In defence, the ability to process information securely and in real-time is more vital than ever. Alongside the all-highs in spending due to geopolitical tensions, there has been a shift towards autonomous technology and next generation drones. However, the sector’s reliance on centralised AI and cloud infrastructure is a major strategic disadvantage.

Defence continues to partly depend on processing in central servers, but these are far away from warzones. In the locations where automation is needed, there is often unreliable connection and low bandwidth, delaying data transfer to cloud platforms and centralised data centres. The high latency that follows presents a huge problem for time-sensitive decision-making, and when connection is cut off, AI systems with a single point of failure become defunct.

Militaries do in part use edge computing. But a more significant transition from centralised cloud infrastructure to privacy-first, edge-native AI networks will be a crucial next step. Distributed inference is ideal for defence because it makes frontline AI fast, local and virtually indestructible.

In this blog, we explain the problems with defence depending on the cloud, and present distributed inference and federated learning as a compelling solution.

[ 👋 Hi there! If you’re here to find out more about FLock, follow us on X and email us at hello@flock.io to learn what we do in distributed inference, federated learning and sovereign AI.]

The problems with cloud dependency

Poor connectivity and even connection loss

In offline, contested and low-bandwidth environments, cloud-dependent AI buckles. Moreover, in combat zones, opponents frequently attempt to cause connection loss by targeting the physical, electronic and digital infrastructure that networks rely on to function. In modern warfare, dominating the electromagnetic spectrum and the digital space is just as critical as controlling physical territory.

More prone to attacks

By transmitting military operation data to the cloud, the attack surface is expanded. This makes data more susceptible to interception or manipulation.

Central hubs are also prime targets for cyber attacks. Due to the single point of failure, opponents do not need to hack every single drone or tank. By simply targeting the central cloud with a distributed denial-of-service (DDoS) attack or a physical attack, they can paralyse entire combat networks at once. A single security breach can also give an enemy insights into troop locations, logistics and strategies.

Slower decision-making

Centralising AI in the cloud creates massive operational drag, delaying decision-making. Satellite and radio networks do not have the bandwidth to send raw, high-definition information back to central servers in real time. Even with strong connections, sending heavy raw video, scans or thermal imaging to a data centre thousands of miles away and waiting for a prediction adds seconds or even minutes of delay. But in counter-drone defence, a two-second delay means missing the target.

Data sovereignty is compromised by centralisation

Enormous quantities of data are needed to train AI for defence. But centralised AI architectures demand that all data be shipped to a single hub or third-party cloud platform for processing and model training. Palantir is the dominant AI and analytics platform used by the US, UK and NATO allies, while AWS, Microsoft Azure and Google Cloud are used for cloud infrastructure.

This creates a severe strategic vulnerability. When nations upload raw intelligence to third-party clouds, they risk violating national security laws and losing control over their data. Centralisation concentrates critical national security capabilities into the hands of a few private corporations, leaving militaries vulnerable to vendor lock-in, services being cut off during diplomatic disputes and losing true sovereignty over its AI.

This principle sits at the heart of FLock’s architecture: we were founded on the principle that AI should be brought to the data; data shouldn’t be brought to the AI if sovereign AI, data privacy and compliance are the aim.

Distributed inference for real-time AI is more resilient, fast and secure

Major militaries such as the US and UK have used edge computing in part since the 1990s. These are called tactical clouds (or combat clouds): a portable, decentralised computer network that brings localised data processing directly to often disconnected combat zones. It allows sharing real-time information locally without relying on vulnerable, distant home-base servers. When connection to a HQ is lost, it processes sensor data and runs AI tools locally.

The global military edge computing market is valued at approximately $3.66 billion in 2026 and is projected to scale to over $10 billion by the mid-2030s. Driven by demands for real-time battlefield AI, low-latency processing, and secure tactical communications, the market grows at a compound annual growth rate of roughly 13.9%.

Edge AI is the above concept applied to AI. It is when a standalone, self-contained model is placed on a physical endpoint (like a drone, tank, reinforced laptop, smart camera or phone) to make immediate, localised predictions. The benefits of edge AI include less power use compared to data centres, reduced bandwidth, more privacy, stronger security, easier scalability and lower latency.

Distributed inference takes it a step further. Where edge AI runs algorithms directly on edge devices, distributed AI uses multiple interconnected systems: central servers, edge devices, and others. It breaks down a single large AI workload or splits a model across multiple nodes to process pieces of the problem cooperatively.

The advantages of distributed inference

No single point of failure

Thanks to its mesh network, distributed inference is highly resilient and secure. Whether a drone gets shot down, an opponent obliterates the infrastructure or launches an electronic attack like noise jamming, or a cyberattack takes place, the system can redistribute workloads across available nodes. Squads and autonomous swarms keep full AI capabilities even when GPS and satellite communications stop working.

Greater speed

Running distributed inference allows edge hardware (like an onboard drone camera or armoured vehicle display) to process the raw data locally in real time. The low latency means intelligence is available quickly on the battlefield, independent of faraway platforms. It cuts out the long trip to the cloud, giving frontline operators instant predictions when seconds count most.

Sovereignty and privacy

Distributed inference keeps sensitive data strictly where it belongs, on local, sovereign hardware. Intelligence is processed locally at the edge, allowing forces to run AI without risking data exposure or giving up physical control of their most critical assets.

Federated learning

Federated learning (FL) is a decentralised way of training a model, while distributed inference involves running a model that has already been trained. In FL, multiple edge devices collaboratively train a single global model using their own local data without sharing that raw data. For allied coalitions like NATO, federated learning allows cross-border collaboration while only sharing the model updates.

FLock has demonstrated this method in healthcare: we proposed a framework using data sets from Europe, North America and Asia, and tested it on glucose management, to tackle how the issues of data privacy are holding back AI development. This can be applied to several sectors that can benefit from cross-border collaboration.

FLock’s approach

FLock enables military branches, allied forces and defence contractors to co-develop powerful AI models directly on local hardware. Raw intelligence and classified feeds never leave sovereign borders. Defence forces gain the predictive capabilities they need for successful missions and counter-attacks. By combining federated learning with blockchain-based verification, FLock ensures 100% data sovereignty, model ownership and operational control

FLock is an AI research and infrastructure company pioneering enterprise-grade federated learning and distributed AI solutions that prioritise data privacy. Its decentralised federated learning architecture and production-ready platforms (AI Arena, FL Alliance, and FLock API Platform) enable organisations to train and deploy their own custom AI models on local hardware while maintaining full data privacy, model ownership, and regulatory alignment by design.

FLock effectively combines FL and blockchain-based verification for a 37% improvement in model accuracy, a 44% reduction in total cost ownership, a reduced risk of data breaches or model poisoning attacks and a 63% shorter deployment time.

FLock was recently spotlighted by the World Economic Forum (WEF)’s MINDS programme for its privacy-preserving AI training for the NHS, specifically in eye disease detection and diabetes management that keep patient data secure and maintain 100% data sovereignty.

Follow FLock on LinkedIn and X and email the team at hello@flock.io.