As the U.S. military increasingly relies on artificial intelligence (AI) to support operations, new challenges are emerging in maintaining the necessary computing power and communication in conflict zones like the Middle East. Drones, sensors, and AI-assisted targeting systems all depend on reliable data centers and connectivity. However, the destruction of a single data center can severely disrupt military capabilities. These issues go beyond just physical hardware, as questions arise about how AI systems can adapt in the absence of communication and how they can make decisions without real-time battlefield data. Traditional data centers in the Middle East are proving to be vulnerable targets, as they can be easily struck by low-cost missiles. In response, companies like Palantir have started using shipping containers filled with high-performance Nvidia hardware as a more flexible and mobile alternative to traditional data centers. These containers provide decentralized computing power at the frontlines. Similarly, Scaleout, a company focused on edge AI infrastructure, has partnered with NATO and defense contractors like BAE Systems to develop systems that can train AI models using distributed, secure data without relying on a central server. According to Andreas Hellander, CEO and co-founder of Scaleout, the goal is not to ensure an unbreakable connection but to create systems that can continue functioning even if part of the network is lost. Their design allows a disconnected node to continue operating using its last approved model, storing data until communication is restored. This approach allows multiple aggregation points to be used, so if one fails, others can take over, ensuring a gradual decline in performance rather than a total system failure. The choice between centralized and distributed AI systems has important trade-offs. Centralized systems offer a unified view of operations and strong computing power but are vulnerable because losing a single point of failure can significantly weaken the system. Distributed systems reduce latency and allow continued local operations when disconnected, but they risk inconsistencies, as decisions may be based on outdated or incomplete information. AI drones and sensors must adapt accordingly, with connected systems able to share data and receive new instructions, while disconnected ones follow pre-set safety protocols. Data collected by drones and sensors helps AI systems adapt to new threats and conditions. Active-learning software identifies the most useful data, such as unclear images or unfamiliar objects, which are then labeled by humans. The AI model is refined locally and tested, with updates shared across the network using federated learning, which reduces the need to move large amounts of data. However, deploying these systems in the field presents challenges such as power consumption, heat management, weight, and environmental conditions. While commercial computing solutions are available, logistics and maintenance remain significant concerns. Companies providing AI and computing services for military use, such as AWS data centers in the Middle East and drone manufacturers in the UK, face the risk of being targeted. Scaleout’s system is designed to avoid this risk by running within the customer’s own infrastructure, without needing a continuous connection to external cloud systems. Looking ahead, improved satellite connectivity from low-Earth-orbit satellites could help, but it's unlikely to connect every sensor directly to space. A more practical approach combines local mesh networks, intermittent synchronization, and selective use of satellite links. While extended operations without a connection are possible, they are often limited by factors like battery life and maintenance. The distinction between operating without a connection and operating without human oversight is crucial, as the latter involves legal and ethical considerations. Future development will focus on ensuring verifiable control, tracking which AI models are used, their origins, and who approved them.