A century ago, innovation in national defense wasn’t defined by data, but by dominance.
After World War I, the aircraft carrier became one of the clearest examples of military might, a floating platform that accelerated the Navy’s ability to attack land targets and ships.
In less than three years, the American carrier fleet grew from seven to more than 100, an impressive growth trajectory. Still, it was the pilots, planes and the engineers behind them that turned the tide.
The pace of growth must have seemed staggering at the time, but it pales next to the acceleration of artificial intelligence. AI capabilities vary just as widely in design, purpose and maturity. The Department of Defense’s rapid adoption of AI – arguably carrier innovation multiplied tenfold – has left agencies grappling with which tools to invest in and how to optimize data centers to meet growing mission demands.
Data centers increasingly serve as the backbone of that mission, providing the computing, storage and connectivity required to train, secure and operationalize AI at scale. For the military, their value extends beyond processing power: they must support workloads that can ultimately deliver trusted information and capabilities wherever missions are conducted.
Choosing the right tools
The AI landscape is so new and fast-moving that many defense agencies risk acquiring emerging tools without a clear sense of how to use them, often gravitating toward a single solution for the sake of simplicity.
That makes defining the right tool for the right job critical. A garden spade is ideal for digging a flower bed, but no one would use it to clear snow from a driveway. The same principle applies to AI, just as it applied when naval leaders determined which aircraft belonged on which carriers. Relying on a single vendor can also limit perspective and capability.
The opposite risk is over-provisioning: buying every new capability without a defined purpose. More tools don’t necessarily lead to better outcomes; often they add complexity, cost and delay. The agencies that succeed will resist the urge to overbuild and focus instead on solutions designed to solve specific mission problems.
Commanders today also have new options, including edge powered solutions that don’t rely on robust cloud architectures. The future of functional AI will depend on speed, scale and stronger security: train models in the data center, then deploy them to the edge.
Success starts with identifying the problem to solve. That problem should drive the AI workloads required and the models best suited to each task, whether computer vision for autonomous systems, natural language processing for translation or multiple models working together to enhance situational awareness.
Before launching any AI project, leaders should ask, what problems are we solving? What medium are we working with: text, voice or video? How autonomous should the system be? Are the data sources trusted, and is there a secure way to manage data at rest, in motion and in use?
Request for proposals should reflect that same discipline, focusing not on “AI” broadly but on the specific capabilities needed to solve the problem at hand. The goal isn’t to deploy as many tools as possible, but to select a limited set of solutions aligned to mission needs, then optimize the infrastructure to support them over time, built on trusted data, clear governance and established retraining practices.
Data governance and model strategy
Once AI systems are in place, the focus shifts to how they’re governed and sustained. The models agencies select determine the capabilities they can deliver and the infrastructure required to support them. A navigation recommender system and an autonomous drone, for example, have fundamentally different requirements that should guide both system design and data strategy.
Supporting these capabilities requires well-defined governance, including compatibility rules for how data flows in and out of AI environments, who owns it, and how it’s managed over time. Equally critical is a plan to continuously refresh and retrain models, as even the most promising AI initiatives quickly become obsolete without it. Agencies must build environments that can expand computing capacity as needed, support diverse workloads, and enable continuous data movement, retraining and redeployment without disruption.
That adaptability must extend to security. Agencies need a true zero trust, defense-everywhere posture, in which no user, device or workload is inherently trusted, and access is continuously verified, so no application, endpoint or data flow becomes a weak link.
In the end, mission success depends on bringing the right tools for the right jobs. A complex tool like a backhoe can be powerful, just as cloud-based AI supports highly demanding workloads. But sometimes a simple garden spade, like an edge AI solution built for a targeted use case, is all that’s needed.
Data centers must stay flexible enough to support the diverse demands of AI models and the environments in which they operate. In this way, the data center becomes the aircraft carrier of AI: a platform that supports the right capabilities, adapts to mission demands and delivers impact where it’s needed most.
