Shield AI

Modern militaries are being asked to fight where GPS is spoofed, communications are jammed, and a human operator cannot be kept in the loop. Those are precisely the conditions under which conventional drones and waypoint autopilots fail. Shield AI builds both halves of the answer: Hivemind, an AI pilot that flies aircraft and completes missions with no GPS, no comms link and no human pilot; and V-BAT, a runway-independent VTOL aircraft delivering combat-proven ISR and strike from ships, rooftops and austere sites at a fraction of the cost and logistics footprint of a large Group 4/5 drone. Shield AI is the clear number two in the modern defense-manufacturer category behind Anduril, and the only player whose autonomy stack is now being procured by the U.S. Air Force as a standalone capability rather than as a feature of someone else’s airframe.

Growth is inflecting. Revenue was roughly $300M in the fiscal year ended March 2025 (up from ~$267M), management guides to more than $540M in FY2026 (>80% YoY), and CEO Gary Steele is targeting ~$1B by FY2028 on 70-100% annual growth. More than 50% of the business is now international. In the last twelve months the company won a U.S. Air Force production contract to supply Hivemind mission autonomy for the Collaborative Combat Aircraft (CCA) program, was selected to compete for up to $800M of U.S. Navy contractor-owned/contractor-operated ISR task orders, and continues to deliver against a $198M U.S. Coast Guard V-BAT IDIQ.

In March 2026 Shield AI raised approximately $2B at a $12.7B post-money valuation, up ~140% in twelve months: a $1.5B Series G led by Advent International and co-led by JPMorganChase’s Security and Resiliency Initiative, plus $500M of fixed-return preferred equity and a $250M delayed-draw facility from Blackstone. Proceeds fund three priorities: scaling Hivemind Enterprise as a licensed autonomy platform for third-party developers, building X-BAT (an AI-piloted VTOL fighter that completed engine light-off with GE Aerospace in July 2026), and integrating Aechelon’s simulation technology to supply the synthetic training data behind a Hivemind foundation model for defense. This memo covers a $2M secondary block of Series G common at $153.68 per share, $161.77 all-in after the one-time Layer 1 broker fee.

Hivemind is the secret sauce. It is a platform-agnostic AI pilot rather than an autopilot: it senses, decides and acts, rerouting around dynamic obstacles, executing collaborative tactics with peer aircraft and crewed fighters, and completing the mission when GPS is denied, the datalink is severed and no operator is reachable. Because it ships as software with a developer SDK, it attaches to other people’s airframes and turns would-be competitors into distribution. Honeywell Aerospace signed an MOU in July 2026 to build its Anthem-based trusted autonomy stack on the Hivemind SDK, and L3Harris flew Hivemind as the decision layer of its DiSCO electromagnetic battle-management ecosystem, with unmanned aircraft detecting and responding to EW threats without human intervention.

For the customer the benefits are concrete. Missions get completed in electronic-warfare conditions where conventional UAS abort. One operator can supervise many aircraft instead of a crew per aircraft. And V-BAT delivers ISR and strike from a ship deck, rooftop or unimproved site with no runway and a small logistics tail, at a fraction of the unit and sustainment cost of a Group 4/5 platform. That is why V-BAT has been fielded by the Marine Corps from Navy ships and by the Coast Guard, and why more than half of revenue is now allied and international. Switching costs compound the advantage: once Hivemind is the certified autonomy layer inside a program of record, displacing it means recertifying the aircraft.

Two product lines plus a developer platform. Hivemind is the AI pilot: platform-agnostic, A-GRA compliant autonomy software that assumes the role of the human operator. It is trained in simulation, hardened on real flight data, and requires no GPS, no datalink and no human in the loop. Hivemind Enterprise wraps the same stack in an SDK so third parties can build their own autonomous behaviors on it. This is the software line, roughly 30% of ARR today and targeted at ~50% by FY2028.

V-BAT is the flying product: a Group 3 ducted-fan VTOL aircraft with 12+ hours of endurance on a heavy-fuel engine, launched and recovered unassisted from ship decks, rooftops and unimproved sites with no runway and no catapult. The enclosed-rotor, single-engine architecture is the hardware IP that makes shipboard and austere operation practical, and it is what the Marine Corps and Coast Guard actually fly today. Manufacturing capacity is approximately 200 aircraft per year.

X-BAT is the option value: an AI-piloted VTOL fighter jet designed to operate from ships without a runway, powered by a GE Aerospace F110-GE-129E fitted with a thrust-vectoring AVEN nozzle. Integration, actuation and engine light-off were completed in July 2026, clearing the path to vertical flight. Revenue from X-BAT is minimal today; it is funded from Series G proceeds and should be underwritten as upside, not base case.

Aechelon Technology, acquired in June 2026, supplies high-fidelity simulation, physics-based sensor modeling and synthetic environments. It is effectively the training-data factory behind the planned Hivemind Foundation Model for Defense.

Shield AI sells into two nested pools. The narrow one is military UAS: roughly $34.9B in 2026 growing to ~$109B by 2031 on the more aggressive estimates (MarketsandMarkets, ~26% CAGR), versus ~$22.5B in 2026 to $52B by 2034 on the conservative side (11% CAGR). The dispersion is itself the signal: forecasters agree the fully autonomous segment is the fastest-growing slice and disagree only on how quickly autonomy displaces remotely-piloted architectures.

The wider and more relevant pool is defense autonomy software and the platforms it flies. The FY2026 U.S. defense budget put more than $14.5B into AI and autonomy programs, up 22% year over year, inside a multi-year autonomy portfolio of roughly $54-55B (Replicator now folded into the DAWG construct); FY2027 earmarks about $1B for CCA aircraft acquisition alone, against a program of record contemplating well over a thousand aircraft. Set against a >$1T U.S. defense budget and ~$2.89T of global military spending in 2025, Shield AI’s ~$540M of FY2026 revenue is a rounding error. The binding constraint is program capture, not market size. Customer spend per program is substantial: the Coast Guard V-BAT IDIQ alone is $198M, the Navy COCO ISR vehicle up to $800M in task orders, and a service-level autonomy license attached to a CCA fleet is a multi-hundred-million-dollar, multi-year line item.

Competition. Anduril is the direct rival and the benchmark: ~$2.2B of 2025 revenue, ~$4.3B guided for 2026, valued at $61B in May 2026 and reported in late July 2026 to be in talks near $100B, so roughly 5x Shield AI on both revenue and valuation, with Lattice competing against Hivemind. The two nonetheless partner on CCA. Beyond Anduril: General Atomics on the CCA airframe side, AeroVironment and Kratos in tactical UAS, Skydio in small UAS, and the traditional primes. The most telling competitive fact is that L3Harris and Honeywell chose to build on Hivemind rather than against it.

Three motions, in order of maturity. First, direct U.S. programs of record, won on flight-demonstrated capability rather than paper proposals. V-BAT is sold both as a system and, increasingly, as a service under contractor-owned/contractor-operated ISR arrangements, which converts episodic capital sales into recurring operating budgets. The $198M Coast Guard IDIQ and the up-to-$800M Navy COCO ISR vehicle are both structured this way.

Second, allied and international sales, now more than half of revenue, sold government-to-government with local industrial participation. Ukraine functions as both customer and the highest-signal reference account in the world; the Netherlands, Japan, Romania and Egypt followed, and the $90M JSW partnership localizes manufacturing in India to satisfy offset requirements. This is the fastest-growing channel and the least exposed to U.S. continuing resolutions.

Third, and strategically the most important, autonomy licensing. Hivemind Enterprise is sold to other primes and integrators as the autonomy layer for their airframes, with an SDK and, post-Aechelon, a simulation environment to develop against. Anduril’s Fury, Honeywell’s Anthem stack and L3Harris’s DiSCO are all Hivemind carriers. Expansion follows a software playbook: land the autonomy layer on one program, expand to adjacent platforms in the same service, then scale price with fleet size. The CCA production award is the proof point that a service will buy autonomy as its own line item.

How they won: operator-founded credibility in Washington, demonstrated performance in Ukraine rather than at a test range, and a deliberate decision to be airframe-agnostic so that every competitor’s platform is a potential distribution channel.

Revenue (fiscal year ends March): ~$267M FY2024, ~$300M FY2025, guided to more than $540M in FY2026 (>80% YoY), with a ~$1B FY2028 target implying sustained 70-100% annual growth. The FY2026 guide implies an exit run-rate of roughly $45M per month.

Honeywell Aerospace and Shield AI to co-develop trusted autonomy on the Hivemind SDK (July 22, 2026). Honeywell will build an autonomy software stack on Hivemind, integrating its Anthem avionics, navigation and sensing portfolio. A prime choosing to build on Hivemind rather than compete with it is the strongest available evidence that the licensing strategy is working well.

GE Aerospace and Shield AI complete X-BAT engine light-off (July 20, 2026). The two completed integration, actuation and light-off of GE’s axisymmetric vectoring exhaust nozzle on an F110-GE-129E, clearing the path to X-BAT vertical flight and de-risking the most technically ambitious program in the portfolio.

Shield AI awarded U.S. Air Force production contract for CCA mission autonomy (June 17, 2026). The Air Force moved Hivemind into production as a standalone autonomy capability intended to span multiple CCA airframes rather than one platform. This is the single most consequential commercial event in the company’s history and the direct driver of the March 2026 re-rating to $12.7B.

Founded in 2015 by brothers Brandon Tseng (U.S. Naval Academy, former Navy SEAL, Harvard MBA) and Ryan Tseng (serial entrepreneur; founder of WiPower, acquired by Qualcomm), with co-founder Andrew Reiter (computer vision and sensor fusion, ex-Draper Laboratory). The founding insight came from Brandon’s deployments: rooms that had to be cleared by hand because no machine could do it. That operator credibility is still the company’s front door in Washington and with allied ministries of defense, and it is not a credential competitors can hire for.

The team has been deliberately re-tooled for scale. Gary Steele, who ran Splunk through its $28B sale to Cisco, became CEO in May 2025, bringing enterprise-software operating discipline to the Hivemind licensing motion; Ryan Tseng moved to President and Chief Strategy Officer. With the Series G, Advent Chairman David Mussafer joined the board and JPMorganChase’s Todd Combs joined as a board observer, adding late-stage governance ahead of an eventual listing. The CEO hire is the clearest signal of intent: a software CEO was brought in to run what has been sold as a hardware company.

Footprint: approximately 1,350 employees as of December 2025, up ~43% year over year and ~82% since 2023, across San Diego (HQ), Dallas, Washington D.C., Abu Dhabi, Kyiv and Melbourne, plus a $90M JSW-partnered manufacturing hub in India. V-BAT production capacity is roughly 200 aircraft per year.

Structurally rising defense budgets. Global military spending reached roughly $2.89T in 2025, the eleventh consecutive annual increase; NATO members committed to a 5%-of-GDP target by 2035; and the U.S. defense budget topped $1T for FY2026. This is a treaty-anchored, decade-long spending cycle rather than one favourable budget year, and it is why more than half of Shield AI’s revenue is now non-U.S.

The Pentagon’s shift from exquisite platforms to autonomous, attritable mass. FY2026 put more than $14.5B into AI and autonomy programs, up 22% year over year, inside a multi-year autonomy portfolio of roughly $54-55B (Replicator now folded into the DAWG construct), with about $1B earmarked in FY2027 for CCA aircraft alone. The persistent bottleneck in that push has been mission-autonomy and swarm-orchestration software, which is precisely Hivemind’s product rather than an adjacent opportunity.

Ukraine has made GPS- and comms-denied operations the default planning assumption. Western procurement offices now write electronic-warfare resilience into requirements, which disqualifies most commercially-derived autonomy stacks and advantages the very small number of systems with jamming-tested flight hours. Shield AI has 170+ such sorties; almost no competitor can make the claim, and the requirement is only tightening.

Structure and entry price. We are buying Series G common through a two-layer SPV. Layer 1 charges a one-time 5.0% broker fee on committed capital ($100,000 on the $2M block) and no management fee or carry; Layer 2 charges 1.5% per annum on committed capital plus 15% carried interest on profit above cost. The price to the seller is $153.68, the Series G reference price, so we pay no premium to the last round. Grossing that up for the one-time fee gives an all-in cost of $161.77 per share, and that is the figure the model now uses: it implies an effective entry valuation of $13.37B, or 24.8x FY2026E revenue against the 23.5x headline. Secondary platforms are quoting $148-$168 (Forge ~$164, Hiive ~$154, Nasdaq Private Market ~$148), so we transact inside the observable market range rather than above it. $1.9M of net capital buys 12,363 shares: 0.0150% at entry and 0.0142% at exit.

Comparables and the exit multiple. Rather than a flat blended average, the model regresses NTM EV/Revenue on forward revenue growth across nine comparables: Palantir 37.6x at 71% growth, Anduril 14.2x at ~95%, Kratos 8.1x at ~20%, AeroVironment 4.6x at ~15%, and the primes (LHX 3.0x, RTX 2.5x, NOC 2.0x, LMT 1.6x) at low-single-digit growth. The fit is Multiple = 1.94 + 26.38 x Growth, R-squared 0.63. The multiple is read off the fitted line at each scenario’s own exit-year growth rate, so deceleration is penalised automatically rather than assumed away. The fit’s weakness deserves naming: Anduril, the closest private comparable and a faster grower, sits well below the fitted line, so the regression prices Shield AI above the one live private mark we can actually observe.

Each case runs its own growth path off the $540M FY2026E base. Deceleration is expressed as a relative haircut to the prior year’s growth rate.

Liquidation preference waterfall. We are buying common, so the model now runs the stack explicitly rather than assuming it away. Blackstone’s $500M fixed-return preferred accretes at an assumed 12% per year, is paid senior and does not participate in upside: $881M in the conservative case, $787M in the base and $703M in the upside. The prior preferred stack, estimated at $2.9B, carries a 1x non-participating preference and elects to convert in all three cases because pro-rata common value comfortably exceeds 1x. The practical effect is therefore modest, a 3.9%, 1.0% and 0.7% haircut to proceeds respectively, because these exit values dwarf the preference, and common is only structurally impaired below roughly $3.8B of exit value.

To ensure that this price per share is defensible, anchoring exit multiples to Anduril’s observable 14.2x mark yields 3.8x net MOIC and 39.5% net IRR in the base case, and 2.6x net with 20.7% net IRR in the conservative case. $12.7B on $540M of FY2026E revenue is 23.5x forward against Anduril at 14.2x on its May 2026 mark, and Anduril was reported in late July 2026 to be raising near $100B, which on ~$4.3B of 2026 revenue is a materially higher multiple than our comp set assumes. The closest comparable has re-rated upward since this round was priced, not downward. Palantir at 37.6x shows what public markets pay for defense software growing at 71%. Shield AI is growing faster than Palantir, with software mix moving from ~30% toward a targeted ~50%, and is priced between the two. Most importantly, we are paying the last-round price for a company whose single most valuable asset, the USAF CCA autonomy franchise, was awarded in production form three months after that round was priced.

The March 2026 Series G, $1.5B at a $12.7B post-money valuation, was led by Advent International and co-led by the Strategic Investment Group of JPMorganChase’s Security and Resiliency Initiative. Funds managed by Blackstone provided $500M of fixed-return preferred equity plus a $250M delayed-draw facility, taking the round to roughly $2B. Existing investors including Snowpoint Ventures, InnovationX, Riot Ventures, Disruptive and Apandion participated. Advent Chairman David Mussafer joined the board and JPMorganChase’s Todd Combs joined as a board observer. Advent has separately committed up to $1B to next-generation defense technology, so this is a platform position for them rather than a one-off. Per-investor ownership stakes have not been disclosed to us.

The signal quality is high but different in character from a classic Tier 1 venture syndicate: a global buyout firm leading, the largest U.S. bank co-leading through its national-security investment vehicle, and Blackstone underwriting the credit-like tranche. That is what a pre-IPO capital structure looks like when the sponsor set is underwriting the durability of government revenue rather than early-stage optionality.

Path to liquidity: no IPO is expected in 2026 and management is prioritising private growth. The primary exit is a 2028-2031 IPO contingent on reaching ~$1B of revenue and a receptive market, consistent with our 3-5 year hold period. Interim liquidity is available through secondaries at $148-$168. A strategic acquisition is unlikely at a $12.7B valuation.

Rich valuation with unproven profitability. At ~$12.7B the company is marked at ~42x trailing and ~23.5x forward revenue versus low-single-digit multiples for the primes; 2025 profitability targets were reportedly missed and margin data is not available to us. Mitigants: we enter at the last-round price with no premium; the exit multiple is derived from a growth-based regression rather than an assumed re-rating; and the base case clears the 25% hurdle by more than twenty points at 46.5% net IRR, while anchoring to Anduril’s observable 14.2x mark instead of the regression still yields 3.8x net and 39.5% net IRR. Anduril’s reported move toward ~$100B suggests category multiples are expanding, not compressing. The low-probability conservative case, which pairs 55% decaying growth with a 20% multiple haircut, returns 1.4x net and 6.7% net IRR and fails our hurdle, so the downside is now modelled rather than assumed away. Residual risk is real: a reset into the Kratos/AeroVironment band (5-8x) combined with a larger growth miss would impair the position outright.

Government and appropriations concentration. Revenue remains heavily tied to U.S. DoD and allied appropriations, exposed to continuing resolutions, shutdowns and program restructurings of the kind that turned Replicator into DAWG. Mitigants: more than 50% international revenue; multi-year vehicles (the $198M Coast Guard IDIQ and the up-to-$800M Navy ISR vehicle); a CCA autonomy award that is a production line item rather than an R&D study; and localized manufacturing via JSW in India that de-risks offset and export politics.

Competitive commoditisation by Anduril. A rival roughly 5x larger, with its own autonomy stack in Lattice, can pressure price and in the worst case relegate Shield AI to a software subcomponent inside someone else’s platform. Mitigants: the airframe-agnostic strategy means Hivemind already flies on Anduril’s YFQ-44A Fury, so Shield AI participates in Anduril’s success; Honeywell and L3Harris are building on the Hivemind SDK; and switching costs are high once autonomy is certified into a program of record. Note the asymmetry, though: if Shield AI becomes purely the autonomy layer, revenue per program shrinks even as strategic relevance holds.

X-BAT execution and capital intensity. X-BAT is a clean-sheet AI-piloted fighter with minimal revenue today, competing for dollars against companies with far larger balance sheets, while V-BAT production scaling consumes cash simultaneously. Mitigants: the GE nozzle integration and engine light-off in July 2026 materially de-risk the propulsion path; the round was sized (~$2B including the delayed-draw facility) to fund it; and X-BAT is underwritten as upside rather than base case, so failure does not break the thesis.

Structural and liquidity risk in this specific instrument. We hold common through two SPV layers, junior to Blackstone’s $500M fixed-return preferred and the entire prior preferred stack, with no information rights, no board access, a 5% entry fee drag and no liquidity until an IPO at least two years away. Mitigants: the secondary market is active enough to provide interim marks and plausibly interim exits; the fee load is front-weighted and now sits inside the $161.77 all-in entry price; Layer 1 charges no ongoing fee or carry; and the preference stack is explicitly modelled, with the senior Blackstone tranche costing common only 0.7% to 3.9% of proceeds across our three cases and common impaired only below roughly $3.8B of exit value.

Tail risk we cannot hedge. A high-profile mishap involving an autonomous weapon system, or an export-control or ITAR violation in a fast-growing international channel, could trigger contract suspension or political blowback. Mitigants: A-GRA compliance, an established DoD compliance posture, and government-to-government sales structures. Low probability, high severity, and only partially mitigable.