DIA shifts to modular data platforms to keep pace with AI

The Defense Intelligence Agency is overhauling its technology infrastructure to address a fundamental mismatch: government acquisition timelines measured in years versus commercial innovation cycles now compressed to months.

DIA Chief Information Officer E.P. Mathew told a recent GDIT summit that semiconductor capabilities can improve 1,000-fold over the government's standard five-year planning cycle. Moore's Law—the traditional doubling of chip capacity every two years—has accelerated to roughly seven months, creating what Mathew characterized as mounting national security vulnerabilities when agencies deploy outdated systems.

Why it matters

Intelligence agencies depend on processing massive data volumes to identify threats before they materialize. When technology refresh cycles outpace acquisition processes by an order of magnitude, defense organizations risk fielding capabilities that are obsolete before deployment—a gap adversaries can exploit. The modular approach DIA is pursuing offers a potential template for other agencies struggling with the same structural problem.

Zero-trust data foundation

Mathew outlined a shift from infrastructure-heavy, application-centric frameworks toward what he called agile, data-centric environments. At the core is a zero-trust architecture where every dataset is tagged, cataloged, and encrypted, with access granted dynamically through centrally managed policies rather than perimeter defenses.

"Our goal is to build a data environment where the focus is on you having access to data by policy only," Mathew explained. Crucially, security controls travel with the data even outside core repositories, with periodic credential retesting to maintain operational integrity.

This policy-driven data layer serves as the mandatory foundation for AI deployment. "An organization cannot do AI unless you do data correctly," Mathew said.

Three-step integration process

To scale AI into intelligence operations, DIA follows a structured approach. First, data must be structured and restricted through granular policy entitlements. Second, the agency is implementing a Modular Component Platform (MCP) that allows diverse components to access data simultaneously. Finally, the network must integrate semantic AI capabilities—including knowledge graphs and entity resolution—to transform raw information streams into actionable intelligence.

The modular architecture enables DIA to swap individual software components as superior commercial tools emerge, avoiding long-term proprietary vendor lock-in and enabling evolution from basic decision support to comprehensive automation.

Workforce challenges

Deploying these architectures requires specialized technical expertise, which has been complicated by a Defense Reform Program reallocation that resulted in the departure of roughly 22 percent of DIA's network and software engineering workforce.

To rebuild internal capability, DIA has moved away from traditional post-sales vendor support models. The agency created an internal training lab and launched a "Training with Industry" initiative that embeds personnel directly within commercial technology companies for six-month rotations. Operators return equipped to serve as internal product leads who can maximize existing software investments.

The agency also began seeking industry partners earlier this year to test, evaluate, verify and validate new AI technologies.

These details were first reported by DefenseScoop, which produced the coverage in partnership with GDIT.