
“From the standpoint of deterring and defending against aggression, A.I.-enabled systems can help accelerate the speed of commanders’ decisions and improve the quality and accuracy of those decisions, which can be decisive in deterring a fight and in winning a fight,” Deputy Defense Secretary Kathleen Hicks said in November.
U.S. intelligence agencies are investing heavily in AI and advanced analytics, which officials say will “improve [the intelligence community’s] ability to provide timely and accurate insights into competitor intentions, capabilities, and actions.”
The U.S. military has been integrating AI into its combat systems since at least 2017, and the Pentagon’s 2025 budget request includes $3.2 billion for research in AI and advanced command and control systems. The intelligence community’s $101.6 billion budget request is classified, but its constituent agencies “expect to maintain amounts of data at a scale comparable to that of a large corporation like Meta or Amazon,” the congressionally chartered National Academies of Sciences, Engineering, and Medicine wrote this year.
Areas earmarked by the Pentagon for funding include tactical autopilot; systems to generate complex simulations for weapons development; detailed studies of the ways humans and AI platforms interact and make decisions together; and systems to synthesize and analyze open source intelligence such as news articles and social media content, and make predictions.
Across the Defense Department, agencies and branches are examining the use cases for AI. Among many examples, the Defense Logistics Agency is researching applications of AI across its supply chains and disaster preparedness and response. The U.S. Army is researching AI in propulsion, weapons platforms, autonomous ground vehicles, programs to aid tactical decisionmaking for commanders, and the improvement of vehicle survivability. And the U.S. Navy is studying AI to aid maneuvers, submarine tactics, target recognition in mine warfare, and decisionmaking for F/A-18 crews.
Civilian agencies are also adopting AI. Customs and Border Protection is developing a system to track illicit border crossings in remote locations. U.S. Immigration and Customs Enforcement is working on AI to analyze photos and video extracted from seized mobile devices. And the National Oceanic and Atmospheric Administration is using AI to analyze beluga whale calls.
The Climate Consequences of the Military’s AI Use
Meanwhile, this AI push could put the national security establishment at odds with U.S. President Joe Biden’s administration’s efforts to reduce planet-heating emissions. Defense Secretary Lloyd Austin has described climate changeas “an existential threat to our nation’s security,” and the Office of the Director of National Intelligence has warned of growing long-term risk to U.S. national security from climate change.
Securing, protecting, and processing AI data will take massive computing and networking resources, and the national security establishment is also investing in cluster computing, cloud environments, advanced networking, and quantum-resistant cryptography. Barring great unforeseen leaps in technology, all this will require huge numbers of physical computer servers. Housed in sprawling climate-controlled data centers in the United States and hubs overseas, these computers will consume vast amounts of electricity and thus also fossil fuels. “Data, analytics, and AI capability development requires tremendous computing power and demand will grow exponentially as adoption scales,” the Pentagon wrote in its 2023 AI strategy.
The Department of Defense is already either the government’s largest user of computing capacity or close to it. The Pentagon’s $11.9-billion IT arm, the Defense Information Systems Agency, manages 40,000 servers containing 38 petabytes of storage and serves millions of users. While little is publicly known about operations at spy agencies that eavesdrop on communications and ingest and process huge amounts of satellite imagery and other raw intelligence, in 2011, the National Security Agency and the U.S. Army Corps of Engineers announced they had broken ground on a $1.5-billion, 1-million-square-foot data center in Utah. At 65 MW, at its peak it consumes enough power for a city of about 20,000 people, the Wall Street Journal reported in 2013.
Data centers and data transmission networks together account for 2–3 percent of global electricity use and about 1 percent of energy-related greenhouse gas emissions, according to the International Energy Agency (IEA). These percentages are climbing, with the IEA calling for “strong government and industry efforts on energy efficiency, renewables procurement and [research and development]” to curb emissions growth.
Training AI models is extremely resource intensive, and the demands on the nation’s power grid will only grow as these models become more sophisticated and their use expands. The biggest growth in energy use in recent years has been among the top data users, with combined energy use by Amazon, Meta, Microsoft, and Google more than doubling between 2017 and 2021, according to the IEA.
The Pentagon’s AI will need continuous retraining to process the ceaseless flow of inputs, similar to the AI models underpinning autonomous vehicles,the University of Washington’s Sajjad Moazeni said in an interview in April 2024. Especially for combat applications, the systems will need to process data and yield results as close to instantly as possible, adding to the energy burden.
Data centers under construction in the United States in the second half of 2023 amounted to 5.3 gigawatts, “enough energy to power all the households in the San Francisco metro area for one year,” said real estate firm JLL in a recent report.In a separate interview, Moazeni estimated that ChatGPT, the breakthrough generative AI system, consumed as much energy per day as 33,000 American households. In Virginia, which houses one of the largest clusters of data centers in the world, state electric utility Dominion Energy plans to spend $9 billion on generative and transmission capacity, citing “increases in electricity consumption which are primarily driven by new and larger data center customers.”
How the Pentagon Can Develop AI and Fight Climate Change
The Biden administration’s AI policy, unveiled over the past year, focuses largely on protecting Americans from perceived risks of AI adoption, building a research infrastructure, and laying the foundations of governing policy. The administration has elsewhere set ambitious goals for decarbonizing the U.S. economy, investing in clean energy, and reducing greenhouse gas emissions, and the AI policy directs agencies to “consider the environmental impact” of AI services, including whether the vendor has implemented methods to improve sustainability and energy efficiency.
In furtherance of that, the Pentagon and other U.S. agencies should launch the discussion by sharing data on AI use. A major example of this was the White House’s agency memorandum “Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence,” which called for agencies to publish AI models and use-cases, among other efforts at transparency. However, it specifically excludes national security applications.
In general, the Pentagon promises to prioritize reducing energy demand and investing in renewable energy technology. But apart from acknowledging that emissions will likely grow, its AI policy does not address the huge consumption of fossil fuels that its drive will entail. The following recommendations could help address this oversight.
Increase Transparency
The Pentagon should acknowledge the use of energy as an issue in its AI and cloud computing efforts. It should also state that it will prioritize efficiency as a research and strategic priority, develop metrics for energy use in national security applications of AI, and begin formulating policies to improve on these metrics. It should publish detailed statistics on energy consumption, weighing national security imperatives against the reality that greater transparency will drive innovation and development of further efficiencies.
Congress’s audit arm, the Government Accountability Office (GAO), could also take up the task if directed by leaders in Congress. It has already urged the Pentagon and the military services to issue guidance on comprehensive AI strategy and acquisitions, an inventory of the AI portfolio and budget, development of performance indicators, and more. “As AI technologies continue to advance at an incredible speed, federal oversight needs to evolve alongside them,” the GAO says.
Boost Efficiency
One might argue that the national security establishment’s job is to keep Americans safe, whatever the resource costs. But there is an operational rationale to energy efficiency that goes beyond the need to slow greenhouse gas emissions. In a current example, fuel efficiencies in the Pentagon’s vehicle fleets over the years have reduced the logistics burden; more efficient computing could reduce energy consumption and storage needs in austere installations closer to the fight. And resupply runs can be dangerous: according to one estimate, more than half of the U.S. casualties in Iraq were associated with fuel and water deliveries.
The White House has pushed agencies, including the Pentagon and presumably the intelligence community, to make their data-intensive computing run more sustainably and efficiently. Agencies have closed physical data centers and moved capacity to commercial providers and shared cloud systems, which are flexible and scalable, and incorporate the latest advances. The policy is expected to achieve some savings in costs and energy use.
Already, the Pentagon has declared that “operations must be resilient [and] sustainable,” and it set goals in its 2022 sustainability plan to adopt renewable energy, electric vehicles, “net-zero” emissions buildings and installations, sustainable procurement, and more. To that end, the Pentagon is adding staff at the Defense Logistics Agency’s energy office with expertise in renewable energy procurement.
The Pentagon should expressly add to its sustainability plan the goal of sustainability in computing and begin to encourage, if not require, energy efficiency in AI tech procurement. It could also integrate research on energy efficiency in computing and emissions reduction into its technology research budget, along the lines of programs underway to develop alternative fuels, improve fuel economy in aviation, reuse jet fuel, and more.
It is conceivable that the efficiencies gained by fundamental integration of AI into warfare could offset or outweigh the emissions generated from the associated computing. More accurate targeting of missiles requires fewer munitions trucked or shipped to the fight; more precise deployment of assets across all domains would save on fuel and other supplies; and autonomous drones, smaller than manned aircraft or surface vessels, would burn less fuel.
Incentivize Sustainability Research in the Private Sector
The sheer size of the government’s investment gives it heft in the market that it could use to encourage sustainability in research, development, and deployment. That isn’t just in the AI space. As Dorothy Robyn noted in a paper for Boston University’s Institute for Global Sustainability, Pentagon research has long accelerated breakthroughs in sustainability technology, such as solar panels, microgrids, and energy storage. “[The Department of Defense] not only develops new technology, it procures it—often at a price premium in exchange for higher performance,” she writes. “As a technology matures and improves with use by the military, it becomes cost competitive.” Producers of small modular nuclear reactors hope that the energy demands of AI will boost their business.
Indeed, the Pentagon’s AI strategy relies heavily on collaboration with the private sector and academic laboratories. Silicon Valley has been pouring billions of dollars into AI companies, some of it with express national security applications. Andreessen Horowitz, the $42 billion venture capital firm that backed Airbnb, Facebook, Okta, and many more critical consumer and enterprise-facing tech companies, boasts of its “American Dynamism” portfolio of companies in sectors critical for national priorities, including aerospace and defense. Among those are companies developing autonomous naval craft, battlefield intelligence and surveillance drones, and AI that can anticipate physical security threats.
Looking Ahead
Unforeseen and unforeseeable applications of cloud and mobile computing will demand efficiencies that are not possible with today’s technology.The Defense Information Systems Agency describes data as a strategic asset, noting in the administration’s 2025 budget request the “inherent power in owning data to control the high ground.” Likewise, data’s production, maintenance, and exploitation are critical strategic priorities, and policymakers and war planners have every incentive to make the United States’ national security computing and data infrastructure as efficient as possible. That conversation cannot begin in earnest until policymakers and researchers have more information about the Pentagon’s AI energy costs.
Beyond discovering and exploiting opportunities for climate sustainability, the greatest potential resource savings from AI is seldom, if ever, expressly discussed in U.S. national security policy documents. By improving strategic and tactical planning, encouraging the efficient and proportionate use of lethal military assets, improving intelligence gathering, and doing jobs long done by humans, AI has the potential to save lives and ease war’s toll on its participants.
