The hidden barrier to aerospace AI: Legacy data infrastructure
Tobie Morgan Hitchcock is CEO & Co-Founder of SurrealDB. He is an experienced tech entrepreneur, developer, and software engineer, with 17 years’ experience in the software and cloud-computing industries. In 2021, he founded SurrealDB to build the ultimate cloud database for tomorrow’s applications.
Of all the industries that stand to benefit from Artificial Intelligence (AI), aerospace is one where the technology is already proving transformational. Across every stage of design, manufacturing, and operations, AI is enhancing efficiency, safety, and the pace of innovation.
In aircraft design, AI-driven simulations and generative tools help engineers create lighter, stronger, and more aerodynamic structures faster than ever. In manufacturing, AI-powered robotics and quality control systems are reducing errors and improving precision. In fleet management, predictive maintenance powered by machine learning identifies potential failures before they occur, cutting costs and downtime.
Operationally, AI is optimising flight paths, air traffic management, and fuel consumption, saving money and improving sustainability. In space exploration, it supports autonomous navigation, data analysis, and mission planning for spacecraft, rovers, and satellites.
The commercial aerospace industry increasingly recognises the need for connected, AI-ready data. The TCS Future-Ready Skies Study 2025 shows that 40% of leaders plan for lights-out manufacturing within 5–7 years, while many warn that MRO revenue and delivery performance will suffer if they cannot scale digital systems by 2028. Crucially, nearly two-thirds are open to agentic AI running supply chains.
This is equally vital for defence, where modern threats compress decision cycles across multiple domains. Only a connected data fabric allows humans and AI agents to coordinate faster and more safely than an adversary.
AI and the connected battlespace: How autonomy is reshaping defence aerospace
A connected battlespace – linking sensors, aircraft, weapons, and commanders so everyone ‘sees the same truth at the same time’ – is becoming essential. AI agents – software and autonomous systems – can perceive, decide, and act across air, land, sea, space, and cyber as one.
There is already proof that trusted air combat autonomy is possible at the tactical edge, not just in simulation. DARPA’s ACE programme, which successfully flew AI-controlled dogfights in real jets, is one recent milestone.
Apart from faster command decisions, recent programmes include collaborative air systems – such as the USAF’s CCA – moving toward operational prototypes of teaming aircraft, accelerating autonomy at scale.
Why legacy infrastructure is holding AI back
As companies are finding, AI is only as effective as the infrastructure beneath it. Data-driven failures are a common and persistent issue (and a huge headache) for architects and engineers.
Aerospace organisations, as in every sector, are struggling with the quality and usability of data, and the expense of normalising manifold strands of data. Without infrastructure that supports learning, adaptation, and integration, even well‑funded AI initiatives tend to underperform in terms of profitability, efficiency, or strategic value.
Legacy technologies (rigid systems, data silos, inconsistent formats, and brittle workflows) are a core part of this problem: they make it hard to embed AI outputs meaningfully and often require workarounds rather than seamless adoption.
Because traditional databases are often rigid, siloed, and optimised for structured data, they hamper the adoption of AI. This makes it difficult to integrate and analyse the vast and disparate datasets – sensor data, maintenance logs, flight records – needed for AI models to succeed.
What’s more, legacy systems frequently lack the scalability and interoperability needed for modern AI and cloud-era patterns. Many run on outdated stacks with limited support for contemporary APIs or data-exchange standards, making integration with AI platforms costly and time-consuming.
This technical debt delays innovation, elevates cybersecurity risks, and makes it harder to realise AI benefits in predictive maintenance, flight optimisation, or autonomy. To unlock AI’s potential, the aerospace sector must modernise its data layer to be more flexible, scalable, and AI-ready.
The gap is not just models – it’s memory and state (the live operational context and its history):
- Silos vs. agents. Traditional stacks split operational facts across PDM/PLM, flight ops, MRO, telemetry, and C2 tools. Legacy databases optimised for static records and overnight ETL simply can’t keep pace. Agents need one place to subscribe to events, reason over relationships, and commit actions – with an audit trail.
- Thread breakage. A digital thread requires end-to-end traceability. Standards like the ISO 10303-239 (STEP AP239/PLCS) and the ASD/AIA S-Series exist, but turning scattered, brittle schemas into a consistent, queryable operational picture is hard when data lives in many silos.
- Governance at mission speed. Autonomy operates under strict rules (e.g., the US DoD Directive 3000.09). You must be able to prove what the agent knew, when, and why it acted – something batch pipelines and ad-hoc data marts struggle to reconstruct.
- Edge reality. Disconnected operations (ranges, forward bases, contested comms) demand local state with eventual reconciliation. Centralised, cloud-only patterns create the very fragility we’re trying to remove.
Unlocking the full potential of AI: How to move beyond the bottleneck
The next leap for aerospace AI won’t come from algorithms alone, but from building smarter, more adaptable data foundations.
A multi‑model approach
Aerospace operations generate and depend on a wide variety of data: telemetry (time-series), maps and tracks (geospatial), documents and logs (unstructured), parts and maintenance (relational), similarity signals for search (vectors), and the relationships between aircraft, subsystems, missions, crews (graph).
Modern multi-model databases (such as SurrealDB) keep all of this together, so digital twins are live (not static files) and digital threads are queryable end-to-end. Key capabilities include: relationships (graph) and direct links for tracing from tail number to component to work order; real-time subscriptions so apps/agents see updates as they happen; change history and replay (CDC) to audit or sync downstream systems; and fine-grained access controls to share the right slice with the right team.
Practically, this gives you one reliable data store where the twin’s current state, its lineage (thread), and an agent’s memory can live together – backed by safe transactions so concurrent updates don’t corrupt data.
Edge-readiness
This industry is characterised by live, in-the-field decision‑making with data from a dizzying array of sources: real‑time sensors, flight control, autonomous or semi‑autonomous vehicles, missile defence, ISR (Intelligence, Surveillance, Reconnaissance) feeds, to name a few.
Batch ETL, data warehouses, and rigid relational databases can’t keep up: data is too stale, pipelines are fragile, and updates are too slow. New models support real‑time queries, local operation (including on edge or air‑gapped/constrained hardware), and eventual reconciliation when links return – so units can act now and sync later.
Context, continuity, and memory for agentic AI
Agents (on aircraft, in operations centres, or in MRO) need more than snapshots: they need context over time – what was observed, what decision was taken, and what happened next.
A multi-model store lets you persist that as a durable, versioned-over-time record of observations, decisions, and outcomes (built from events + relationships + change history). That’s explainable, auditable memory – not just a cache.
Secure operations
Missions span organisations and clearances. Flexible schemas and built-in, least-privilege access let you expose only the necessary data to primes, depots, or allies. Because the core is unified, teams avoid stitching together many tools – reducing fragility and attack surface.
Leadership priorities for enabling AI transformation in aerospace
As aerospace companies push the boundaries of what’s possible with AI, it’s clear that putting in place the right technical foundations will significantly improve the outcomes and long-term viability of AI projects.
In TCS Future-Ready Skies Study 2025, Ozgur Tohumcu, General Manager, Automotive & Manufacturing, Amazon Web Services, confirms that “data continues to be a differentiator for our customers with many rethinking their enterprise-wide data strategies to fully leverage the benefits of AI”; and Taylor Brown, COO and Co-founder, Fivetran agrees that “in aerospace, resilience and precision depend on reliable, transparent data – it’s the foundation that enables AI, digital twins, and the next generation of intelligent operations”.
The Study data tells us the aerospace ecosystem anticipates agentic operations, real-time MRO, and connected supply chains – and is ready to trust agents with them. The defence side is already proving autonomy and accelerating command decisions.
The path forward is not just about building smarter algorithms, but also about creating the right environment for AI to succeed. The common denominator is a database layer that treats relationships, time, space, semantics, and change as first-class citizens.
For aerospace AI to truly take flight, the reform must start at the data layer, because the future of AI rests on database innovation.
