Listen to the audio version of this article (generated by AI).
Editor’s note: “AI Has Entered the Chain of Command” was previously published in July 2026 with the title, “The Bearish AI Headline That’s Actually the Most Bullish Signal of the Year.” It has since been updated to include the most relevant information available.
Count the number of times the Pentagon has created a brand-new command in the past 40 years. You can do it on one hand.
Special Operations Command was born in 1987, after a failed hostage rescue in the Iranian desert exposed that America’s elite units couldn’t operate together. Northern Command arrived in 2002, after 9/11. Cyber Command followed a 2008 worm that crawled into the military’s classified networks from a thumb drive. Space Force stood up in 2019, once rivals began treating orbit as a battlefield.
Each new command marked the moment the military admitted its old structure couldn’t handle a new reality.
On Sept. 30, it added another.
Secretary of War Pete Hegseth announced plans for a four-star Autonomous Warfare Command – “Autowarcom,” in Pentagon shorthand – charged with scaling drones, robotics, and AI across the entire joint force, with what he called “service-like authorities.” The doctrine behind it is a “high-low mix”: a handful of exquisite, expensive weapons alongside swarms of cheap, expendable ones.
The next day, the Army ordered up its own Futures and Autonomous Systems Command, along with a dedicated acquisition executive for autonomy.
Then, the Army selected Amazon Web Services and World Wide Technology for Project Griffin, a pilot designed to use autonomous AI to defend military networks at machine speed.
Three moves in less than a month.
Instead of simply experimenting with AI, the U.S. military is starting to organize around it.
There is a short list of technologies governments decide they cannot afford to lose:
Nuclear power. Semiconductors. Satellites. GPS. The internet.
AI just joined them.
And once the military begins building around a technology, the infrastructure behind it – and the companies that supply it – become much harder to ignore.
AI Is Now a National-Security Asset
For years, Washington mostly treated frontier AI as a commercial technology.
Powerful? Absolutely. Worth watching and regulating? Sure – but still something built by technology companies and sold to customers.
This summer, that began to change.
The government briefly forced Anthropic to restrict access to Claude Fable 5 and Mythos 5 on national-security grounds. The Commerce Department lifted those controls a few weeks later, and access was restored July 1.
Though the restriction was short-lived, Washington had shown that it was willing to treat an AI model like a strategic asset.
Now the Pentagon is going further, creating commands, acquisition offices, technical jobs, and working military programs around AI and autonomy. These systems have already seen combat, too: back in March, Hegseth confirmed that AI-enabled autonomous systems were being used in the campaign against Iran.
The question is no longer whether the military will build around these technologies. It’s how quickly it can.
Sovereign AI Is Becoming a Full-Stack Infrastructure Buildout
In 1942, when the U.S. government decided that atomic weapons were a national-security imperative, it built an industrial pipeline to ensure it succeeded – from uranium mining to enrichment to delivery systems – at a scale and speed that had never been attempted in peacetime.
We are watching the early stages of something structurally analogous.
The difference is that the ‘Manhattan Project’ of sovereign AI requires not one centralized government program but an entire ecosystem: domestic semiconductor fabs, secure data center campuses, high-bandwidth networking, stable power grids, and model development labs operating under strict security protocols.
The proof is in the usage logs.
This spring, the Pentagon signed agreements with eight major AI companies to bring advanced models onto classified networks. Within five months of launch, its GenAI.mil platform passed 1.3 million users, with military personnel running hundreds of thousands of agents.
The Army’s plans go further.
By 2028, it wants autonomous systems working across combat vehicles, resupply, reconnaissance, breaching, fires, and teams of crewed and uncrewed machines.
That sounds like a technology the military expects to use every day.
And the United States isn’t building alone.
The Sovereign AI Race Is Already Global
Japan became the first international partner in the U.S.’ Genesis Mission, committing $500 million alongside a matching $500 million from the U.S. Department of Energy – a combined $1 billion over five years to advance AI science, next-generation computing, and autonomous laboratory systems through joint teams spanning 12 DOE National Laboratories and 12 leading Japanese research institutions.
Saudi Arabia’s Project Transcendence is deploying $100 billion toward AI infrastructure, model development, and data centers.
The UAE has launched G42 as its sovereign AI vehicle, with Abu Dhabi committing billions to domestic compute capacity.
And China has been quietly building sovereign AI infrastructure for years – ChangXin Memory Technologies scaling domestic HBM production, Huawei developing its own GPU stack, and state-directed capital flowing into data center construction at a pace that rivals the hyperscalers.
Every one of those commitments reinforces the others. Sovereign AI is now a race – and races don’t have off switches.
Why Military AI Spending Gets Harder to Cut
Once a technology is classified as critical to national security, the political cost of underfunding it becomes unacceptably high. That means capital will flow regardless of economic cycles, earnings misses, or Fed policy.
But what has to be built before this strategy can work?
The answer runs through the entire AI infrastructure stack.
- Secure compute: Foreign-access restrictions mean domestic, sovereign, security-hardened data centers become a requirement, not a preference. Hyperscaler buildout just got a policy tailwind.
- Chips and memory: If frontier models are strategic assets, the chips that run them are, too. Domestic semiconductor production, Nvidia (NVDA) allocations, high-bandwidth memory supply – all become matters of national priority. Networking and optics: AI infrastructure communicates, constantly, at scales that dwarf anything the internet was originally designed to handle. All of it runs across physical fiber, switches, and optical transceivers.
- Power and cooling: Sovereign AI clusters run continuously, consume extraordinary amounts of power, and generate heat that requires industrial-scale cooling systems. That demand grows with every new model generation.
- Cybersecurity: If AI models are now in the same category as military hardware, then the security perimeter around them will be built to military-grade standards.
- Autonomy and robotics: Someone has to build the drones, robotic vehicles, and unmanned systems. Someone else has to provide the edge processors, sensors, batteries, motors, and secure communications inside them. As the military moves from small pilots to full fleets, that supplier base becomes part of the trade.
Together, these trades form a single investment thesis: own the infrastructure layer of a technology that governments have decided they cannot afford to lose.
How the Sovereign AI Race Expands the Infrastructure Trade
No country wants to discover that a rival controls the better models, the larger computing base, or the autonomous systems that could decide a future conflict.
So, one government builds. The others respond.
More data centers go up. More chips get ordered. More power is secured. More autonomous systems move from testing into production.
The race itself creates the demand.
Core AI infrastructure names – like NVDA, AVGO, ANET, and VRT – are precisely the companies that benefit most when AI infrastructure becomes a sovereign imperative rather than an enterprise discretionary.
We are watching closely for:
- New government AI infrastructure contracts and sovereign AI fund announcements
- Allied-nation buildout cadence
- Accelerated domestic fab investment, particularly anything related to secure, export-controlled advanced packaging and HBM production
- Security hardware specs for AI data centers – when DoD and allied governments start publishing requirements for secure AI infrastructure, those spec sheets will be a roadmap for which companies win.
The Bottom Line
AI entering the chain of command raises a larger question:
Can America build enough of the physical equipment to keep up?
Remember the Pentagon’s own blueprint: a “high-low mix” of exquisite, expensive systems and masses of cheap, expendable ones. Right now, America can’t produce enough of either. Drones need batteries, motors, sensors, and secure communications. Autonomous vehicles need processors and factories. AI systems need chips, power, networking, and domestic manufacturing capacity.
And the military needs a supply base that can deliver all of it quickly – after decades of letting that capacity drift overseas.
Rebuilding it is going to take more than Pentagon budgets. It’s why some of the biggest names on Wall Street are now sitting at the same table as defense leaders and emerging technology companies, working out how private capital gets pulled into the effort.
That is the opportunity I’ve been researching.
AeroVironment is one of five companies I’ve identified to profit from it – a group that spans established producers, the emerging technologies replacing them, and the critical materials both depend on. I lay out all five in a new presentation, along with the developments I believe could bring them to the market’s attention very soon.
