Golden Dome’s AI Playbook Comes From Its Own Contractor

Golden Dome cannot run without artificial intelligence fast enough to out-think a missile in seconds, and Booz Allen Hamilton just won a contract to help build it. The consulting and technology firm is also the one telling the Pentagon what that AI needs before it can be trusted with the job.

A senior Booz Allen advisor laid out four engineering conditions for AI-run battle management in a new essay. Golden Dome, the Trump administration’s planned shield against ballistic, hypersonic and cruise missiles, already carries cost estimates ranging from $175 billion to $1.2 trillion, and Congress says it still cannot see the program’s full architecture.

Four Conditions Before the Algorithm Fires

Lt. Gen. Trey Obering spent decades as an Air Force officer before becoming a senior executive advisor at Booz Allen Hamilton, a consulting and technology firm that works across the U.S. defense industrial base. In a new essay on advanced battle management, he starts with Ukraine.

Software called GIS Arta, often nicknamed a rideshare for artillery, matches sensors, drones, forward observers, even smartphones, to the nearest available gun the way a rideshare app pairs a rider with a driver. What used to take Ukrainian gunners 15 to 20 minutes of manual coordination now takes under a minute, Obering wrote, producing firing solutions that regularly confounded Russian counter-battery crews.

Golden Dome scales that same problem by orders of magnitude, from a few thousand guns on one front line to hundreds of thousands of sensors, interceptors and satellites spread from the ocean floor to low Earth orbit.

Obering’s point is that simply adding AI to that architecture will not get the job done. He lists four engineering conditions that have to be met first.

• Edge-first, not cloud-first. The system has to make split-second calls in space and at sea with no guarantee of a steady connection, and none of the on-demand computing power a data center provides.
• Hardened against adversaries. Booz Allen researchers have shown how attackers can manipulate live video feeds inside computer vision systems, and Obering warns that untrusted, open-source models built by rival governments are already showing up in U.S. infrastructure.
• Deterministic, not just accurate. Commanders need models that behave the same way every time and can explain their own outputs, which means fusing many narrow, purpose-built neural networks rather than trusting one general model across every kind of sensor.
• Fielded in days, not years. Obering points to Ukraine’s compressed innovation cycles and argues engineers should test high-risk scenarios in digital proving grounds instead of waiting on traditional acquisition timelines.

The same tension shows up well beyond missile defense. Financial institutions face pressure toward building real operating models before trusting AI with compliance decisions, and a majority of Indian chief executives say they are being pushed to adopt AI systems they cannot yet measure. Golden Dome just raises the stakes from a compliance fine to a live intercept.

The Space Force has also put Booz Allen’s name on a contract to help build exactly what that list describes.

What Is Golden Dome Actually Building?

Golden Dome is the Trump administration’s proposed shield against ballistic, hypersonic and cruise missiles, combining space-based sensors, ground and sea interceptors, and an AI-run command network meant to link them at machine speed. Trump signed an executive order on January 27, 2025, initially calling for what it labeled Iron Dome for America; by the time he detailed the plan at the White House that May, it had been renamed Golden Dome. The Pentagon has set a goal of an initial capability by 2028, with the full system not expected until the mid-2030s.

Senior officials showcased the program’s first concrete milestones on April 23, 2026, at an event in Virginia: a completed architecture blueprint, a new Command-and-Control (C2) Consortium meant to link multiple vendors, and active contracts for key system components. Gen. Michael Guetlein, the Space Force general Trump picked to run the program, told the crowd the effort would forge “a shield that is layered, integrated, and automated.”

Defense Secretary Pete Hegseth has described Golden Dome as a “system of systems” that will be fielded in phases, prioritizing defense where the threat is greatest, leaning on sensors and interceptors the Pentagon already owns rather than starting from nothing.

What We Know:

• The architecture blueprint, C2 Consortium and initial component contracts were confirmed at the April 2026 Virginia event.
• The Space Force has named 12 companies to develop competing space-based interceptor (SBI) prototypes, on top of a separate demonstration deal with Northrop Grumman targeted for 2027.
• The Pentagon’s fiscal 2027 budget request seeks $17.5 billion for the program.

What’s Unconfirmed:

• The “objective architecture,” including exactly how many satellites and interceptors will be built, has not been made public.
• The program’s real long-term cost, which the Congressional Budget Office says is impossible to estimate without that architecture.

Booz Allen’s contract sits inside that gap between what is confirmed and what stays classified.

The Firm Writing the Rules Also Won the Job

Booz Allen Hamilton is being paid to help build the AI it just told the Pentagon how to engineer. The Space Force’s Space Systems Command awarded the firm a contract to assemble a testbed interceptor system called Brilliant Swarms, a constellation of small satellites that uses its own battle management and command-and-control algorithm to track, lock onto and intercept targets with minimal human input at each step.

Equity analysts tracking the program count Booz Allen among the firms doing AI targeting and data fusion work inside Golden Dome, alongside Amentum’s enterprise architecture and Elbit Systems’ interceptor heritage. More than 2,000 companies have qualified to bid on Golden Dome task orders, according to one industry tracker, everyone from Northrop Grumman down to venture-backed startups.

The Space Force separately named 12 companies to build competing SBI prototypes, backed by agreements worth a combined $3.2 billion. Northrop Grumman is teaming with commercial spacecraft manufacturer Apex to demonstrate an orbital interceptor by 2027, a step the companies say could prove a single demonstrator can grow into a real orbital network for homeland defense.

Obering’s four conditions still read like sound engineering advice on their own terms. Cato Institute researchers argue the winners of a project this size are the contractors who win the bids and the service branches running it, calling it the work of the Iron Triangle at its finest. Booz Allen wrote the conditions its own contract now has to satisfy.

A Price Tag Between $175 Billion and $3.6 Trillion

Ask what Golden Dome costs, and the answer depends entirely on who is doing the asking.

The gap comes down mostly to one layer: space-based interceptors. A notional constellation of 7,800 satellites in low Earth orbit, enough to engage 10 intercontinental ballistic missiles (ICBMs) launched at once, would cost $723 billion to acquire and deploy on its own, the Congressional Budget Office (CBO) found in its May 2026 report, versus roughly $139 billion for ground- or sea-based systems.

That single space layer eats up about 70 percent of total acquisition costs. Each satellite would run about $22 million on average, and low orbits decay fast enough that the whole layer needs replacing every few years.

Gen. B. Chance Saltzman, the Space Force’s chief of space operations, has waved off the sticker shock. Asked about an earlier $542 billion estimate at a security summit, he said he had “never seen an early estimate that was too high.” Even at $1.2 trillion, the CBO concluded, the system would not be an “impenetrable shield” against an all-out attack from Russia or China.

Lawmakers Say They Are Flying Blind

Congress has now funded Golden Dome for two budget cycles without seeing what it is actually paying for. Appropriators on the House and Senate defense subcommittees wrote in January that they were “unable to effectively assess resources available to specific program elements” for fiscal 2026, because the Pentagon had not supplied a deployment schedule, cost schedule, performance metrics or a finalized architecture.

Ten members of Congress went further. In a letter led by Sens. Ed Markey, D-Mass., and Jeff Merkley, D-Ore., they warned the plan could become “prohibitively expensive, operationally ineffective, massively corrupt” and capable of reigniting a nuclear arms race with Russia and China.

Nothing more than a massive giveaway to defense contractors paid for entirely by working Americans.

Merkley said that after the CBO’s estimate landed in May. The Union of Concerned Scientists, a science advocacy nonprofit, said the entire premise remains “technically unrealistic in practice, profoundly destabilizing and wildly expensive” in a statement marking the executive order’s first anniversary.

Guetlein’s response has been that outside groups “take the cost of a legacy system and they multiply it out” to reach inflated totals, and he has pledged the Pentagon will not build space-based interceptors at all if they turn out to be unaffordable.

Ukraine’s Battlefield Lesson, Applied to Orbit

The kill web Obering describes is not just a slogan. Ukraine’s shift, in his account, moved coordination from single platforms toward distributed networks, replaced a linear chain of command with simultaneous strikes, and let software do work that used to require officers on radios. That shift happened inside one country’s artillery corps, with a handful of sensor types feeding a few thousand guns.

Golden Dome asks similar software to coordinate a system stretching from the ocean floor to geostationary orbit, deciding in milliseconds whether a heat signature over the Pacific is a decoy, a hypersonic glide vehicle, or nothing at all. A September 2025 American Enterprise Institute study, citing Defense Intelligence Agency estimates, projects China and Russia could field more than 16,000 missiles capable of reaching the U.S. homeland by 2035.

America has tried a version of this before. President Ronald Reagan’s Strategic Defense Initiative promised a space-based shield against Soviet missiles in 1983 and never got past research and prototypes, even after billions in spending. Space Force officials developing Golden Dome’s orbital interceptors describe their work as revisiting those Reagan-era concepts with cheaper launch costs and better computing than existed forty years ago.

Tom Karako, director of the Missile Defense Project at the Center for Strategic and International Studies, has walked through Golden Dome’s biggest feasibility questions alongside the think tank’s Aerospace Security Project director, Kari Bingen. His read on the administration’s original three-year completion promise: “really ambitious.”

The Next Test Happens in Orbit, Not a Simulator

The digital proving grounds Obering describes, simulations meant to catch failures before satellites reach orbit, get their real test soon. Northrop Grumman is due to demonstrate an orbital interceptor prototype by 2027. The Missile Defense Agency’s (MDA) Next-Generation Interceptor (NGI), built by Lockheed Martin, is set to close out its design phase by the end of 2026 and fly for the first time in 2029.

Sen. Angus King, an independent from Maine, pressed Guetlein in April on whether the math scales at all. Point defense of Guam alone, using existing Patriot, THAAD and Aegis systems, costs $3.5 billion, Army Gen. Francisco Lozano testified. Guetlein’s answer was that Golden Dome’s networked, self-contained weapons should not be compared to a single-island defense in the first place.

Congress gets its next chance to test that answer when the Pentagon’s fiscal 2027 budget request comes up for a vote, funding another year of work on a shield that, by its own advisor’s account, still needs four engineering conditions nobody has fully met.