The U.S. Army has committed to a million small drones by 2028. The Department of Defense has not decided who will authorize how they fire.
That is not a materiel gap. The volume is on order; Nvidia’s latest earnings guidance says the compute to run it is on the way. It is a governance gap, and it is now the single largest exposure in the American national security posture entering the fourth quarter of 2026.
Consider what landed inside a seventy-two-hour window this week. On Tuesday, Nvidia posted $96.2 billion in second-quarter revenue, with $89 billion of it in the data-center segment and guided to $108 billion in the September quarter — an annualized run rate through American hyperscalers of more than $400 billion, the largest defense-adjacent capital cycle in modern history. The same day, Boeing, Lockheed Martin, and RTX filed their responses to the Deputy Defense Secretary’s twenty-one-day production ultimatum on accelerated interceptor delivery. Nine days earlier, the Pentagon had opened the vendor portal for the $185 billion Golden Dome missile-defense program.
The chips are being funded. The primes are being pushed. The perimeter is being paid for. None of it tells you who decides when a fielded drone force pushes beyond its vendor’s safety rails during a live engagement.
That is not a hypothetical. Since March, the Department of Defense has formally designated Anthropic — the American AI lab whose Claude model is in active use within U.S. Defense environments — a “supply chain risk.” The two sides opened arguments in DC court over the blacklisting on May 19. In May, the Department expanded its classified AI work to eight companies, explicitly excluding Anthropic. On February 28, OpenAI took the other side of the same question, signing a deal to run its models on a Pentagon classified network. The day before, more than 580 Google employees, including DeepMind researchers, signed an open letter urging Google’s CEO to refuse the same kind of arrangement.
The industrial base of American artificial intelligence is not aligned on whether it will perform the classified work the drone doctrine assumes. The Pentagon’s response has been to fund compute, sign one lab, blacklist a second, and litigate.
Picture April 2027. A deployed American drone force is airborne over a contested Indo-Pacific corridor. A vendor safety guardrail on the targeting AI refuses a strike within a sixty-second window. The commander needs an override. No office at the Department of Defense — including the Chief Digital and AI Officer — holds explicit arbitration authority between the vendor’s guardrail and the commander’s judgment at operational tempo. The decision will be made ad hoc, by whichever lawyer or flag officer picks up the phone.
That is the arbitration gap. It is not the debt-service bill. It is not the interceptor shortfall. It is not even the shipbuilding backlog. It is the fact that the United States is standing up the largest defense-adjacent capital cycle in modern history — Golden Dome’s $185 billion, a million-drone target by 2028, and $1 billion for 340,000 attritable systems by 2027 — with no institutional authority designated to arbitrate how any of it fires. It is a supervisory function no institution has yet been chartered to perform, and no doctrine has yet been written to describe it.
Beijing runs the inverse problem. Paul Scharre laid out the industrial-base half of this argument in “Losing the War of the Future,” published in Foreign Affairs on June 23. What his piece understates is the demand side: even if the United States builds the drones, the doctrine and civil-military arbitration required to employ them do not yet exist at scale. The People’s Liberation Army’s answer is on the record. Three PLA researchers — Lieutenant Colonels Li Hengrui and Wang Haiyuan and Senior Colonel Cheng Jinming — published “all-domain megacity incapacitation warfare” in Military Art, the Central Military Commission’s restricted journal, in March 2025. Nathan Beauchamp-Mustafaga and Andrew Wang brought the article into the English-language conversation this summer in “Total War on Taipei,” published in War on the Rocks. The concept identifies the operational meaning of Xi Jinping’s “national total war” as an integrated arbitration architecture that fuses military, civilian, media, and party assets under a single narrative apparatus.
The PLA may struggle to execute it. The Central Military Commission, which began this term with seven members, has been reduced to two, hollowed out by disciplinary investigations. But the design is on the page. The American design is not.
Three signals will indicate by year-end whether the gap is closing or widening.
First, does the CDAO’s office designate a named override authority for wartime AI lab guardrails? Second, is the Anthropic-DoD blacklisting resolved into a joint framework for classified use, or does the litigation calcify? Third, does the FY27 defense-authorization conference name a joint force to operate Golden Dome, or fund the program without operator language? As of this week, all three are moving the wrong way.
The next war will not be lost because America bought the wrong drone. It will be lost because America bought the right one and never decided who could tell it to fire.
The Cipher Brief is committed to publishing a range of perspectives on national security issues submitted by deeply experienced national security professionals. Opinions expressed are those of the author and do not represent the views or opinions of The Cipher Brief.
Have a perspective to share based on your experience in the national security field? Send it to Editor@thecipherbrief.com for publication consideration.
Read more expert-driven national security insights, perspective and analysis in The Cipher Brief
