Defence has committed the money and set the date.
Delivery now rests on data quality, integration and speed of insight - the least glamorous and most decisive part of the programme.
The Defence Investment Plan published on 30 June 2026 commits £7.5bn to digital across the next four years: £1.8bn for the Digital Targeting Web and £5.5bn for the Digital Backbone, rising to at least £17bn between 2030 and 2035. The Strategic Defence Review before it set the delivery date - a digital targeting web connecting sensors, deciders and effectors by 2027. The Chief of Defence Staff put the underlying logic plainly at London Tech Week this June: the side able to diffuse and adopt technology faster than an opponent will win.
None of it works on bad data. Strip away the doctrine and a targeting web is a data integration problem: many sources of uneven quality, arriving at different tempos and different classifications, that must be resolved into a single trustworthy picture fast enough to act on. The Data Strategy for Defence named the six rules that make this possible: sovereignty, standardisation, exploitability, security, curation and endurance - and they are still the right rules. What remains incomplete is the engineering underneath them.
That engineering is not novel, and it is not defence-specific. It is the same discipline we apply where a wrong number costs money, reputation or safety.
Taskforce RAID, established under the Chief of Defence Staff with £100m ringfenced, is pursuing four challenge sets: machine-augmented intelligence fusion, mapping the electromagnetic environment, automating operational planning, and AI-enabled swarms. Every one of them is, at root, a fusion problem. A model fed poorly curated data does not fail loudly it fuses confidently and wrongly, which is considerably worse than not fusing at all. Curation is where the risk actually sits.
As an example, our partners Qlik Talend Cloud does the unglamorous work here: profiling data at source to quantify how bad it genuinely is, applying rules for cleansing, matching and de-duplication, assigning stewardship so a named human owns each domain, and producing a trust score you can put in front of an authoriser. Our partners Microsoft Purview complements it by cataloguing and classifying assets across the estate and attaching sensitivity labels that travel with the data which is precisely what a Defence Data Catalogue needs to be, and what makes cross-domain handling auditable rather than manual.
“Better data isn't the outcome. Knowing where it came from and how far to trust it is”
Seatal Patel - Founder and Executive Director
The Digital Backbone’s stated purpose is to replace fragmented services and networks with an enterprise platform, with a Defence-wide Secret Cloud minimum viable product due later this year. In practice, fragmentation will persist for years: legacy systems that cannot yet be retired, coalition feeds MoD does not own, sensor volumes that never belonged in a warehouse, and business platforms mid-migration.
The pattern that works is layered landing and curation: raw ingest, conformed and quality-assured, then business-ready all built on open formats. Databricks with Delta Lake handles the high-volume, semi-structured and streaming end: imagery, telemetry, geospatial, sensor and log data at scale, with Unity Catalog carrying lineage and access control across it. Snowflake handles governed structured analytics and, critically, secure sharing between organisations without copying data are valuable where Commands, DE&S, primes and allies each need a view but not custody. Microsoft Fabric and Azure Data Factory tie the wider Microsoft estate together, and since most of Defence’s business data already lives there, that is usually the shortest route to a first delivered outcome.
Choosing open table formats Delta, Iceberg, Parquet is a sovereignty decision as much as a technical one. Data written in open formats can still be read by a different engine in five years. That is the endurance rule made real, and it answers directly the common standards, open architectures and data frameworks approach the (Strategic Defence Review) SDR mandated through the Uncrewed Systems Centre.
Fast analytics is where the investment becomes visible to the people it is meant to serve. Qlik’s associative engine lets an analyst pursue a question sideways, following an anomaly through related datasets without commissioning a new report on structured and unstructured data - which suits intelligence, readiness and support analysis far better than fixed dashboards. Databricks SQL and notebooks serve the data science end, from geospatial exploitation to predictive maintenance across platforms and fleets.
The metric that matters is cycle time. If a commander’s question takes days to answer, the answer is irrelevant. Getting it to minutes is an engineering achievement with operational consequence and it comes from modelling, semantic layers and governance, not from buying another tool.
The Defence AI Centre’s AI Model Arena signals how MoD intends to buy AI: evaluated, compared, evidenced. Alongside it, UK policy holds that humans remain accountable for lethal force, with context-appropriate human involvement in AI-enabled systems. Both requirements land squarely on the data layer. A capability that cannot show what it was trained on, which version produced a given output, and how that output was validated will not clear assurance however well it performs in a demonstration.
Unity Catalog and MLflow on Databricks, Snowflake’s governance and access history, and Azure AI Foundry’s evaluation and content-safety tooling produce that audit trail as a by-product of the build rather than as retrospective paperwork. Project FRONTIER’s £100m for foundational AI infrastructure is money for exactly this: the plumbing that makes models deployable and defensible. So is the £80m for land AI command and control and the £20m for underwater AI capability lines that will each stand or fall on data readiness.
The SDR also requires at least 20% of HR, finance and commercial functions to be automated by July 2028, and the £320m equipment management platform awarded in October 2025 shows the same intent across support and logistics. These are lower-classification, high-volume, well-understood data problems and they are where a delivery partner can prove value in weeks rather than years, building the credibility and the reusable foundations that harder operational work then depends on.
Procurement has moved to match. The SDR set contracting in three months for rapid commercial exploitation, mandated at least 10% of the equipment procurement budget against novel technology, and Commercial X reports 580 contracts at an average 31 days to contract. The routes in are the Neutral Vendor Framework for Innovation, the Dynamic Market and the Defence Sourcing Portal are open, and the MOD SME Action Plan published in July 2026 is explicit about widening the supplier base.
At Data Technology we have spent decades delivering data strategy, platforms, integration and analytics for organisations where the data is messy, the estate is federated and the decisions carry real cost across financial services, transport, automotive, construction and national-scale retail operations. The tooling is the same tooling. The disciplines are the same: profile before you promise, govern before you scale, put something usable in a user’s hands in the first sprint. What Defence adds is classification, sovereignty and consequence - design constraints, not obstacles.
The £7.5bn is committed and 2027 is close. The programmes that hit that date will be the ones that fixed the data first.