The Arms Race is Measured in Gigawatts

The Arms Race is Measured in Gigawatts

Why America’s military edge runs through a civilian grid.

Katherine Lambert

Oct 22, 2026

20

1

Share

Katherine Lambert is a University of Toronto student and AI researcher focused on critical systems. She is a former Palantir FDE intern.

The Dawn of a New Era

Before sunrise on March 1, 2026, Iranian Shahed drones

struck

two Amazon Web Services (AWS) data centers in the United Arab Emirates (UAE). Debris from a third strike damaged an AWS facility in Bahrain. By morning,

two of three

UAE availability zones were offline. This was enough to rupture the central architectural premise of cloud infrastructure, built on the assumption that zones fail one at a time.

The cascade was immediate. Leading financial institutions, such as the Abu Dhabi Commercial Bank and Emirates NBD,

reported

digital-service disruptions. The AI data platform company, Snowflake, experienced

service errors

, and the ride-hailing platform Careem

declined

rides to customers who had no idea that the failure came from a drone attack. The Islamic Revolutionary Guard Corps (IRGC)

claimed

responsibility and stated that AWS was hosting U.S. military AI, making it a legitimate target of war. Weeks later, Iran published a list of technology companies, including Amazon, Meta, Microsoft, Google, Nvidia, and Palantir, all of which it classified as valid military targets.

The IRGC strike was the first deliberate kinetic strike by a state on commercial cloud infrastructure in history.

The IRGC almost certainly didn’t know whether U.S. military workloads were actually running on those specific servers. Nor did the engineers who built the servers imagine that it would matter if they were used for military operations. Amazon selected the UAE for its cheap land, fiber connectivity, and reliable power. Those features would harden the facilities against the threats most engineers worry about, like periodic outages, software failures, and cyberattacks. Instead, the data centers were hit by drones. With the IRGC strikes, the United States distributed some war risk onto companies unequipped to carry it.

Every dominant era of military power has been governed by an enabling layer: the infrastructure without which all else fails, which defeats strategy before movement can even begin, and which no general can conjure up by issuing an order. Coal and steam in the nineteenth century. Oil from 1912 onward. And now, electricity.

The shift to AI as the dominant technology of warfare has already occurred. The Pentagon’s

$9 billion

Joint Warfighting Cloud Capability contract puts AWS, Microsoft Azure, Google Cloud, and Oracle at the center of how U.S.forces plan, target, and fight.

Maven Smart System

, a military AI operating system, is deeply integrated throughout the military. The Defense Logistics Agency, managing end-to-end supply for the military services, awarded Google a

$48 million

contract in 2025 to modernize its supply chain operations on an AI-ready cloud platform.

A growing share of critical military capability relies on data centers, which run on electricity, which in turn comes from a civilian grid. And that civilian grid is largely in private hands, with investor-owned utilities accounting for

80 percent

of the U.S. transmission network. Strip away the software and AI and what remains is an unbroken line of dependency terminating in electricity. No power means no compute and no military AI, which translates to deficiency in modern warfighting capability.

A nation’s ability to project force, sustain operations, and deter adversaries has always been a function of its industrial base. And in the coming decades, the ability to assert sovereignty for any nation may be determined less by what it can build or buy, than by what it can reliably power.

The new arms race is measured in gigawatts.

The Constraint

U.S. data center power demand was

projected

by Goldman Sachs to more than double, from 31 GW in early 2025 to 66 GW by 2027. However, S&P Global

reported

that U.S. data-center demand had already reached 64.4 GW in late 2025, exceeding the earlier projected forecast. It now puts the 2030 number at

183.2 GW

. Data centers will draw

nearly 12 percent

of all U.S. electricity, up from four percent just a few years ago, and from the same grid that powers hospitals, factories, and water treatment plants.

In parallel, more than

2.2 terawatts

of generation and storage projects are sitting in interconnection queues, which is the formal regulatory process that new electricity projects must follow to connect to the grid. Today, the entire installed generating capacity of the United States is only around

1.2 terawatts.

Behind the demand figures sits a different problem entirely. The United States is not running short of electrons, but of time and institutional capacity to move electrons from decision to delivery.

The median wait time for a new power source to connect to the grid now exceeds

five years

. In the largest grid operator in the country, PJM, it can take up to

eight

. Further, around

75 percent

of the capacity that entered interconnection queues from 2000 to 2020 had been withdrawn by the end of 2025, while only

13 percent

was operating.

The power problem isn’t just quantity and wait time. Military AI workloads need a dispatchable, 24/7 base load system that works regardless of weather and can survive grid disruption from enemy strikes. The civilian grid strategy of adding solar and wind at scale to route around the problem works for average household demand, but it cannot guarantee 500 MW at 3 AM in December. This is why nuclear is back, why Microsoft

signed

a 20-year power purchase deal to restart the Three Mile Island nuclear plant, and why gas peakers are being signed at premium rates by companies whose press releases still say

100 percent

renewable.

Energy capacity has decided wars before

In 1912, Winston Churchill faced a choice that looked like a bad trade:

convert

the Royal Navy from coal to oil at the peak of an arms race with Germany. Britain had abundant coal resources—the South Wales coalfields were among the richest in the world—and no domestic oil. Exchanging a secure state resource for a foreign dependency mid-competition was strategically illogical.

Churchill converted regardless, commissioning

Admiral Fisher

to find the oil and prove that it could be stored cheaply in peacetime and secured with certainty in wartime. From this came the Anglo-Persian Oil Company, in which the British government took a

51 percent

controlling stake. Churchill then introduced the investment legislation in

June 1914.

Oil-powered ships ran faster and quieter than coal-powered ones, could refuel at sea, and needed far fewer crew members to operate. In a naval arms race measured in knots and range, that was decisive.

Forty years later,

Hyman Rickover

applied the same logic to submarine warfare, routing backward from required capability to the necessary resource to fuel it. Diesel-electric submarines were constrained by battery life and the need to surface. His approach was to eliminate the constraint at its source. USS

Nautilus

went to sea in

1955

powered by a reactor, and submarine endurance became effectively unlimited.

The German case study illustrates the failure mode.

Germany’s defeat in World War II traces back beyond military strategy. Beneath the operational failures lay an economic model that could not sustain the industrial demands of a prolonged war, with fuel being the sharpest expression of this. Lacking sufficient petroleum resources to power a modern military campaign, Germany built synthetic fuel plants that converted domestic coal into liquid fuels—a remarkable feat of industrial chemistry, papering over a fundamental scarcity. The Allies understood this exact constraint, and when the RAF and USAAF struck the Leuna and Pölitz plants in 1944, German aviation gasoline output from synthetic plants fell from 175,000 tons in April to just

5,000

in September.

The technology had been sophisticated, but it lacked a secure enabling layer underneath. It was capital-intensive, slow to repair, and geographically vulnerable. Once the Allies recognized that, the air war was effectively over.

The Institutional Gap

Churchill’s successful coal-to-oil transition had a structural advantage that is easy to overlook. The Royal Navy was the customer, the operator, and the policy-setter all housed within a single institution. When the decision was made to convert, Churchill could commission a study, take a controlling stake in an oil company, sign a twenty-year supply contract, and issue an order. The enabling layer was all within a single chain of command.

Meanwhile, the electricity infrastructure of the United States answers to no one in uniform.

Fractured across private utilities regulated by the Federal Energy Regulatory Commission (FERC), fifty state public utility commissions, regional transmission organizations, private sector-owned infrastructure, and dozens of local permitting authorities, the grid that powers the United States has no single owner or point of accountability. National security is not its primary concern. The Department of War can decide that it needs a 500 MW AI facility at Fort Bliss, but cannot by itself make 500 MW appear there. The chain—from site selection, interconnection request, queue position, environmental review, to transmission upgrade, construction, commissioning—runs through institutions that, individually, are doing what they were designed to do. Collectively, they cannot move at the speed a defense requirement demands.

As Dan Wang argued in

Breakneck

, the United States has become a lawyerly society, one in which physical infrastructure is the hardest component. Some fast energy projects have sought speed by leaving the grid queue and building on-site power instead, a process known as behind-the-meter. Enverus projects that behind-the-meter projects will account for roughly

40 percent

of installed data center capacity by 2030, as developers try to escape grid interconnection timelines.

But this is not a solution for the Department of War. Self-contained, off-grid power installations measured in hundreds of megawatts cannot win a contest in the scale of hundreds of gigawatts. Concentrated, fixed, high-value installations (whether synthetic fuel plants in 1944 or availability zones in 2026) are also precisely the targets an adversary finds easiest to name.

When Churchill advanced Britain’s situation, he didn’t build a Navy-only refinery, but took a state position in an entire supply chain, rewiring national policy around it.

To degrade U.S. military AI capability, no adversary needs to strike the physical grid. The permitting system is already doing the job. The Department of Energy’s own projection—a hundredfold increase in grid outage risk by 2030—is a rational self-assessment. The front line of this arms race runs through FERC’s interconnection queue.

The Directive

Churchill’s model was not a policy mandate, but a directive with a specific institutional form, backed by an organization with the authority and continuity to execute it. The contemporary problem demands the same sequence: a mandate, then something with the authority and continuity to carry it out.

A credible response from Washington rests on four concrete pillars:

First, military compute demand needs to be forecast in megawatts, by site and by year. Industry, including grid operators and utility providers, can then bid to supply it. The government should offer accelerated permitting and tax incentives for companies to build capacity to serve the increased demand.

Second, interconnection queue position, transmission rights, and long-term firm capacity need to be held as defense assets. The strategic petroleum reserve exists because Washington decided oil supply was too important to leave entirely to markets. The same logic would apply here.

Third, Federal siting authority for defense-designated load needs to exist, as every downstream intervention (from financing, to construction and commissioning) runs on a schedule the DoW currently has no power to accelerate.

Fourth, the United States has distributed its war risk onto commercial companies that were never set up to carry it. If cloud infrastructure serves as a critical component in military infrastructure, then its physical defense is a military obligation.

The Will to Win

The arms race is measured in gigawatts. Wars, conflicts, and strategic competition have not historically been decided by the side which possesses the most resources, but by the country that has an effective strategy and the institutions to deploy it. Britain had less oil than its competitors but won because Churchill built the machine to secure it. Germany had world-leading chemistry and superior military strategy, yet lost because it couldn’t secure what that chemistry depended on. The United States has abundant electricity but is currently losing the institutional question, and not to a faster adversary but to a zoning board.

The engineering capability and capital exist. What remains is institutional architecture, and the will to reorganize policy around the problem. That will remains slow to arrive within Washington’s defense establishment. But as shown on the sunrise of March 1, 2026, unless Washington acts now, it will be forced to do so under fire.

Thanks for reading First Breakfast! Subscribe for free to receive new posts and support our work.

20

1

Share

Previous

Next

A guest post by

Katherine Lambert

AI/ML researcher focused on AI security and defense.

Subscribe to Katherine