ABOUT THE AUTHOR: Patrick McSpadden is a retired U.S. Air Force lieutenant colonel and former intelligence officer with more than 21 years of service, including multiple operational deployments. He writes on defense policy, military technology and national security. The views expressed are his own and do not reflect the official position of the Department of Defense or the U.S. government.

Artificial intelligence is no longer just another defense buzzword. It is becoming a source of real military power. The Department of Defense clearly recognizes that at the strategic level. The 2025 National Security Strategy made it explicit that AI dominance is central to U.S. competitiveness across all instruments of power, and the upcoming National Defense Strategy will almost certainly reinforce that view.

The question is not whether the Pentagon understands the importance of AI. It does. The question is whether it is moving fast and deliberately enough to turn that understanding into something operators actually trust and use in combat.

To truly win the AI race across all instruments of U.S. power, the Pentagon will have to move from understanding to deploying with scale, coordination and focus in the field as soon as possible.

To be fair, there has been real progress. The creation of the Chief Digital and Artificial Intelligence Office brought much-needed coordination to what was once a scattered and uneven effort. Task Force Lima, in particular, deserves credit for taking a clear-eyed look at generative AI.

But translating potential into power requires more than task forces and strategy documents. It requires hard choices about infrastructure, doctrine and people.

If the next National Defense Strategy is serious about AI as a pillar of deterrence and warfighting advantage, three priorities should stand out.

First, the Pentagon has to start treating compute like the strategic resource it is.

For all the attention paid to models and software, the biggest constraint on operational AI today is compute. Access to high-end infrastructure is uneven across the department. GPU availability varies widely. No one has a clean, department-wide picture of who needs what and when.

That makes it difficult to move beyond limited pilots and one-off demonstrations. The solution is not more experimentation. It is organization. A Unified AI Compute Command would give the department a way to manage high-performance compute the way it manages other scarce resources.

If AI systems are going to be relied on in real missions, then operators and analysts need confidence that the infrastructure will be there when it matters. Without that confidence, trust erodes quickly.

Second, AI has to be integrated into doctrine and training in a way that operators actually accept.

The department has demonstrated that AI tools can help. Planning assistants speed up staff work. Intelligence tools sift through data faster than any human team could.

Predictive maintenance systems save time and money. None of that matters if operators are not comfortable using these tools under pressure.

The Pentagon needs to codify how AI fits into decision-making, fires, logistics and maneuver. It should establish AI-focused red teams that stress systems under realistic conditions, test them against adversary interference, and surface failure modes before they show up in combat.

Most importantly, AI must feel like it belongs to the operator, not imposed by distant headquarters. If people do not trust it, they will work around it. And unused technology does not deliver military advantage.

Third, AI export controls should be used deliberately as a tool of strategy.

Advanced chips and compute infrastructure are now geopolitical assets. Who gets access matters, but so does sustaining a strong, innovative U.S. industrial base. The United States should treat AI export policy as part of alliance management and economic leadership, not just a compliance checklist.

That means working closely with State and Commerce departments to enable timely, predictable access for trusted allies that share security standards, cyber protections, and interoperability goals. Healthy, rules-based export markets also allow U.S. firms to scale production, spread costs, and reinvest in the next generation of AI and semiconductor technology that underpins long-term military advantage.

There is also a case for allowing sales of older-generation technologies: letting companies move legacy inventory keeps production lines healthy, supports the workforce, and turns yesterday’s technology into funding for tomorrow’s breakthroughs, while the U.S. retains its lead at the cutting edge. Reports that the Trump administration will allow sales of 2024-era chips to China show the administration is taking serious steps toward blunting Chinese tech firms rise and independence.

Handled properly, AI access becomes a way to reinforce our strategic advantage over China, shore up alliances, and sustain U.S. technological leadership.

The upcoming National Defense Strategy will almost certainly frame AI as central to future conflict. That framing is correct. But strategy only matters if it changes how institutions operate and how people fight.

Progress is happening. Offices like the CDAO have laid important groundwork. Task Force Lima helped set sensible boundaries and expectations. The next step is follow-through.

AI will not replace people in war. But it will reshape how power is generated, how decisions are made, how conflicts are fought, and which nations win when it comes to national security and prosperity. The Pentagon has an opportunity to get this right, but only if it moves from enthusiasm to execution.