{"id":13658,"date":"2026-07-14T20:35:35","date_gmt":"2026-07-14T20:35:35","guid":{"rendered":"https:\/\/sawahsolutions.com\/range\/how-a-former-marine-is-rewriting-the-future-of-battlefield-ai\/"},"modified":"2026-07-14T20:35:35","modified_gmt":"2026-07-14T20:35:35","slug":"how-a-former-marine-is-rewriting-the-future-of-battlefield-ai","status":"publish","type":"post","link":"https:\/\/sawahsolutions.com\/range\/how-a-former-marine-is-rewriting-the-future-of-battlefield-ai\/","title":{"rendered":"How a former Marine is rewriting the future of battlefield AI"},"content":{"rendered":"
\n

One stubborn former Marine is changing how the Pentagon moves data to farflung troops\u2014a change intended to enable them to use advanced AI tools, and one that just might reshape how lightweight models are developed.<\/p>\n

Some of these new workflows have already been used during the U.S. war on Iran.<\/p>\n

\u201cIf you look at what’s happened with Operation Epic Fury, in particular, we were able to incorporate dozens of new feeds in real time that allow us to not only serve up that data in the right format, right structure, and everything else for those, for those applications to leverage, but also get data at the speed of conflict,\u201d Cameron Stanley, the Pentagon\u2019s Chief Digital and Artificial Intelligence Officer, said at the AWS Summit in Washington, D.C., last week.<\/p>\n

Stanley was talking specifically about the War Data Platform, a branch of the Pentagon\u2019s Advana data system being developed to digest thousands of data feeds. The WDP is a key part of defense leaders\u2019 plan to use AI to help commanders make decisions faster.\u00a0<\/p>\n

But another key to the plan is developing ways to get those tools and that data to troops in and around the battlefield, where connectivity is minimal or absent. Nowhere is the challenge more acute than in Indo-Pacom Command, with its vast distances and powerful potential adversary.<\/p>\n

That\u2019s the main challenge picked up by Bala Selvam, an entrepreneur and a former Marine, when he stepped into the job of chief technical officer for Special Operations Command Pacific in early 2025.\u00a0<\/p>\n

A race judged in milliseconds<\/strong><\/p>\n

When Selvam took the job, he found that the Defense Department\u2019s data workflows were largely being developed in places with excellent networks and plenty of computing resources. That had produced a culture of shipping huge amounts of data with little concern for latency.<\/p>\n

In the continental United States, or even European or Central Command, \u201cIt didn’t matter because you had all the compute you needed,\u201d Selvam said at the AWS event.<\/p>\n

But that wouldn\u2019t work at INDOPACOM.<\/p>\n

The distance from the command\u2019s Hawaii headquarters to its second-largest data center is 9,000 miles\u2014too far for efficient data transfer, even with low-earth-orbit satellites, Selvam said.\u00a0<\/p>\n

Analysis and bureaucracy might stretch the data\u2019s journey by yet more thousands of miles before it gets to warfighters at the edge.<\/p>\n

\u201cIf you’re SOCOM, it goes back to Tampa, or if you’re any other service, back over here\u201d\u2014that is, Washington, D.C..<\/p>\n

All of that was limiting how quickly operators could merge top-secret data and open-source data, analyze it, send it to troops, and share it with partners. The deadline is a matter of milliseconds, with China working 83 percent as quickly. So staying ahead of that number is critical to moving information faster than the adversary. Losing that race just might mean losing a conflict.<\/p>\n

The problem only got harder as more data streams emerged. And it also meant that troops in or near conflicts would be unable to use advanced AI tools that require heavy use of enterprise compute resources.<\/p>\n

The approach<\/strong><\/p>\n

Rather than try to solve each problem he discovered, Selvam decided to figure out just what everyone actually needed. It\u2019s the kind of analysis that a big U.S. consultancy might require millions of dollars to complete.\u00a0<\/p>\n

Selvam started by going to INDOPACOM senior leadership. \u201c\u2018Write down everything that you expect everyone to do,\u2019\u201d he told them. “I think a list is going to be in the hundreds. It was actually like 10 things.\u201d<\/p>\n

Selvam took this list of \u201ccommandments\u201d and worked his way down chains of command, asking everyone to lay out what they were using and how. He found many commanders buying software and tech tools that didn\u2019t match up with what they needed to accomplish.<\/p>\n

One category, in particular, stood out.<\/p>\n

\u201cThey kept saying, \u2018the AI.\u2019 I got so mad that I said, \u2018If you do it, I might, like, throw a wild dog in your house,\u201d he said. \u201cIf you’re gonna say AI, tell me what exactly you’re doing\u2026If you can\u2019t tell me that, you’re not using AI.\u201d<\/p>\n

To force commanders to define their tasks, Selvam began to physically deprive them of their technology.<\/p>\n

\u201cI actually almost got, like, fired. I took everyone’s computer away, and then I hid it. Then I put whiteboards in everyone’s rooms with their actual task. I forced everyone to walk into the [Joint Operations Center] to put in their inputs. When they said the words, \u2018I wish I had this technology; I did this virtually,\u2019 I gave them back the computer with only that thing. It was like an Excel sheet and a PowerPoint slide. And then when they said, \u2018Man, I wish I could do this faster,\u2019 then I gave them back Palantir and Databricks [a cloud-based data analysis and AI platform] and everything else.\u201d<\/p>\n

For Selvam, this wasn\u2019t about instilling tech discipline; it was implementing business best practices. The process yielded a record of what everyone actually needed to do their jobs.\u00a0<\/p>\n

\u201cIf you want to know what a good tech company looks like, look at their documentation. Same concept here,\u201d he said. \u201cSo, for every bit of our work, we will write processes, procedure business rules, and we will build based on those process procedures or business rules.\u201d<\/p>\n

A key meeting<\/strong><\/p>\n

Armed with a much better understanding of what people actually needed to get done, Selvam turned his attention to the problem of getting data where it needed to go. For that, he needed help.<\/p>\n

Around January, he traveled to the Crystal City, Virginia, offices of AWS, which makes billions of dollars selling the Pentagon the sort of enterprise cloud capability that won\u2019t be available at the edge of combat. There he met Joshua Hobgood, who leads AWS\u2019 efforts to sell AI-related services to the world\u2019s militaries.<\/p>\n

\u201cI want to solve for AI at the edge,\u201d Selvam told Hobgood, who described the meeting to the AWS Summit audience. Hobgood answered that it was a common problem, to which Selvam replied, \u201cYeah. But we’re going to do it right.\u201d<\/p>\n

Over the next two hours, they hammered out a new plan. First, they would build smaller cloud compute clusters, or nodes, that would require less power and infrastructure, and, hopefully, be harder to target in a conflict. These would be designed to support battalions\u2014a few thousand troops\u2014and would replace larger nodes meant to cover entire divisions of 20,000 or more, Hobgood said.<\/p>\n

He said that making this work will require commanders to follow Selvam\u2019s procedures governing which tools people use and for what purpose. When SOF commanders \u201cgive out tasks, they will specifically say, \u2018Humans, you will do these [six] out of these 10 tasks\u2026These other four, here’s how you’re gonna let the [AI] agents take over that.\u2019\u201d<\/p>\n

The next step was to make sure those agents, or whatever other AI tools were running at the edge, were feeding data back to the cloud.<\/p>\n

\u201cWe based the solution on the idea that\u2026up to the point of a conflict scenario, there will be connectivity. So we wanted to create a system that could continuously learn,\u201d Hobgood said. \u201cThe core assumption to that was that the majority of your AI inference, your data analysis, would happen in the cloud, in the AWS cloud.\u201d\u00a0<\/p>\n

But as they worked to flesh out the ideas over the next few months, Selvam and Hobgood realized that AI tools could be built to sip power and compute\u2014and still deliver results to troops in the field.<\/p>\n

\u201cWe’re at a point where we can use smaller and smaller models\u2026to get, actually get, the best results,\u201d he said.<\/p>\n

That\u2019s going to be vital as the Pentagon works to integrate more and more information into the War Data Platform. In May, the Chief Digital and Artificial Intelligence Office said the WDP was drawing upon some 4,000 data sources from more than 55 organizations.<\/p>\n

\u201cWhat we’re trying to do is really create that single pane of our single repository of truth for a lot of data that was disaggregated and come together into a single platform that allows for very quick connection of data sources to different types of applications,\u201d said Stanley, who leads the CDAO.<\/p>\n

It\u2019s unclear how Selvam\u2019s proposed brigade-sized compute clusters will relate to the platform.<\/p>\n

But, said Stanley, \u201cThat’s going to become a program of record here pretty soon.\u201d<\/p>\n

Selvam said the effort was supremely worth it.<\/p>\n

\u201cI’m not gonna lie to you; my entire 18 months has been like, \u2018If I don’t solve the [cloud-defined storage] problem\u2026my life would just be meaningless.\u2019 But if I did solve it, I’m going to put it on my tombstone\u2026It’s going to say, \u2018I solved the [Department of War] CDS problem.\u2019 That’s what it’s going to say, I swear to God.\u201d<\/p>\n

Nextgov\u2019s Alexandra Kelly contributed to this story. <\/em>\n\n<\/svg><\/p>\n<\/div>\n