$600M for a Classified AI Supercomputer: Inside the Pentagon's $1.5B Reprogramming Bid

On September 4, Inside Defense obtained a budgetary reprogramming package that shows the Pentagon asking Congress for permission to move nearly $1.5 billion across fiscal year 2026 and FY-25 accounts. The destination is not a new fighter jet or a shipbuilding program. It is artificial intelligence — with the single largest tranche, $600 million, earmarked to build a classified, high-performance AI compute center at the Top Secret/Sensitive Compartmented Information (TS/SCI) level.

The request lands at a telling moment. The Department of War’s own AI strategy, released in February 2026, promised an “AI first” warfighting force, but as analysts at War on the Rocks noted in July, the strategy has a funding problem: Operation Epic Fury — the campaign launched on February 28 against Iranian missile and naval capabilities — imposed costs no budget anticipated, and money that would normally be re-scrambled in September, the last month of the fiscal year, started moving months early. This reprogramming package is the latest installment of that scramble, and it doubles as a window into where the Pentagon’s AI money is actually going when the budgeting formalities are stripped away.

What the package asks for

Reprogramming requests are the Pentagon’s mechanism for moving money between accounts after Congress has already appropriated it. Anything above certain thresholds requires congressional approval, which is why these packages — usually leaked or obtained rather than announced — are among the most honest documents in the defense budget. This one covers increases and decreases across FY-26 and FY-25 funds, and according to Inside Defense it ties the shifts to two drivers: a “major expansion of artificial intelligence capabilities” and “changing requirements in the Middle East following Operation Epic Fury.”

The centerpiece is the $600 million classified compute center. The funds would procure advanced computing infrastructure, hardware, and integration services to rapidly expand AI, machine learning, and enterprise data analytics across the Joint Worldwide Intelligence Communications System (JWICS) — the top-secret network that carries the intelligence community’s most sensitive traffic. In plain terms: the Pentagon wants frontier-class compute inside the classified enclave, because models that ingest classified intelligence data cannot be trained or run on commercial clouds.

Describing the effort as an “urgent requirement,” the package also directs money toward AI computing more broadly, the Maven smart system, aviation, counter-UAS, and other emerging priorities, according to Inside Defense’s summary of the document.

Why a classified AI data center, and why now

The push reflects a structural problem the military has been vocal about all year. In April, Reuters reported on a Pentagon memo making Palantir’s Maven AI system an official program of record and a core military system — a move that put sensor fusion, targeting analytics, and decision support on a path to standardization across commands. In May, DefenseScoop reported that the department wants nearly $30 billion in fiscal 2027 just to modernize its AI supercomputing arsenal. The $600 million JWICS compute center is the classified, near-term down payment on that trajectory: you cannot run Maven-class analytics — or the growing family of cyber and intelligence models — on infrastructure that cannot hold TS/SCI data.

There is also a battlefield logic. Epic Fury has been the Pentagon’s live test case for AI-assisted operations: CENTCOM’s commander publicly credited AI tools with helping plan the campaign, TRANSCOM used AI agents for logistics replanning when Spain closed its airspace to U.S. aircraft, and Pentagon leaders told Congress in April the operation carried a $25 billion price tag. A March AEI assessment noted the Pentagon had already submitted a $1.56 billion reprogramming tied to the war’s early weeks. Wars, in other words, are generating both the demand signal and the fiscal pressure — and AI is winning the internal fight for where redirected dollars land.

The context that makes this more than a budget line

Three strands of 2026 policy converge in this package. First, the February AI strategy committed the department to agentic AI, autonomous systems, and decision advantage — but strategy documents do not buy GPUs. Second, Congress itself has been pushing: the Senate’s FY-26 NDAA report flagged both a Computing Modernization Program and counter-UAS as priorities, and the Pentagon requested $2.3 billion over five years for Maven expansion back in April. Third, the fiscal calendar is merciless: with FY-26 ending September 30, unobligated money must be re-scoped or lost, and reprogramming is faster than waiting for a new appropriations cycle.

The choice of JWICS also says something about maturity. Two years ago, the debate was whether large models could be trusted with classified data at all. Today the question is procurement logistics: racks, integration services, and accreditation inside the intelligence community’s most restricted network. The Pentagon is no longer piloting AI on the side — it is building permanent classified infrastructure for it.

What to watch

Congressional reaction is the immediate variable. Reprogramming requests can be approved silently, modified, or held up by a single committee objection, and lawmakers have shown willingness to flex: the historic pattern of Congress blocking or trimming Pentagon AI transfers (Project Maven’s own rocky transition in 2022 is the cautionary tale) has given way recently to enthusiasm, but the personnel-cost reprogramming fights of this summer show appropriators are in a combative mood. Watch whether the $600 million survives intact or gets renegotiated into a smaller pilot.

The longer arc matters more. A classified AI compute center on JWICS is the kind of asset that, once built, becomes the foundation for everything that follows — classified model training, secure agent deployment, intelligence analytics at scale. It is a bet that the military’s AI future runs through networks no commercial provider can touch. If Congress signs off, the Pentagon will have moved from renting intelligence-grade AI to owning the infrastructure it runs on — one quiet budget transfer at a time.

Sources are listed in the post metadata. Figures cited from Inside Defense’s reporting on the reprogramming package obtained September 4, 2026.