Why the Pentagon Is Spending $1 Billion on Bilingual AI for Combat Translation

Military instructors seldom discuss one type of frustration in public: the waiting. Imagine spending eleven months creating a training program for officers from twelve Latin American countries, perfecting the curriculum, and then giving it to a translation team that promises to finish it within a year.

The course has already been updated by the time the translation is delivered. The entire process is restarted. At WHINSEC, the Army’s Western Hemisphere Institute for Security Cooperation at Fort Benning, that was exactly how things operated for many years. Then LILT appeared.

The California-based AI company recently received a flexible contract from the Pentagon’s Chief Digital and AI Office to expand an AI translation platform across military operations worldwide. This type of contract allows for rapid scaling without the typical procurement delays. The announcement was made discreetly, just before it was made public on a Tuesday. This could be due to either standard timing or the kind of purposeful understatement that major programs occasionally favor. There was no disclosure of the contract value. It’s still unclear.

What transpired at WHINSEC when they actually used it is evident. In just a few weeks, the entire eleven-month Command and General Staff Officer Course—which is taught in Spanish to students from partner countries throughout the Western Hemisphere—was translated. Not months. weeks. The institute’s commandant, Army Col. Eldridge Singleton, said it was “catapulted into the future.” Officers typically don’t use this type of language unless they truly mean it.

The timing is important. This push by the Pentagon is not taking place in a vacuum. Secretary Hegseth nearly immediately followed the Department’s January 2026 release of an AI Acceleration Strategy with a speech denouncing what he referred to as the “linear” model, the outdated approach in which a technology advances from the lab to the battlefield over years or even decades. Wartime speed is the new benchmark, at least in theory. Honest observers are still debating whether the bureaucracy can really move that quickly.

Beyond the technology, the LILT contract is intriguing because it shows where military communication has been lacking. Historically, translation has been viewed as a logistical issue that can be resolved by employing contractors, waiting, and hoping the outcome is precise enough to be significant. The issue is that military terminology behaves differently from everyday language. During a live exercise on a coalition radio channel, a standard phrase from a training manual can mean something quite different.

That is completely missed by generic translation tools. The platform is trained on military-domain-specific vocabulary, according to LILT’s pitch, which was verified by the Defense Innovation Unit prior to this contract being awarded. The WHINSEC results indicate that there is a substantial difference between this and previous approaches, though it is still possible that edge cases will break it in ways that no one has yet predicted.

The larger picture is striking. Eight significant tech firms, including OpenAI, Google, Microsoft, NVIDIA, Amazon Web Services, Oracle, SpaceX, and Reflection, have now partnered with the Defense Department to use their models on classified networks. Tens of millions of prompts have already been generated by more than 1.3 million active users of the department’s internal AI platform, GenAI.mil. It’s difficult to ignore the fact that this is no longer a think-tank idea or a pilot program. The equipment is already in motion.

It is arguable whether combat translation is the military’s most pressing AI issue. However, it’s an exceptionally clean use case because the stakes are real, the resistance is low, and the before and after are quantifiable. There is no denying that the system of waiting a full year for a translation was effective. Now, the question is what happens when technology advances more quickly than the doctrine, training, or confidence needed to rely on it in a real emergency. That response is still pending.