Shield AI took its drones from the ‘Batcave’ to the battlefield. Now the $5.6 billion defense-tech startup’s new CEO says it’s at an inflection point

There are plenty of conventional indicators that signal that a product is turning heads: Weekly active user figures start to soar, products fly off the shelves, there is unsolicited praise. But for San Diego-based Shield AI, validation has looked a little different. In April of this year, Russian armed forces fired two HESA Shahed 136 missiles into a hangar in Kyiv, where a team of 30 Shield AI employees had been doing research and development just two weeks earlier. The missiles turned the facility into a skeleton of twisted metal and rubble, according to a photo and video footage reviewed by Fortune. Incredibly, no one was harmed. James Lythgoe, a former U.K. Royal Marine who is now Shield AI’s managing director of Ukrainian operations, had moved the Shield AI employees to a new site, as he had been concerned about the newfound attention that its sprawling nine-foot-tall surveillance drone, the V-BAT, was picking up. “We were advised that the Russians were very aware of a new capability on the battlefield,” Lythgoe says. On the frontlines in Ukraine, Russian jammers intersect communications and radio signals, leading drones to veer off course or even fall from the sky and crash. Many U.S. drones haven’t been able to perform. But after an eight-month iteration period in 2024, Shield AI’s V-BAT cleared rigorous Ukrainian jamming tests. In 2025 alone, the drones have executed more than 35 missions and identified more than 200 Russian targets in the warzone, according to the company. The initial success Shield AI has seen with V-BAT in Ukraine and on U.S. shores with the Coast Guard and Marines has helped the startup land a $5.6 billion valuation and positioned it as one of the hottest defense startups of 2025, right behind its higher-valued and more hardware-heavy rival Anduril Industries. Major government contractors, known as the “primes,” have begun to pilot Shield AI’s autonomous aircraft software system, Hivemind, for the experimental aircraft they are building for the U.S. military. Foreign allies and U.S. partners like Romania, Indonesia, and Japan have purchased its surveillance drones. Shield AI wants to harness this traction and turn it into meaningful financial results. It’s looking to a brand-new autonomous fighter jet it’s building, the X-BAT, to help make it happen. It’s also looking to a new CEO. In May, the company brought in a new chief executive—Gary Steele—who has a track record of taking tech companies to multi-billion exits. With Shield AI’s cofounder and former CEO, Ryan Tseng, stepping into another leadership position, Steele has plans to grow the company’s revenue 70%-100% each year until it hits $1 billion in annual revenue for 2028, up from the approximately $300 million Shield AI notched in 2025. “I think the number one thing I think about is: How do we scale this?” says Steele, who spoke with Fortune over two interviews, his first since being named Shield AI’s CEO. It won’t be easy. As part of Shield AI’s strategy, the 1,200-person company will need to convince legacy defense shops that the AI-powered autonomous software Hivemind can do more than power Shield AI’s own drone. A gruesome accident in 2024—in which a U.S. Navy servicemember had the tips of his fingers effectively sliced off during a drill with the V-BAT—put a damper on last year’s revenue, and gave the company a public black eye that its executives are anxious to put behind them. And Steele, who is likable and seemingly adept at navigating internal politics, has walked into a leadership position notoriously difficult in the startup world: a CEO seat at a company where the founders maintain key leadership roles, board seats, and stakes in the business they created. Shield AI is at an inflection point. Now Steele will have to prove that he’s the one who can take it to the next level. ‘This inflection was happening’ Even before Anduril, there was Shield AI. Brandon Tseng, a former Navy SEAL, partnered up with his brother after he, Ryan Tseng, had sold a startup to Qualcomm. The two of them, with cofounder Andrew Reiter, wanted to take the autonomy that Elon Musk and Jeff Bezos were promising would transform the auto and e-commerce industries and translate it to the battlefield. This was back in 2015—two years before Anduril started to take shape, and not long before protests erupted within Google over a contract it was renewing with the Department of Defense. While Palantir had been securing government contracts for years, building military technology was rare among Silicon Valley tech-types at the time, not to mention exceedingly controversial. The Shield AI team turned down an initial $5 million investment because it had been contingent on Shield AI ditching its intended military focus and going commercial—which its founders weren’t willing to do. “It was really, really uncommon, if non-existent, for venture firms to be doing DoD-first companies,” says Peter Levine, a general partner at Andreessen Horowitz, who sits on Shield AI’s board. As the venture capital-backed defense tech industry has matured, however, the Tseng duo have become synonymous with the industry and with the traction the sector has garnered since geopolitical tensions started climbing in 2021. That climb sped up, of course, in 2022, when Russia invaded Ukraine and views on the space shifted dramatically. Shield AI had started with the now-discontinued quadcopter called the “Nova,” which, on first glance, looks like a superbly beef-ed up version of a drone you might buy at Radio Shack. Its innovation was in its tech stack, the AI-powered autonomous software system Shield AI calls “Hivemind,” which ingests data from onboard sensors—things like infrared cameras, radar, signals intelligence, and satellites—to build a model of its environment, then use AI to navigate, plan routes, avoid threats, and execute missions without the need for remote control. With Hivemind, the quadcopter could go into the most dangerous parts of a building and gather intelligence of potential ambushes or hidden combatants, so soldiers wouldn’t have to walk in blind. The Nova has been used for several missions in the Middle East, inlcuding in October 2023, when Israeli forces used it to explore Hamas’ tunnel network below the Gaza Strip. The Defense Department’s budget for quadcopters is relatively small, however, according to Ryan Tseng, so Shield AI pivoted in 2021 via its acquisition of the V-BAT, a towering surveillance drone capable of flying up to 18,000 feet and for 13 hours into enemy territory. The drone, which takes off and lands vertically, can fly from a ship or boat without a runway or launch mechanism, which has helped it notch contracts with the U.S. Coast Guard and Marines. But it’s the war in Ukraine that has really put V-BAT on the map. Like many other U.S. defense startups, Shield AI donated technology and hardware to Ukraine’s military for testing and experimentation—for proof that their drones could stand up in a conflict zone. Many of those companies quickly came to realize that they couldn’t, including Shield AI. The drones weren’t equipped to operate in areas where combatants could jam their communication signals or GPS, says Nathan Michael, Chief Technology Officer at Shield AI, who says the V-BATs they initially sent to Ukraine didn’t have Hivemind on board. “We had to come back and revisit our strategy,” he says. It took roughly eight months for Shield AI’s tech team to incorporate Hivemind into the V-BAT. After the update, V-BAT underwent two new rounds of intense testing in summer 2024: a two-day test-run where seven jammers tried to knock it down, as well as a 60-mile test mission, where the V-BAT was used in jammed airspace to spot a Russian surface-to-air missile system and alert the Ukrainians, who hit it with a rocket. Both tests were successful, according to Ukrainian documents reviewed by Fortune, and Shield AI eventually sent over 16 V-BAT drones to Ukraine—most of them purchased by European allies—and they’ve been serving in the field ever since. One of its most noteworthy missions thus far was in April, when a V-BAT flew some 80 kilometers into Russian-held territory, south of Zaporizhzhia, over two days to identify—then help destroy—two military headquarters and barracks, where Russian pilots and operators were remotely controlling the country’s highly-lethal FPV drone fleet. New business has been pouring in in the months since, according to Steele. Shield AI started selling its V-BATs to the Netherlands, Ukraine, and Egypt this year. Steele wouldn’t give specifics, but said that Shield has “hundreds of millions” of dollars worth of new contracts in Asia, Europe, and the Middle East alone. And this summer, in late August, the Ministry of Defence of Ukraine formally named Shield AI one of its “verified business partners,” allowing it to compete for state procurement contracts and access programs—and making it a true player in the war effort. “I suspect that this year, more than half of our business is international,” Steele says, noting that he arrived at the company “as this inflection was happening.” Shield AI is currently manufacturing the V-BATs out of its 200,000‑square‑foot “Batcave” production and engineering facility outside of Dallas, where the company is building 200 aircraft per year, though it just inked a deal with the manufacturer JSW to eventually start producing them in India as well. Shield AI either sells the V-BAT outright, or, as is the case for nearly all of its contracts with the U.S. military, serves as a contractor operating the V-BATs for the customer, and the orders or contracts range from 4 to 300 aircraft, according to the company. For purchase, each V-BAT costs about $1 million, though the cost can vary depending on how many the customer is purchasing or the tech that is integrated into the system. Shield AI also licenses Hivemind to customers, including Singapore and South Korea, as an autonomy software suite and developer platform. Hivemind made up approximately 30% of the company’s revenue in the 12 months ending in March 2025. While the company says it makes “some revenue” from the early demonstrations and integration work it is doing with primes, including Airbus, RTX, and Northrup Grumman, the future of that business line will largely depend on whether the Department of Defense eventually opts to purchase those products. ‘Every single investor made money’ Steele was almost gliding around the light brown wooden floors of his San Francisco condo when we first met in August. He had left his loafers in his office and was enthusiastically sliding about in his grey slacks and socks, pointing out various paintings that scatter the walls of his second home, a corner apartment with floor-to-ceiling windows on the top floor of a skyrise near the Ferry Building. “It’s hard to get the colors right,” Steele says as he points to a painting hanging in a guest bathroom. The artist, Doron Langberg, is one of many recent art school graduates that Steele began following on Instagram shortly after they graduated—a habit he picked up after he started collecting art in 2014. Steele—with his kind smile and knack for an emerging artist—was not the pick one might have expected at the helm of Shield AI, whose drones have helped destroy some $400 million worth of Russian weapons. Steele’s background is in software, running the companies Splunk and Proofpoint, which focused on data analytics and cybersecurity. Steele founded Proofpoint and says he scaled it to $1.5 billion in revenue before Thoma Bravo purchased it in an all-cash $12.3 billion deal in 2021. At Splunk, Steele came in when it was losing money, then sold it to Cisco two years later for $28 billion in 2024. Cisco kept him on, making him president of the company’s $55 billion go-to-market strategy. He is confident—maybe even a bit smug—in his track record of returns. “If you look at my history at Proofpoint, literally every single investor made money,” Steele says. “Every single one.” That, he says, is one of the reasons that Shield AI’s board, lined with Silicon Valley investors from Andreessen Horowitz and Point72 Ventures that have backed the company, thought Steele would do well in the CEO seat. “He has scaled very large companies,” Andreessen Horowitz’s Levine says. “We wanted an emphasis on software, because as we go forward, we intend to make that software available to many other organizations who will use that software on their hardware. And Gary had that background.” Steele joined the company just as Shield AI had announced its most recent funding round, $240 million at a $5.3 billion valuation. Shortly after the round closed, Shield AI extended the round by raising an additional $300 million, hoisting its valuation to $5.6 billion, Fortune is first to report. In total, the company has raised $1.4 billion in equity and $200 million in debt—taking it from a GPS-denied quadcopter company to one of the most well-funded private defense companies in the U.S. and one of the definitive players working on autonomy in the private markets. “They’re right there with Anduril,” says Ali Javaheri, an emerging tech analyst at PitchBook. “They have serious venture backing from the big firms. They have serious backing from the Primes. They are winning contracts.” But Shield AI hasn’t enjoyed the same scale that Anduril has. Anduril said it had notched $1 billion in revenue in 2024. Shield AI, comparatively, hit $300 million in 2025, according to the company. That was a $100 million shortfall of the $400 million it had been aiming for. Shield AI credits the shortfall to an incident that took place during a test with the U.S. Navy in 2024, which was first reported by Forbes earlier this year. One of its V-BAT drones had tipped over during a test, and a Navy servicemember who rushed to capture it inadvertently grabbed the propeller and severed the tops of three fingers, according to a summary of the subsequent investigation, which was obtained by Fortune via a records request. The Navy’s investigation said that, because of poor signal, it took 45 minutes for anyone to get a hold of emergency services before the servicemember, as well as the pieces of his fingers on ice, could be transported to the hospital, according to witness testimony and findings from the Navy’s investigation. Shield AI says it had a Tactical Combat Casualty Care-qualified employee who provided immediate medical care on site and then initiated immediate ground transport to the nearest medical facility. The incident was gruesome and publicly embarrassing. While most of the findings of the Navy’s subsequent investigation were redacted, the Navy documents say that Shield AI’s preflight brief packet didn’t have sufficient instructions for emergency procedures, and that Shield AI’s tip-over training did not include practical training exercises, according to the records. The V-BAT—even the drones operational and in the field—was grounded for two weeks as the investigation ensued, and it ended up delaying a series of contracts. “Aviation is dangerous. Machines are complicated, and through a Swiss cheese situation, a person lost their fingertips, and it was an unfortunate event,” says Ryan Tseng, who was still CEO at the time of the incident. After the incident, the company added a warning on the duct surrounding the propeller, along with “extensive” hands-on practical exercise requirements. It later rolled out an unassisted launch and landing capability that eliminated the need for a person to be involved at all. Tseng described the Forbes story about the incident as “sensationalized” and contested the notion that there were any deeper-rooted safety issues at the company, or that the accident had any relation to his decision to step aside. While “many purchasing decisions were delayed as a consequence of that investigation,” Tseng says, “for a long time, it’s been back to normal.” In interviews, Ryan Tseng and Levine emphasized that it was Tseng’s idea to step into the chief strategy officer role and bring on a new CEO. “He wasn’t pushed out,” Levine insists, adding: “It’s not like he did anything wrong.” Ryan Tseng says that, as the company hit 1,000 employees, he questioned whether he was the person to take it to 5,000 people. “I’ve told people, and I don’t think they believe me, but I’ve never felt a particular attachment to the CEO role,” Tseng says. Tseng says he first approached the board this past winter, but they encouraged him to stay on. After the funding round closed, he suggested they revisit the conversation. About seven months into the leadership transition, the Tseng brothers and Steele say they have found a balance and that they talk every day. Ryan Tseng has moved into the strategy role, where he oversees corporate development and M&A. Brandon Tseng, who is based out of Washington, D.C., continues to lead growth and is focused on customers and investor relationships. Steele is focused on running the business, making money, and bringing on new people, having hired four new executives since he joined, including a Chief Legal Officer and Chief Marketing Officer. “This transition between Ryan and Gary has been the best transition from a founder to a new CEO that I’ve ever seen. And I’ve been around for a while,” Levine says. But proof will come with time, as these kinds of transitions can be exceptionally difficult to pull off. Sometimes cofounders struggle to give up control in the company they’ve built themselves, or become skeptical their replacements are doing an adequate job. Bumble founder Whitney Wolf Herd, for example, stepped back as CEO in 2023, only to come back around one year later after a rocky few months at the company. Or at Uber, when CEO Travis Kalanick stepped back from his position but remained on the board, there were reports of conflicting vision and power struggles. When asked about the dynamic between himself and the Tseng brothers, Steele says he was well aware of the importance of their roles, because he was a founder himself. “I understand what that means,” he says, noting that he wouldn’t have joined the company if he didn’t feel like they could work well together. “I needed to feel like we saw the world in a similar way,” he says. For him, he says he was convinced that the Tseng brothers approached the world with the same instincts as him, a “relentless” work ethic, and a “hands-on, problem solver’s mindset.” The company wouldn’t share what voting power the brothers still have, only that they are “still significant shareholders.” The company said that Shield AI “operates with a mature governance structure and an independent Board. No single individual has the ability to make leadership changes on their own; those decisions rest with the Board as a whole, just like any well-run company.” What’s coming next At the end of October, Shield AI unveiled a brand-new product: an autonomous fighter jet with a 2,000-mile nautical range called the X-BAT. Shield AI has been working on the X-BAT for 18 months, designing a massive vertical take off and landing aircraft that wouldn’t need a runway, according to Brandon Tseng. Shield is aiming to have its first test flight sometime next year, and start production in 2029. The X-BAT is intended to complement the V-BAT, which is proving to be the company’s workhorse—at least for now. But in the meantime, Shield AI wants to put more emphasis on the Hivemind software to meet its lofty revenue goals—hoping that product will make up 50% of the company’s revenue by 2028. While the company currently licenses its software out to foreign governments to use on their defense systems, it also wants to lean further into partnerships with the “primes”—the behemoth military contractors that have been the primary customers of the U.S. military for decades—so that Hivemind can eventually be incorporated into everything from helicopters to fighter jets. So far, Shield AI is working with eight of the military’s main 25 contractors, according to Ryan Tseng. For starters, it is being incorporated into General Atomics’ MQ-20 unmanned combat aerial vehicle, a Kratos BQM-177A target drone, and an Airbus H145 twin-engine light utility helicopter. But, importantly, these have been demonstrations, not deployments, with little revenue. Shield AI still has to prove its capabilities to these primes—and eventually to the Defense Department—before they would roll the technology out widely. “The customer has to have confidence to go do this,” Steele says. One of those early partners is Airbus, which started working with Shield AI in spring 2025 on an Airbus DT25 target drone as well as an autonomous developmental Lakota helicopter that it hopes to deliver to the Marine Corps in the next “couple of years,” according to Carl Forsling, director of business development and strategy at Airbus. “If that’s successful, then that market is going to continue to expand—both with the Lakota and potentially other platforms,” Forsling says. Steele emphasized that the company wants to position itself across a series of platforms. “While we’ve been very focused on aircraft, because that’s the place we started, there’s tremendous opportunity as we cross domains,” he says. PitchBook’s Javaheri pointed out that Shield AI is likely to benefit from the Defense Department’s recent decision to hone in its 14 priorities down to six, one of which is “applied artificial intelligence” systems, which would include autonomy. “Aerospace and defense autonomy is the name of the game, and Shield AI is one of the leaders in that,” he says. On the front lines While defense tech companies are becoming increasingly prevalent in Silicon Valley—and Washington, D.C.—there is something intrinsically different about a defense company than its enterprise or consumer counterparts, even if the same storied venture capital firms have begun backing all of them. Shield AI is a case in point. For one, its makeup: 18% of its 1,200 employees are veterans, including Shield AI’s head of communications, Lily Hinz, who served in the Navy. Nearly all of the 30 employees stationed in Kyiv are former Ukrainian soldiers. But more importantly, there’s a difference in mentality and approach—perhaps due to the high stakes and real-life consequences of the projects people work on and the soldiers they work on them for. This is very evident from Shield AI’s 41-page document explaining its culture, which the company publishes on its website. In it, cofounder Brandon Tseng lays out a personal anecdote behind one of the company’s values—“do what honor dictates.” He writes about how one of his Navy SEAL instructors had dragged a team member to safety with one arm after being shot in the other. “While there are many ways to conduct ourselves, we choose to act in a manner that is moral, good, and of high standards—leaving the world better than we found it, simply because it’s the right thing to do,” Tseng wrote. There are ethical grey areas for defense tech companies that don’t exist in the rest of Silicon Valley—when you build a surveillance machine or a weapon, and when the thing that you build is responsible for saving human life, or for taking it. “It’s a huge responsibility to get it right,” says Ryan Gury, who had a background in consumer drones before he started the defense drone company PDW. “You’re selling equipment that is going to extend the life and lethality of our operators.” Garrett Smith, an active Marine Officer who is CEO of the tactical edge tech company Reveal Technology, says that, when a product lives in a “life-and-death” environment, it “changes everything.” “You prioritize reliability, safety, and mission outcomes over vanity metrics. You also have to think about escalation dynamics and law-of-war implications in a way a typical startup never does,” he says. “‘Move fast and break things’ is the wrong mantra when ‘things’ are people and escalation paths.” Several tech companies that operate in this space have set up teams to wrestle with these topics. Palantir has a “Privacy & Civil Liberties Engineering” team designed to “foster a culture of responsibility” around how their technology is used. Even then, Palantir is extraordinarily controversial among many, particularly because of its contracts with Immigration and Customs Enforcement. Risk is very real for Shield AI employees. In contractor-operated deals, as well as in complex, high-risk environments, employees are often stationed for months on the ground (or at sea) where its drones are deployed. In Ukraine, its 30 operators regularly travel between cities to support mission planning, monitor sorties, and troubleshoot in real time to adapt to new threats and feed lessons learned back into the V-BAT. That level of proximity is all about trust, according to Lythgoe, Shield AI’s head of Ukrainian operations, who says that, if you are going to ask a soldier to trust their life with your technology, you need to be able to prove that you are just as committed to them. That has meant Lythgoe has only been home with his wife back in the U.S. four weeks over the last year, which is “not ideal,” he admits. “That is the job, I believe,” Lythgoe says. “Inherently, it’s the role of the defense sector to understand problems and to give the war fighter the edge. And to do that, you have to understand the problem, otherwise you’re guessing. And so you really do need to be close to the problem to do that.” It’s curious, then, that Shield AI’s new CEO talks in circles about whether he feels a heightened sense of responsibility running a defense tech business, and seems uncomfortable to be asked about it at all. When asked about increasing disagreement about U.S. involvement in Ukraine or the controversy around the Coast Guard carrying out the Trump Administration’s agenda for Venezuela, he said: “We literally spend no time talking about the politics of particular missions.” While Steele acknowledged Shield AI has different protocols and processes because there is “human life involved,” he repeatedly stated that Shield AI isn’t much different from other tech companies. His focus is on the “mission,” he says, and how to “deliver the customer outcomes.” Update, December 21, 2025: This story was updated to reflect that the Batcave facility is now 200,000 square feet. Editor’s note: On March 26, three months after this article was first published, Shield AI informed Fortune that the previous financial data it had provided for 2025 relied on a different reporting period. The company informed Fortune that the previous data, which came from Shield AI, had been inaccurate. This story has been updated to reflect the correct reporting period.