“If India were to be attacked, we’ll be there to help,” President Donald Trump told Prime Minister Narendra Modi on the sidelines of the G7 meeting in June. The comment was offered warmly, in the spirit of a friendship the president said “cannot be closer.” But whether the two countries could fight together effectively is a question that’s getting harder to answer.

As the nature of warfare evolves with unprecedented U.S. progress in artificial intelligence (AI), that answer is increasingly about operational reality rather than political will. On the one factor that increasingly decides modern wars, the two are drifting apart fast.

Interoperability has long been a challenge in the U.S.-India defense relationship, and the AI gap will only widen it. However, the gap is not inevitable, and addressing it needs to become a priority now. At the same time, India’s military AI architecture is still being built, and the country is making progress in developing its own AI systems.

To ensure future interoperability, the United States and India should make AI interoperability an explicit part of their broader defense agenda, especially by stress-testing through existing bilateral and multilateral military exercises.

The past year showed how far apart the two countries now sit within the AI kill chain. In May 2025, India carried out Operation Sindoor, the first cross-border operation it has described as “AI-enabled.” Indian forces used AI to analyze intelligence and sharpen targeting against nine sites across the border with Pakistan.

>>> AI Can Play a Limited but Crucial Role With Nuclear Weapons

Roughly nine months later, the United States launched Operation Epic Fury against Iran and struck more than 13,000 targets in 38 days, roughly 1,000 of them in the first 24 hours. Both were called AI-enabled operations, but the tempo, scale, and inherent capability differed vastly.

The United States began building this capability in 2017, when the volume of drone footage from the counter-ISIS campaign became too difficult to be reviewed by human analysts alone. At the time, U.S. Central Command was collecting 700,000 hours (80 years) of full-motion video annually. So, the Pentagon started building Project Maven to use computer vision to flag objects of interest.

Over time, it grew into the Maven Smart System (MSS), fusing diverse feeds—video, satellite imagery, signals intercepts, radar—into a single targeting picture. They later layered large language models on top so commanders could query and interact with the system in plain language. The key was a system that evolved organically through civil-military fusion, where commercial firms supplied the technological core and battlefield data, and targeting continuously improved the model and system as a whole. During Epic Fury, the Pentagon used MSS and ran through roughly 20 billion AI tokens a day at peak.

India has nothing comparable, and the reason is timing as much as capability. Its systems are less battle-tested; they are first-generation and are only now being fielded. Project Sanjay, the battlefield surveillance system, began rolling out to formations in phases through 2025. The electronic-intelligence system, or ECAS, used in Sindoor was still being modified mid-operation.

India’s homegrown military-grade large language model for cyber-vulnerability discovery is only now out for expression of interest through DRDO’s technology development fund, meaning the model is not in active service and is only now being procured. These are real AI tools, and they identify, prioritize, and fuse, accelerating the decisions a commander makes.

Maven keeps a human in the loop too. Still, the difference is what the machine does before the human acts: it has spent nine years in continuous combat use learning to generate the targeting itself, at a scale and tempo India’s just-fielded systems are years from matching.

The operational gap also shows up in funding. India’s dedicated defense AI agency was set up with about $12 million spread across five years, and Indian military AI spending overall has been estimated at roughly $50 million a year. In sharp contrast, the Pentagon requested $13.4 billion for AI-enabled systems in a single fiscal year.

The gap in military AI use and development is a partnership problem. For two decades, interoperability between the United States and India has meant platforms and plumbing. The foundational agreements signed by the two countries between 2002 and 2020—GSOMIA, LEMOA, COMCASA, BECA and ISA—were built to let the two militaries share secure communications, cooperate on logistics, and share geospatial intelligence.

That work is not even finished. India’s P-8I maritime aircraft were delivered without the Link-16 secure data links other partner navies receive, and a CNA study found that ad hoc workarounds have closed interoperability gaps only temporarily.

The AI gap complicates interoperability in an entirely new way. Without equally capable systems working in tandem, the two militaries cannot share a live targeting picture, hand off targets, or reconcile what two AI systems each conclude about the same area of interest. NATO is already wrestling with this among allies, with a senior alliance intelligence officer warning in May 2026 that AI-enabled systems risk handing commanders conflicting machine-generated assessments, and that the alliance must agree on common standards before the technology outruns the frameworks. The same principle also applies to the United States and India, which have not yet started discussing this AI interoperability problem, at least publicly.

It’s not as simple as India buying its way onto the American kill chain, the way it bought P-8Is and Apache helicopters and the various other U.S. platforms that now power the Indian military. India’s focus on AI sovereignty emphasizes sovereign ownership of the stack, and Indian analysts argue that defense datasets such as satellite imagery and electronic intelligence are too sensitive to process through foreign platforms.

Similarly, Washington, for its part, will not hand over its best AI technology. In June, the U.S. government suspended access to Anthropic’s most capable Mythos and Fable models for all foreign nationals, inside or outside the country, even though the models are commercial rather than military.

>>> What Recent Wars Mean for U.S.-India Defense Cooperation

Complicating matters, Washington and New Delhi share an adversary that is working the same problem from the other end. China is building AI of its own, and CSET estimates its military AI spending $1.6 billion a year while the People’s Liberation Army (PLA) is building AI decision-support systems for targeting.

China is not just building its own AI; it has also begun to publish how it intends to defeat others’ AI kill chains through counter-AI or anti-AI. In May 2025, the PLA Daily set out to attack an adversary’s AI at three levels: its data, its algorithms, and its computing power, through methods such as polluting data to corrupt a model and logical deception. Chinese forces are already rehearsing it with decoys that lure AI-assisted targeting into firing at phantoms.

The foundation for interoperability that the United States and India have built through platforms and secure networks is now being tested by a new variable: AI. The platform era of the partnership—built on foreign military sales and 20th-century foundational agreements—took two decades to construct and is still a work in progress. A new layer is now reshaping warfare, and the two sides cannot yet operate together.

But the gap can be breached, and the time to do this is now, while India’s system is still being built and standards can be designed in rather than bolted on later. Since the United States is India’s largest military exercise partner, the exercises the two already run every year are a natural place to test what works and what breaks.

Interoperability has always been a major conversation in the U.S.-India defense partnership. The future conversations around U.S.-India defense cooperation and interoperability need to factor in the constraints that AI adds—and build around it.

This piece originally appeared in The National Interest.