Climbing the Aviation Food Chain With Shield AI

May 2, 2024. Edwards Air Force Base, California. Secretary of the Air Force Frank Kendall climbs into the cockpit of a modified F-16 fighter jet. The jet takes to the sky and proceeds to engage in a series of dog-fighting scenarios versus another F-16.

After roughly an hour hurtling through the California sky at 550 mph, and pulling 5 Gs, the jet is back on the ground. The invigorated Air Force Secretary steps out, grinning.

Kendall was a passenger in a jet piloted by AI. It went toe-to-toe with another F-16 manned by a human, and Kendall had seen everything he needed to be convinced of its promise.

The AI pilot responsible was built by Shield AI — a San Diego defense company, founded in 2015, now valued at $12.7 billion.

How did it get here?

Brandon Tseng served as a Navy SEAL officer for seven years, completing multiple deployments overseas, including two in Afghanistan.

It wasn't until he left the Navy and started applying to business schools that his older brother, Ryan, began to understand what those deployments had actually involved.

Ryan was helping Brandon prepare for admission interviews. What's the most challenging circumstance you've ever faced, and what did you do about it?

But Shield AI has no single origin story. Its purpose is grounded in a multitude of painful combat experiences.

An 8-year-old boy, caught in crossfire during a Taliban ambush on Brandon's platoon, shot in the stomach. Brandon couldn't call a medevac — the helicopter might have been shot down. So they carried the boy ten kilometers to a base. Somehow he survived.

A fighter pilot hovering over a confirmed ISIS training camp. He was cleared to drop his ordnance, but kept circling. Something didn't feel right. Then dozens of children emerged from the building. He returned to his carrier without releasing his weapons. He is still haunted by what could have been.

A 2012 raid in Afghanistan where operators, taking fire from a building, exhausted every option before calling in an airstrike. They went through the rubble afterward and found that the fighters had been holding a family hostage inside.

Through these experiences runs a common thread. Danger was in the absence of information at the critical moment.

Not knowing who was in the building. Not knowing whether the compound was a training camp or a school. Not knowing, until it was too late, that a family was being held inside.

Brandon had seen what that uncertainty cost, measured in lives and in the decisions soldiers and commanders had to make without the intelligence they needed.

The problem Brandon identified was one that had plagued urban combat since long before he served. It is known as the fatal funnel.

Any doorway you have to cross not knowing what's on the other side. In the close-quarters fighting that defined so much of the post-9/11 wars, clearing buildings was always a gamble. You never knew who or what was waiting inside.

When Brandon put the idea to Ryan in 2015, his brother's initial reaction was skepticism. Surely the military, with all its technological resources, had already figured this out?

It hadn't.

Two problem statements define Shield AI's evolution as a company:

Problem 1:

Room clearance with limited intelligence presents an unacceptable risk to military operators and civilians, exposing them to lethal threat at the moment of entry with no means of knowing what lies on the other side of the door.

Value Proposition 1:

An AI-powered quadcopter capable of autonomously navigating and mapping a building without GPS or human control, feeding real-time threat intelligence back to operators before any human crosses the threshold.

As US military deployments shifted away from counterinsurgency operations, the room-clearing problem became less central to the battlefield. The rise of peer adversaries, electronic warfare, and contested airspace exposed a larger and more strategically consequential version of the same underlying challenge.

Problem 2:

Unmanned autonomous vehicles operating in hostile environments where GPS is denied and communications are jammed represent the critical capability gap of modern warfare, leaving operators blind and platforms mission-incapable at precisely the moments they are needed most.

Value Proposition 2:

An AI pilot — platform-agnostic, GPS-independent, and capable of autonomous decision-making when communications are severed — that enables unmanned systems to complete their missions in fully contested electromagnetic environments.

Shield AI hadn't pivoted. It had scaled the same thesis to a much bigger stage.

In the spring of 2015, Brandon and Ryan Tseng, along with engineer Andrew Reiter, had a business plan and $100,000 scraped together from friends and family.

Thirty meetings with potential investors. Thirty rejections. When a term sheet eventually arrived, it came with a condition: change the mission. Develop a consumer drone instead — something for selfies.

They turned it down.

A year later they got another term sheet for $800,000. Then Andreessen Horowitz led the Series A in 2017 at $10.5 million.

With that initial investment, and a first contract from the newly formed Defense Innovation Unit, Shield AI deployed the Nova drone with US special operations forces in the Middle East in the winter of 2018.

The company that had been told repeatedly it was a bad investment had just put its technology into active combat. The next question was how far it could scale.

Hivemind is the AI software at the core of every Shield AI product. It is much more than a drone operating system.

The architecture is, in principle, platform-agnostic. It doesn't care what it's flying. Which meant Shield AI could start with the smallest, cheapest platforms and work upward, accumulating credibility at each step.

Nobody was handing the keys to a fighter jet to a three-person startup. So they earned them. Nova proved the AI worked indoors, V-BAT moved it outdoors and into contested airspace.

Hivemind has now flown on fifteen different platforms — including the US Air Force's F-16, General Atomics' MQ-20 Avenger, and Anduril's YFQ-44A Fury.

At the top of the food chain sits X-BAT — Shield AI's own fighter-class aircraft, unveiled in October 2025. Vertical takeoff and landing (VTOL). No runway required. No human pilot required.

It boasts a 2,100 nautical mile range, a ceiling above 50,000 feet, and a target acquisition cost of $27.5 million (roughly a quarter of an F-35, at a fraction of the operating cost).

From a quadcopter clearing rooms in Afghanistan to an AI-piloted fighter jet that needs neither a runway nor a human — the food chain strategy, a decade in the making, is starting to look like a blueprint.

In July 2021, Shield AI made two acquisitions in the same month. The first was Heron Systems, whose AI had beaten an experienced F-16 pilot 5-0 in DARPA's AlphaDogfight Trials the previous year. The second was Martin UAV, maker of the V-BAT.

By 2024, the V-BAT was operational with Ukrainian forces in one of the most electromagnetically contested environments on earth.

The missions that followed destroyed hundreds of millions of dollars of Russian equipment. All of it without GPS.

By April 2025, Shield AI had logged more than 130 V-BAT sorties inside Ukraine, every one of them in a fully jammed environment.

The Nova 2 drone was deployed by Israeli forces following October 7, 2023, for indoor reconnaissance in Gaza.

In September 2025, Shield AI was selected as one of two autonomy providers for the US Air Force's Collaborative Combat Aircraft (CCA) program, with Hivemind chosen to fly Anduril's YFQ-44A Fury prototype.

In March 2026, Shield AI raised $1.5 billion in a Series G round led by Advent International and JPMorgan, at a post-money valuation of $12.7 billion. This more than doubled its worth in twelve months.

From an $800,000 term sheet to a $12.7 billion valuation in under a decade.

Shield AI's central claim is deceptively simple: autonomy is a software product, and the airframe is a commodity wrapped around it.

Where the defense primes bundle AI into specific platforms — autonomy as a feature of a Lockheed aircraft, a Raytheon system — Shield AI sells the intelligence layer separately. Hivemind doesn't care what it's flying.

A government customer already locked into a hardware contract doesn't have to walk away from that commitment to work with Shield AI. They can bring the AI pilot in afterward.

This is the strategic moat. Not the drone, but the intelligence that flies any drone.

Anduril builds the YFQ-44A Fury airframe for the US Air Force's CCA program and Shield AI's Hivemind software flies it. Two of the most prominent voices in defense tech, simultaneously collaborating and competing on the same aircraft.

Anduril also wanted the autonomy contract. It didn't get it. Shield AI's software is now embedded in its rival's most important hardware program.

Whether that dynamic favors Shield AI long-term — or quietly creates a dependency on a competitor's hardware roadmap — is an open question. For now, it is the most concrete proof that the platform-agnostic thesis works.

By the time Shield AI was bidding for the CCA autonomy contract, it had a decade of combat deployments behind it, something none of its competitors could match. That is a moat the hardware-first companies can't easily acquire retroactively.

A company that fully believed its own software thesis wouldn't need to build a fighter jet. The platform-agnostic model is, by definition, most valuable when the platforms belong to someone else.

And yet, in October 2025, Shield AI unveiled X-BAT — its own VTOL fighter-class aircraft, self-funded, with a production target and a unit cost that puts it in direct competition with the hardware primes it has spent a decade positioning itself alongside rather than against.

There are two credible readings. The first is that the Pentagon still fundamentally buys hardware, and the most reliable way to ensure Hivemind is always in the room is to own the airframe it flies.

The second is that Brandon Tseng, having climbed the aviation food chain, has decided to own the top of it. Neither reading undermines the software thesis entirely, but both suggest its limits.

What Shield AI is building, in practice, may be something more interesting than either a software company or a defense prime: a new kind of vertically integrated autonomy business, where the software is the product and the hardware is the proof.

April 2024. A US Navy test range. A V-BAT drone completes its mission and begins its descent.

As it comes down to land, it tips. A serviceman moves in to steady it. The propeller partially severs three of his fingers.

A brutal, unglamorous moment for a company that had spent nearly a decade arguing that putting machines in harm's way was precisely how you kept humans out of it.

The V-BAT's vertical landing sequence required human assistance. Someone had to be there, close to the aircraft, as it came down. That design assumption had just cost that someone dearly.

Shield AI grounded the aircraft, redesigned the launch and recovery sequence to remove the need for human assistance, and subsequently passed two government safety audits.

But the commercial damage was real. Revenue came in at $300 million for 2025 against a $400 million target. Newly-minted CEO Gary Steele attributed the miss directly to the V-BAT safety pause and the customer flight restrictions that followed.

Shield AI operates in an increasingly crowded field. Anduril, its CCA collaborator and most credible rival, is vertically integrated at a scale Shield AI does not yet match.

Anduril also combines its hardware with its own Lattice software platform, as well as the heft of Palmer Luckey's profile and Peter Thiel's backing.

Shield AI's differentiator is clear: no rival has Hivemind's combat record and platform flexibility.

That argument has been persuasive so far. Whether it remains so, as rivals build their own operational track records, is the question the next few years will answer.

The X-BAT timeline is aggressive. First vertical takeoff tests in 2026, full flight demonstrations by 2028, production by 2029. Anduril, with considerably more hardware experience, took roughly five years to get the Fury airborne.

At $12.7 billion on approximately $300 million in 2025 revenue, the valuation requires the market to believe in a trajectory CEO Gary Steele has publicly committed to: roughly $1 billion in revenue by fiscal year ending March 2028.

The math depends on the CCA program going to production, V-BAT scaling internationally, X-BAT staying on schedule, and Hivemind Enterprise finding paying OEM customers beyond Shield AI's own airframes.

Each of those is plausible. All of them together, on that timeline, is a demanding ask.

In May 2025, Shield AI appointed Gary Steele — who sold Splunk to Cisco for $28 billion — as its first non-founder CEO.

The founders remain in senior roles, but the signal is clear. This is a company preparing for $1 billion in revenue, a public markets conversation, and a seat at the table where the future of American airpower gets decided.

Brandon frames X-BAT as this generation's F-16, and Hivemind Enterprise — already purchased by the Singapore Air Force — as the longer bet: Shield AI as the operating system on which allied nations build their own AI pilot capability.

But the technology, however impressive, is ultimately replicable. What isn't is the human capital behind it.

Brandon's combat experience opens doors no venture capital can buy. Ryan's instincts built a business thirty investors had called idiotic. Andrew Reiter built the engineering foundation that made any of it possible.

That combination — operator, entrepreneur, engineer — is as much a competitive moat as Hivemind itself.

The harder question remains open. Secretary Kendall said on the record that he'd trust the AI to decide whether to release weapons. Shield AI's public position is more cautious — human oversight remains in the loop.

The gap between those two statements is where the most consequential debates in defense ethics, international law, and geopolitics will play out. Shield AI is at the center of that conversation whether it chooses to be or not.

It is a long way from a $100,000 prototype and a quadcopter that could clear a room. But the thread connecting them is the same one Brandon Tseng has been pulling since 2015.

The most dangerous moment in any conflict is the one where a human being has to go somewhere without knowing what's waiting for them. That was the problem, and solving it is still the mission of Shield AI.

To send the machine first.

The founding story:

A Navy SEAL, a Quadcopter, and a Quest to Save Lives in Combat — Elliot Ackerman, WIRED, December 2020

The Kendall flight:

An AI-Controlled Fighter Jet Flew Against a Human Pilot in a Secret Dogfight — Fortune/AP, May 2024

Inside the AI-Enabled Pilot that Flew Air Force Secretary Kendall Through a Dogfight — DefenseScoop, May 2024

The V-BAT incident:

How a Grisly Injury Threw $5 Billion Drone Startup Shield AI Off Course — David Jeans, Forbes, May 2025

Ukraine deployment:

Shield AI Surpasses 130 V-BAT Sorties in Ukraine — Shield AI, 2025

Funding & valuation:

Shield AI Raises $1.5B Series G at $12.7B Valuation — Fortune, March 2026

Israeli deployment context:

Nova 2, Legion-X, and the AI Political Declaration — Lieber Institute at West Point, November 2023

Interviews:

Ryan Tseng — CSIS / Wadhwani AI Center, June 2025

Ryan Tseng — Future of Defense: Shield AI’s Ryan Tseng & Axios’ Colin Demarest

Brandon Tseng — Shawn Ryan Show, October 2025

Brandon Tseng — WSJ Bold Names podcast, April 2026