For centuries, military strength was measured in steel and firepower: fleets of battleships, tanks, and jets — the key indicators of strategic advantage. Today, supremacy is defined not just by physical presence, but reaction time: how fast a force can sense, decide, and act in rapidly-evolving environments. Artificial intelligence is at the centre of this shift, promising autonomous systems, predictive logistics, real-time intelligence, and resilient cyber defence.
However, AI in defence will stand or fall on one thing: how the underlying data layer performs.
When it comes to putting compute power on military devices, we’ve come a long way. As far back as 2019, the US Air Force was already using open source technologies (Kubernetes and Istio) in F-16 fighters. Now, the Air Force Battle Lab is experimenting with AI under the Maven Smart System, to better “prioritize targets” and enhance military operations, processing full-motion video and imagery from drones.
So the processing power is there. The management platforms exist. And yet without strong, agile, and resilient data foundations, will military AI be just another shiny promise?
Why defence data is fundamentally broken
As most industries are discovering, the architecture of yesterday was not designed to address today’s challenges, or to capitalise on Artificial Intelligence (AI). Something the defence sector is acutely aware of.
Inherited data practices are fragmented, siloed, and painfully slow. In most cases, ISR (intelligence, surveillance, reconnaissance) data feeds are delivered live via one system, logistics data originate in another system, cyber threat telemetry in yet another. Analysts and operators stitch data together through brittle pipelines and manual workflows.
Latency continues to be the major, and most serious, side-effect of this fragmentation. Where decisions are made on a second-by-second basis, the immediacy and accuracy of data is critical. If data is ingested, normalised, and delivered from multiple systems it quickly becomes outdated. Added to this, AI development is siloed, often for security reasons. This means training pipelines are batch-oriented, meaning models learn from yesterday’s battle, not today’s.
In the current paradigm, autonomous systems risk acting on partial or outdated information, and commanders cannot trust AI outputs for real-time decision-making.
Where legacy infrastructure falls short
Yesterday’s infrastructure simply cannot keep pace with escalating cyber defence threats or autonomous drone swarms. Traditional relational databases and batch ETL (extract, transform, and load) data pipelines were designed for back-office accounting, not live combat environments, and have their shortcomings.
Firstly, the rigid schemas of traditional models don’t like a varied mix of structured, semi-structured and unstructured data sets. Fixed relational tables cannot handle the messy mix of data generated in rapidly evolving defence scenarios.
Secondly, centralised architectures are not suited to complex defence contracts. A single data warehouse may work for finance, but it collapses under the needs of distributed, multi-domain operations. Finally, traditional models lack deep enough context for AI-legacy stores and lack vector search, graph reasoning, and temporal memory — features that modern AI workloads demand.
As the defence industry rallies around AI to help it transform, how mission-critical data is collected, analysed, and delivered is fast becoming the next frontier.
Agentic AI demands a new defence data layer
There are 3 key trends driving the need for a more unified, and adaptive database strategy:
- The rise of agentic AI and autonomy means drones, UGVs, and cyber agents need persistent, context-aware memory to act independently.
- Multi-domain operations require that land, sea, air, cyber, and space domains are fused into a single operational picture, something that current database methodologies struggle with.
- The volume of ISR data overwhelms most legacy systems. Satellites, UAVs, and sensors generate terabytes of raw data per hour, demanding real-time filtering and graph-based correlation.
Autonomous and agentic AI do not simply “query” data. They live in it. For them to operate safely and effectively, the data layer must provide several non-negotiable services. Without these capabilities, autonomous systems risk becoming brittle, unsafe, or even exploitable.
It must provide persistence; a durable memory of prior actions, decisions, and observations. It must also offer contextual reasoning including Graph traversal to link events, actors, and places. Real-time responsiveness is crucial; with millisecond-fresh updates through live queries. Finally, it must bring secure compartmentalisation in the form of row- and field-level permissions for coalition or mixed-trust environments.
The benefits of a more unified approach
As demand for AI surges in defence, fresh innovation in the database is quickly emerging. I expect to see the arrival of a more unified data layer that is real-time, multi-model, secure, and edge-ready that better equips defence organisations to act faster, more accurately, and that builds confidence in AI.
There are several novel models, including ours at SurrealDB, particularly suited to military use cases. In a cyber defence scenario, these new technologies can power agents that detect anomalies, correlate them across networks, and autonomously isolate compromised nodes, without waiting for human-in-the-loop approvals:
- Real-time decision-making allows live queries to push data directly to operators or agents, ensuring ISR feeds, cyber alerts, or logistics updates are always current.
- Multi-model reasoning means a single SurrealQL query can combine similarity search (vector), knowledge-graph reasoning (graph edges), and transactional updates (SQL) in one round-trip.
- Tactical edge deployment allows a 3MB build to run on any device with an operating system or a standard library without saturating its resources, making it possible to run close to the action.
- Agent memory allows AI agents to view versioned records: to “time-travel” through data, replaying prior states for audit, reproducibility, or situational awareness.
Data-centric warfare is different
Defence is shifting from platform-centric to data-centric warfare. AI will only be as effective as the database beneath it. The fragmented, brittle systems of the past cannot support autonomous agents, ISR fusion, or multi-domain command. The message is clear: winning tomorrow’s conflicts will not only depend on superior weapons, but on superior data. And that begins with the database.
Join us and 200+ hackers from all across Europe and beyond for the European Defense Tech Hackathon in London on September 25-28, 2025, in London, UK, as part of London Defence Tech Week. SurrealDB is proud to sponsor this year’s hackathon.
