Listen to the article
The cloud connectivity that makes big AI models work is also the reason they won’t work in war. As the Marine Corps explores ways to bring AI tools to the battlefield, it is looking at one software company’s proposal to keep troops computing when broader access gets cut off.
Ditto will announce Monday that the Marines’ Project Dynamis will evaluate their technology for turning radios, cell phones, even drones, into a local network that can keep data flowing and AI tools running locally when cloud access disappears.
“Ditto’s position is that we do not leverage a server-client model,” Eric Hanft, Ditto’s senior vice president for public sector and a former Army infantry officer, said in an interview. “If you continue to architect your data flows around that, you never remove this fundamental dependency that, if two edge nodes need to communicate and have to go to and from a server—and that server is not available—there’s no communication.”
A digital and data communication failure is a nightmare scenario that the United States has largely been able to avoid in past operations. But over the past two decades, those operations have increasingly focused on the Middle East, where the United States did not face adversaries that were capable of disrupting critical communications infrastructure.
Project Dynamis is the Marine Corps’ addition to the Pentagon’s joint all-domain command and control, or JADC2, efforts. Increasingly, such efforts focus on binding troops, satellites, drones, and more in a data web to enable much faster sharing of information and, as a consequence, faster and more successful operations. According to Marine Corps documents about these efforts, they test solutions in situations where communication is “degraded or denied.” But details beyond that, such as whether cloud connectivity itself was denied during experiments or tests, are hard to come by.
Marine Corps leaders are more likely to discuss the need to move to the cloud.
“Implementing cloud on our networks is the critical requirement for CJADC2. Without cloud, JADC2 just is not going to come to fruition,” Col. Jason Quinter, who led Project Dynamis as commanding officer of Marine Air Control Group 38, said in 2024.
As sensors—flying, orbiting, marching—proliferate across the battlefield, and as the military charges ahead with ambitious AI plans to better use that sensed data in combat, the growing reliance on cloud computing makes sense.
But Ukraine’s experience against Russian electronic warfare suggests that cloud architectures are exactly the sort of target high-tech adversaries will come after first.
That is already becoming a problem for the military as it seeks to put large language models into combat settings. One of Anthropic’s arguments against allowing the Pentagon to use its tools in battle is that if data is wrong or is missing, transformer models may hallucinate the missing pieces. It’s a low-stakes problem when trying to write a term paper, a high-stakes issue in war.
OpenAI appears to have similar concerns, announcing in March that its own military contracts “would not permit powering fully autonomous weapons, as this would require edge deployment.”
The Pentagon has responded to such concerns by looking for models less prone to errors on the edge. And the Marines have added a goal to Dynamis: testing assured communication technologies.That’s where Ditto comes in. Hanft argues that relying on the cloud during a conflict is “going to become a problem” no matter which AI company is making the tools. Local networks can help ensure important data aren’t lost when the cloud goes offline, but the military’s existing peer-to-peer tools are limited and difficult to use, he said.
So Ditto offers a way to use “whatever transports the customer brings—in the case of a commercial environment, that could be the radios that are already built into your phone—and deal with Bluetooth and Wi-Fi. We can create a peer-to-peer data mesh with no new hardware, just, you know, your consumer phones.”
That aligns with the Pentagon’s efforts to prioritize commercially available or dual-use technology rather than military-specific tech.
The company was founded eight years ago to connect civilian and commercial devices when their cloud connectivity vanishes, as when, for instance, a business closes its doors for the day because the credit-card machine is down, he offered by way of example.
Now it’s trying to apply its methods to the battlefield. Hanft said that the company has already done experiments with U.S. special operations elements and Nordic military forces. He didn’t say whether the company had worked with Ukraine directly but acknowledged that they have “worked in the past with organizations that are operating within those boundaries.”
“One of the greatest strengths of Project Dynamis is in its methodology. For the last several months we’ve been in a continuous cycle of designing, testing, and refining warfighting solutions—Marines and industry engineers side-by-side—so we can find best-of-breed solutions and get them into the hands of our Marines,” said Col. Arlon Smith, the current director of Project Dynamis, in an early version of Ditto’s press release.
Read the full article here

6 Comments
Interesting update on Marines eye cloudless networks to keep AI tools running when the cloud goes down. Looking forward to seeing how this develops.
Great insights on Defense. Thanks for sharing!
Solid analysis. Will be watching this space.
Good point. Watching closely.
This is very helpful information. Appreciate the detailed analysis.
I’ve been following this closely. Good to see the latest updates.