The head of a fast-growing tech firm has noticed that the promise of rapid advancements in space defense is bringing around some top public decision-makers to deep truths about the free market system.
“I believe that the Department of War and the U.S. government — governments across the world, actually — are finally understanding that there’s more private capital out there than there is taxpayer money,” Pete Cannito, chairman and CEO of the Jacksonville, Florida, space infrastructure company Redwire Corp., told a space defense conference in December 2025. “And they want to tap into that.”
Nowhere is that private capital snowballing like it is in artificial intelligence and machine learning (AI/ML), where just seven major tech companies — including some whose systems form the foundations of AI/ML — were expected to spend more than $300 billion on the technology in 2025. That’s equal to the value of Coca-Cola, United Health or Chevron.
Tapping into the advancements fueling this trend has become a matter of policy for the United States government. “America must have the most powerful AI systems in the world,” according to America’s AI Action Plan, issued by the White House in July 2025. “Achieving these goals requires the Federal government to create the conditions where private-sector-led innovation can flourish.”
Air Force Secretary Troy E. Meink, speaking in December 2025 at the Space Force Association’s Spacepower Conference in Orlando, Florida, said, “We need this skill set in everything we do or we’re not going to be successful.” AI/ML skills are emerging across a host of Pentagon initiatives, many of them within the U.S. Space Force — the military’s first digital service and the one charged with securing the nation’s interests in space.
“In terms of AI, industry is definitely driving the path forward,” U.S. Space Force Capt. Drake Williams, who runs an AI prototyping program as an acquisitions manager with Space Systems Command (SSC), told Apogee.
Among the examples: AI/ML capabilities play a growing part in the Space Force’s Industry Days, where acquisitions personnel bring in companies to seek help answering identified needs, and in Reverse Industry Days, where companies showcase their developing technology in hopes it might be of use. The service’s innovation arm, SpaceWERX, and the Pentagon’s Defense Innovation Unit conduct similar outreach to the space industry.
Industry also partners with Space Force and academia at TAP labs — for tools, application and processing — such as the Space Domain Awareness (SDA) TAP Lab in Colorado Springs, Colorado. The lab works on innovation accelerator cycles of two or three months. The latest cohort of companies selected includes Soresu LLC, a Colorado software firm founded in 2025 with just a few employees that leverages AI in orchestrating ground, space and air sensors for autonomous detection of events.
Companies such as Colorado-based Slingshot Aerospace already are bringing AI/ML to SDA — the term that describes an understanding of the realm and of the satellites moving within it. By using AI tools to fuse different data sources, operators can “get actionable information on operationally relevant timelines,” Audrey Schaffer, Slingshot’s vice president of strategy and policy, told the Constellations podcast in June 2025. “What we’re seeing is better exploitation of the data sources that we have.”
There’s a lot of data coming down all the time and we don’t really have the human bandwidth to analyze all of it.” ~ Capt. Drake Williams, U.S. Space Force
Data fusion for SDA delivers a better picture of a space object’s identity, positioning, movements, behavior and context. The next step, Schaffer said, is doing something with that AI-enhanced data — perhaps using the creative form known as generative AI to evaluate options and make operational recommendations.
Aerospace giant RTX Corp. and its missile-making subsidiary, Raytheon, are working to incorporate AI/ML into new missions such as space-based interceptors, Jeff McCall, Raytheon’s vice president for mission solutions and payloads, told the Spacepower Conference. “For algorithm development, algorithm optimization, light software development, developing fire control solutions for some of the new and exciting missions out there,” McCall said during a missile defense panel discussion.
For four decades, Virginia-based Raytheon has been building exoatmospheric kill vehicles that target an adversary’s ballistic missiles once they enter outer space. The Space Force in December 2025 solicited proposals for a new generation of space-based interceptors that would take out missiles at an earlier phase of flight. AI technology uses algorithms, a mathematical process for solving a problem using several steps, as its instructions for interacting with training materials to understand what outcome it should work toward.
Florida-based Redwire brings its AI/ML expertise to space defense through products such as the new Acorn 2.0 software, featuring AI-powered digital engineering tools. Acorn 2.0 uses agent-based modeling and AI to solve complex problems beyond the capacity of other modeling tools.
“An agent can be anything you want — a plane, a tank, an unmanned aircraft system, or it can be a satellite or multiple satellites,” Dean Bellamy, Redwire’s executive vice president for national security space, told Apogee. “I think this is going to allow us to smartly move AI onboard and actually demonstrate virtually how AI would work to an operator and prove it out. I think it’s going to be core to how not only Redwire but a lot of our partners build satellites.”
AI initiatives by Virginia-based Peraton aim to enhance national security, in part, by creating a relationship between humans and AI. Jessica Kalmanson, Peraton’s chief technology officer for space and intelligence, said one priority — the one that “keeps her up at night” — is developing systems resistant to rising attacks from adversaries. “AI in cyberattacks are more and more in the news, and we have adversarial AI and data poisoning. We now have to have systems that think about that,” Kalmanson said at the Spacepower Conference. “Can your AI detect when it’s being deceived? Can it detect, isolate and keep operating in a degraded state so the operator can keep focusing on the mission to be successful?”
Our economy depends on that constant disruption from innovative new companies.” ~ Pete Cannito, Redwire chairman and CEO
Companies using AI/ML for space defense, as is the case with AI/ML users everywhere, are working from a foundation created by a handful of pioneering research labs and tech behemoths. By some measures, there are eight of them: the U.S. firms OpenAI, Anthropic, Google, Microsoft, Meta and xAI, and the Chinese corporations DeepSeek and Moonshot AI. They are developers of massive, large language models (LLM), based on text, and the foundation models that contain text and other forms of data. “They are the architects of a new kind of internet, capable of generating code, analyzing complex data, and engaging in nuanced, human-like conversations,” according to the website Search Engine World. These companies are “pushing the boundaries of what’s possible and reshaping how we interact with technology.”
Said Williams, “It takes an incredible amount of resources to create a new large language model. It’s not something that the government really has the resources or expertise to do, so we’ve been really leaning on [industry] as we create the engines we can use for our specific use cases.”
The Space Force has rolled out a three-level model for leveraging AI/ML, from general to more specific uses. Level 1 is known as enterprise for its widespread, basic application; Level 2 is functional AI, narrowed some to specific activities within the service; and Level 3 is known as mission-specific AI.
Level 1 work draws from the data available through the LLM companies. Through its 2025 Space Force Artificial Intelligence Challenge, the service recently honored a half-dozen teams that created breakthrough products in areas such as onboarding new Guardians, tracking fitness and launch operations. A new Department of War (DOW) web tool called GenAI.mil enables all service members to access these resources to create their own work, whatever their area of responsibility.
Developers can train an LLM to perform a specific function or they can use its ability to communicate in natural language to pull out a piece and create a small language model, one “that you could run off a laptop,” said Bartley Stewart, head of the SSC Chief Digital and Artificial Intelligence Office and Information Dominance Division. “There’s advantages of that approach, especially when you get into things like Level 2 and Level 3 that are operational, where you want to be on the edge.” The edge refers to processing data close to the sensor that is its source, often used to describe autonomous satellites.
One tool for accomplishing these tasks comes from private industry. Ask Sage — billed as a secure, government-grade generative AI platform designed for defense, national security and regulated industries — lets users ask questions and get answers from vast, private datasets using natural language. Ask Sage, based in Virginia and founded by a former chief software officer with the Space Force and Air Force, supports more than 100,000 DOW users, 16,000 government teams and hundreds of commercial companies.
There are many potential mission-specific uses of AI/ML in space operations. “We’ve got a lot of satellites on orbit doing missile warning, weather, GPS activities, space sensing, taking images,” Williams said. “There’s a lot of data coming down all the time and we don’t really have the human bandwidth to analyze all of it. So we use algorithmic sorting to eliminate the things that maybe don’t matter so much and call to the attention of the human operators the things that do matter. Help them sort through different courses of action and present options so we reach this state of decision superiority where we can assess the situation, orient ourselves, make a decision and act — much more quickly than we could if we only have humans in the loop on these processes.”
Barbara Golf, strategic advisor for space domain awareness with the SSC, drove home the challenge of analyzing a growing deluge of data during an SDA panel discussion at the Spacepower Conference. “I’ve got 8,000 commercial sensors coming out of my ears, so I don’t think we have a sensing problem anymore. We’re drowning in data, literally drowning in it … Where do we put it?” The SDA TAP Lab, Golf said, has a backlog of 80 app requests.
Among issues to deal with are the sharing of proprietary code among competing vendors, hosting apps at the secret and top-secret level, and the different platforms required for different types of processing. “There is no one platform to rule them all,” she said.
Today, the major large language models aren’t truly tied to operations in a way that could help with space defense missions. “I don’t think that we trust them to that level yet,” Williams said — not to run, say, constellations of satellite sensors. The service is getting to the point, though, where it can test and accredit LLMs and eventually transfer more responsibility to automated tasks.
Initial prototypes focus on the human-machine interface — enabling an operator, for example, to forgo clicking through all the forms on a graphical interface as is now required to send a command to a system. Instead, with AI/ML, “you could say, ‘Please send a request to view this satellite using this sensor in the next 12 hours.’ You could give the larger language model that natural input, and then it could format that into a machine-readable command.”
The technology also could help discern adversarial activity from normal activity that’s detected by ground-facing Space Force satellites. “They can detect missiles, but they can also detect bright clouds or sun reflecting off of a plane and rocket launches that aren’t a threat,” Williams said. “That takes quite a bit of skill and training on the job to figure that out. So we use the LLM to basically look at the image of a missile track and determine, ‘Is that a missile?’ And then start to sort it into the broad categories of missiles so you can understand the threat better.”
LLMs help accomplish this by unraveling images into data that consist of 224-character strings of pixels. They’re not readable by humans. “Then it can, for lack of a better word, see the image. These data points represent distinctive differences among the many pixels that are being analyzed.”
Williams is working to stand up an AI hub within Space Force acquisitions, Stewart said, “so no matter what specific technology industry comes out with, what the next models look like, we can adapt them.”
The effort so far consists of a few building blocks. The first is infrastructure and data, ensuring there is computing power that can run the required algorithms on government-accredited hardware and with the right data that’s ready to go for the right sources. “Because if you get the latest model, but you don’t have anything to give it, it’s not very useful,” Williams said.
The second building block is talent and the workforce, “making sure we have the right experts in place, making sure the workforce understands how to use these models, if we put them in front of them.” The third is partnerships and ecosystem. “That’s really having the right connections to industry so we know what’s going on. Having the right connections to academia and at the labs and with federally funded research and development centers so that we can all share solutions.”
The fourth and fifth building blocks are change management and process reengineering to help speed up the accreditation of new systems, and a responsible AI governance framework “to make sure that we do all these things in an ethical manner.”
The push to leverage space commerce notwithstanding, the Space Force’s own research contributes to the advancement of AI/ML, too. Stewart tells the story of a respected private vendor who had done business with the government for a long time and “showed me a tool — and it was a good tool. It appeared to do what it was supposed to do. It filled a need we had for Level 2 tooling.” At around $500,000, “I don’t have any reason to think they were not asking a fair price.”
“The very next week,” Stewart continued, “there was a Space Force Guardian in a civilian institution who called me up and wanted to show me this tool he had developed. It was an 80% overlap of the tool I had seen previously.” Stewart asked the Guardian how long it took and how much it cost. “He said about two weeks … and about two pizzas.” Stewart added, “It wasn’t quite as polished, but it was real close to good enough.”
In general, though, the role of industry is to do the engineering and build the products, while the role of government is to make decisions, Stewart said, “whether it’s a decision around what do we buy, a decision around what company or companies do we go to to procure it, or it’s a decision around what do we think the adversary is doing or how do we react or respond? It’s all decisions.”
AI/ML can help operators reduce what’s known as cognitive load. “How long do you have. … How much are you thinking about? How much can you hold in your memory?” Another factor is information certainty, using AI/ML to fuse a number of inputs to be surer about a decision. “Simplifying them, combining them, so now what the human decision-maker has to make a decision based on isn’t as large, isn’t as complicated,” Stewart said. “You’re doing better. You’re really making a difference.”
Another role for government in its partnership with the space industry — perhaps its No. 1 role — is setting the standards and architecture for all players to follow, said Redwire CEO Cannito. “Because they’re really the only ones who are in a position to get everybody to play together.” Standards are needed for systems engineering, digital engineering, advanced modeling and simulation, “to make sure that all that stuff can work together, seamlessly, because that’s really where the power comes from,” Cannito told the Spacepower Conference.
Exercising this power also serves another goal of government, Cannito said: Leveling the playing field for the small startup companies where so much innovation resides. “Standards are absolutely critical to that,” he said, “because if you build these monolithic closed systems and you don’t publish the standard and then you find this innovating group of young professionals in the garage who have the next killer app, and they say, ‘All right, we’re ready to plug in in this one area because we do this better than everybody else.’ Sometimes you have to bypass that product just because it doesn’t fit in your architecture. That’s a real missed opportunity, not only for the user not getting the best capability but for our economy. Our economy depends on that constant disruption from innovative new companies.”
Stewart with SSC echoed the need for the acquisition system to protect small businesses — and to balance this against the need to keep down costs. “If we can get tools so cheap that we save the taxpayer a bunch of money, but by doing that we’re driving the small, innovative companies out of business, it’s not really a good value to the taxpayer because at the end of the day, our mission is to secure and defend space. If we degrade our industrial base’s capability to support the defense of space, then we haven’t done a good job.”
At the same time, the military must ensure the highest levels of reliability for systems using AI/ML. “I think it’s definitely different for commercial versus military,” Williams said. “The military has a unique mission; we’re talking about lives on the line, and mistakes are costly when the military makes them. So the standard has to be much, much higher. I think the leadership stance now is, ‘We understand that we’re going to need to implement AI into operations, to keep up with our competitors. But we have to do it very carefully and we have to take each step incrementally and fully test things out.’”
AI/ML capabilities aren’t end goals in themselves but a means of reaching them, Stewart said. “We’ll often talk about ‘AI Tool and Die’ … basically, a machine shop where you make other tools. … Not necessarily using AI to do the thing, but to develop the algorithm to do the thing. That’s where a lot of the power is.” This transcends the use of AI/ML for simply reducing toil and taking over mundane tasks. “The capacity for developing tools is super strong.”
Added Williams, “AI is itself not inherently a priority. Space Force has its priorities. … We’re just evaluating how do we get to our number 1, 2, 3 priorities? How does AI help us get after those faster?”