Defence technology used to be the category Silicon Valley refused to touch. In 2018, Google’s Project Maven contract triggered a revolt: more than 3,000 workers signed a letter of protest, and Google walked away.
That era is over. Defence startups raised more than $14.6 billion through mid-2026, beating 2025’s $9.6 billion full-year record. Helsing closed a $1.8 billion Series E at an $18 billion valuation. Anduril is reportedly chasing a valuation near $100 billion, on par with Northrop Grumman and Lockheed Martin.
The Pentagon’s $200 million Anthropic contract collapsed over two red lines: no autonomous weapons, no mass domestic surveillance. OpenAI signed an equivalent deal within 24 hours, and the same capital, talent and AI vendors now operate in the markets where you hire, buy and plan.
In This Series
- How Defence Tech Went From Taboo to Venture Capital Asset Class: the reversal and the valley of death.
- Is Defence Tech a Bubble? The Anduril $100 Billion Question: the multiples and whether Anduril can displace the primes.
- Why Anthropic Refused the Pentagon Deal and OpenAI Accepted Within 24 Hours: the collapsed contract and what postures mean for buyers.
- Who Is Accountable When AI Picks the Target?: the doctrine shift and who answers for a strike.
- The Defence Tech Hiring Boom Is Reshaping Engineering Salaries and Retention: how mega-rounds reshape salaries and build-vs-buy calls.
How did defence technology become a venture-capital asset class?
Defence technology turned from a sector venture funds avoided into one of their fastest-growing asset classes when three forces converged. The war in Ukraine proved software-defined warfare at scale; the Pentagon began buying finished products off the shelf; and contract revenue became legible to investors. Palantir, which inherited Project Maven, is the long-run proof that defence software scales, and the result is SaaS-like revenue visibility combined with hardware economics.
Your market is thousands of customers renewing annually; the defence buyer is one concentrated, budget-backed customer with multi-year programmes. Software carries 80 percent-plus gross margins; hardware carries manufacturing cost and capital intensity.
The catch is the valley of death, the funding gap between a working prototype and production-scale revenue. Winning a contract does not close it. Shield AI, funded to a $12.7 billion valuation while still pre-Program-of-Record, is the case to watch.
Read the full arc: How Defence Tech Went From Taboo to Venture Capital Asset Class.
Is defence tech a bubble — and what would that mean for you?
Parts of the sector are priced beyond their fundamentals and parts are not. Early-stage defence startups raise at 17–50x forward revenue, multiples that assume flawless execution, sustained budgets and contract-to-production conversion. Anduril’s reported $100 billion target, on par with Northrop Grumman and Lockheed Martin, is the test case for whether venture-backed builders can displace the primes. For you, a supplier priced at bubble-level multiples carries survival risk.
Near $114 billion, Anduril trades at 53 times trailing revenue and 26 times forecast 2026 revenue. Helsing’s $18 billion valuation — 32x forward revenue after a $1.8 billion Series E, Europe’s largest-ever VC round for a German startup, reads as a second centre of gravity in Europe.
The debate has named voices on both sides. Anduril cofounder Trae Stephens warns there is too much venture capital chasing too few deals. The bear case cites crowding in drones and battlefield AI, plus the Space-Eyes SPAC — a $638 million listing on roughly $1 million of revenue.
Dig into the maths: Is Defence Tech a Bubble? The Anduril $100 Billion Question.
What does the Anthropic–OpenAI split mean for your AI vendor choices?
When the Pentagon’s $200 million Anthropic contract collapsed and OpenAI announced an equivalent deal the same day, acceptable-use policy became a commercial variable. Anthropic refused to let its models power autonomous weapons or domestic surveillance; OpenAI accepted the same terms. For your vendor decisions, that posture is a procurement risk and reputational signal — not a footnote — and it can change your supply-chain exposure.
Anthropic was comfortable with nearly all use cases, bar two. The Pentagon insisted on an “any lawful purpose” clause, Anthropic refused, and the DoD labelled it a supply-chain risk, a first for an American company. Judge Rita Lin called the designation an attempt to cripple Anthropic.
Most enterprise AI is one contract away from defence adjacency, because foundation models are dual-use. Due diligence must go past features and price: what does a vendor’s acceptable-use policy bar, and which uses would your engineers refuse to build? After the split, ChatGPT uninstalls jumped and Claude downloads rose.
Read the full story: Why Anthropic Refused the Pentagon Deal and OpenAI Accepted Within 24 Hours.
Who is accountable when AI picks the target?
The Pentagon’s updated targeting doctrine shifts the human role from initiating strikes to monitoring AI-initiated ones in a compressed sensor-to-shooter cycle. The monitor supervises AI-generated target recommendations, real-time environment assessment and collateral-damage checks, with veto authority at machine speed. Accountability is unresolved: DoD Directive 3000.09 and its companion instruction never define “appropriate levels of human judgment” — language widely judged too vague for systems that outrun human intervention.
AI can turn intelligence into options faster than a human staff can, so the Pentagon is redesigning the human role. In places like Ukraine, autonomous engagement is happening despite the stated position.
“Meaningful human control” is the stronger standard campaigners want, but hard to operationalise at machine speed. The international track, from the Political Declaration to REAIM, remains largely nonbinding. Researchers point to self-driving-car hallucinations as a preview of irreversible false positives in targeting, with China and Russia the strategic accelerant.
Read the full background: Who Is Accountable When AI Picks the Target?.
How will the defence hiring boom reshape your engineering team?
Defence mega-rounds — Helsing’s $1.8 billion Series E, Anduril’s reported $100 billion target — convert directly into hiring budgets and equity packages that compete with yours for AI, robotics, systems and platform engineers. The boom is double-edged: it raises raid risk and compensation pressure on defence-adjacent skills, while an ethics-driven cohort refuses defence work entirely, so the market both poaches and repels. Your build-vs-buy and supplier-viability assumptions need re-benchmarking.
Anduril added more than 1,000 employees in nine months and now sits above 6,200. These are the same senior backend, ML and embedded engineers that fintech and applied AI shops want.
Map the skill families you share with the defence cohort and watch offer-accept rates and band drift. The same pull reaches Australia, where tech hiring is rising while attrition stays elevated. A venture-rich defence-adjacent supplier can outbid you for the same engineers, raising your delivery risk.
Read the full analysis: The Defence Tech Hiring Boom Is Reshaping Engineering Salaries and Retention.
Suggested reading order: money, valuation, vendor split, accountability, hiring. Glossary: forward revenue is contracted future revenue, the valley of death is the gap to production revenue, meaningful human control is the accountability standard, and a neo-prime is a venture-backed builder. This cluster uses Australian English and an APAC framing. Start with the question closest to your next decision and follow the cluster links back here for the full story.
Frequently Asked Questions
What was Project Maven, and why did Google’s 2018 walkout over it still matter?
Project Maven was the Pentagon programme that applied AI to surveillance footage for data analysis and target selection, and Google held the original contract. When more than 3,000 employees signed an open letter in 2018 objecting to the work, Google walked away and published principles barring technology that could cause or directly facilitate injury. Palantir took the work over. The walkout defined defence tech’s taboo era and remains the baseline for measuring how far the sector has come.
Didn’t Google ban weapons work after Project Maven?
It did, and the ban no longer stands. Google’s 2018 principles barred technology that could cause or directly facilitate injury to people, but the company has since removed that weapons language from its AI principles and launched Gemini for Government. The reversal tracks the wider market: once the reputational penalty for military work collapsed, even the sector’s most visible holdout softened its stance, and defence now competes with everyone else for the same money, people and vendors.
How much money is flowing into defence tech right now?
More than $14.6 billion globally through the first five months of 2026, already beating the $9.6 billion full-year record set in 2025. Defence now represents roughly 8 percent of all global venture capital, about double its share in 2020. The round sizes match the totals. Helsing raised a $1.8 billion Series E at an $18 billion valuation, and Anduril is reportedly chasing a valuation near $100 billion.
What actually counts as defence tech?
More than weapons manufacturers. The category spans AI analysis software such as Palantir and Helsing, drone and attritable systems builders such as Shield AI and Saronic, robotics ventures, space companies and dual-use AI vendors whose models now sit inside military programmes. The working definition is revenue: if a company’s fortunes depend on a defence budget line, from software prime down to a contract-funded startup, it is part of the sector and competes for the same capital and talent.
My company doesn’t work with the military. Why should I care about defence tech?
Because the boom does not stay inside the defence industry. Defence startups now compete with you for the same senior engineers, the same AI vendors and increasingly the same suppliers, which makes a startup’s valuation an input to your hiring plan and a vendor’s military posture an input to your procurement. A supplier’s survival odds feed your build-vs-buy call too, because a bubble-priced defence-adjacent vendor that fails can take your roadmap down with it.
What is a “prime” contractor, and why does it matter that Palantir is now one?
A prime is the lead contractor the government holds responsible for an entire programme, integrating subcontractors and delivering the finished system, a role historically dominated by Northrop Grumman and Lockheed Martin. The label matters because Palantir reached that position as a software company, moving from the Project Maven handoff to the prime position on the Army’s TITAN ground station. It is the proof point that software can scale inside the defence system.
Is the defence tech boom a US-only story?
No, but the US still sets the scale. Europe now has a genuine second centre of gravity in Helsing, the Munich defence AI company, whose $1.8 billion Series E at an $18 billion valuation was Europe’s largest venture round ever for a German startup and made its three cofounders billionaires. Other allied markets are building sovereign capability programmes of their own, yet none outside the US has yet produced a round of comparable size.
What does the defence tech boom mean for Australian tech companies?
A new competitor for talent and a new customer pool at the same time. US defence startups can hire Australian engineers remotely on US-dollar packages, and local tech hiring has already risen sharply year on year while attrition stays elevated, so engineering compensation pressure is likely to continue. The government has floated a co-investment plan to pull private capital into defence, a signal to treat sovereign capability work as a serious option in your build-vs-buy planning.
Which venture firms are behind the defence tech surge?
The firms that spent a decade avoiding the sector. Founders Fund, a16z, 8VC and General Catalyst all hold the defence thesis, and Founders Fund carries double exposure because its partner Trae Stephens cofounded Anduril, though he now warns that too much venture supply is chasing too few deals. Much of the money behind the drone and battlefield AI crowd, from Shield AI to Saronic and Castelion, sits in less visible funds, but the direction of travel is uniform.
Is it still controversial for engineers to work in defence tech?
Yes, but the controversy has changed shape. The blanket taboo of the Project Maven era is gone, replaced by case-by-case red lines. An ethics-driven cohort still refuses defence work entirely, while mission and national-security relevance pull a growing share of candidates the other way. Anthropic shows where the line now sits, comfortable with nearly all military uses except autonomous weapons and mass domestic surveillance. The open question is no longer whether to engage but where your own line is drawn.
