Start a trial

Operational notes Observatory

Anduril replaces Microsoft: the $22 billion headset

5 min read

Period interior of a munitions factory: workbenches and long rows of stacked metal crates
The modern arsenal is built before the order arrives: the off-the-shelf product overturns the balance of power in procurement.

There was a handover, just over a year ago now, that says a great deal about how states are changing the way they buy technology. In February 2025 Microsoft — a $3 trillion colossus by market capitalisation — announced it would cede the lead on IVAS to Anduril Industries, a company founded in 2017: the US Army’s augmented-reality military headset programme, with a potential value of $22 billion. This was not a subcontract: the giant stepped aside and the start-up took the helm, with Pentagon approval following in spring. For anyone watching the critical-technology market — even from Europe, even outside defence — this is a textbook case. Let’s set out the facts, then draw the lesson.

The facts, in order

  • The original contract: in 2021 the US Army awarded Microsoft the IVAS programme — augmented-reality headsets derived from HoloLens, with night vision and integrated tactical data — worth up to $22 billion over ten years. Difficult years followed: delays, versions rejected by soldiers for causing nausea and headaches, and funds frozen by Congress pending proof of maturity.
  • The handover: on 11 February 2025 Microsoft and Anduril announced the agreement — Anduril takes on production, hardware and software development, and delivery schedules; Microsoft remains as the “preferred” cloud provider on Azure. In April 2025 the Army approved the transfer.
  • The factory: in January 2025 Anduril announced Arsenal-1, a “hyperscale” plant near Columbus, Ohio: $1 billion in the company’s own funds, up to 5 million square feet at full capacity, around 4,000 jobs projected, and production of autonomous systems — the Fury, Roadrunner and Barracuda aircraft — with start-up targeted for summer 2026.
  • The capital: in June 2025 the company raised $2.5 billion in a round led by Founders Fund, with a valuation of $30.5 billion — more than doubled in a year, against 2024 revenues of around $1 billion.
  • The new competition: in September 2025 the programme, renamed Soldier Borne Mission Command, entered a competitive phase: the Army awarded over $350 million in prototyping contracts to Anduril ($159 million, teamed with Meta, Qualcomm and Gentex) and to rival Rivet ($195 million). There is no definitive winner yet: the decision will be made in the field.

Lesson one: “products, not programmes” — whoever pays for development sets the timetable

The traditional defence-procurement model is the programme: the state defines the requirements, funds the development, and pays costs plus a margin. Anduril has turned this on its head: it develops off-the-shelf products with private capital — its own and its investors’ — and sells them once they work. For buyers, the advantages are real: shorter timelines, development risk shifted onto the supplier, list prices rather than cost reimbursement. But the flip side needs careful reading: intellectual property stays with the supplier, the supplier decides the roadmap, and a closed catalogue is a new form of lock-in. The right question does not change from classic contracts — if anything it matters more: what is left for the buyer — data, interfaces, documentation — if the relationship ends? It is the criterion by which we design our platforms: the product accelerates, but the architecture must remain open and reversible. It is no accident that the US Army, too, has put the programme back out to competition between two rival suppliers.

Lesson two: human control is a design requirement, not a slogan

Anduril builds autonomous weapon systems, and criticism is not in short supply. The UN Secretary-General has called lethal autonomous weapons “politically unacceptable and morally repugnant”; the international Stop Killer Robots campaign has been calling since 2012 for a ban on systems lacking “meaningful human control”. The company’s position deserves to be reported with the same precision: founder Palmer Luckey maintains that every Anduril system has a “kill switch” allowing the operator to intervene, that “responsibility must remain with human beings”, and that the real choice is not between autonomous weapons and controlled weapons, but between intelligent weapons and stupid ones — a landmine, he argues, cannot tell a school bus from an armoured vehicle. It is not for us to say who wins the ethical debate. But for anyone deploying AI in critical processes — civilian or military — the operational lesson is clear: human control must be designed, documented and verifiable — who authorises, how it is interrupted, who is accountable. In Europe this is not optional: the law requires it for high-risk systems.

Lesson three: for European suppliers — and buyers — the contest has already begun

As Europe rolls out rearmament plans worth hundreds of billions, the Anduril model sets the new benchmark: private capital, ready-made catalogue products, factories built ahead of orders. European suppliers will have to decide whether to chase that model or compete on the guarantees it does not offer: data sovereignty, verifiable supply chains, demonstrable human control, contractual reversibility. And European public buyers — defence ministries and agencies included — would do well to bring into their tender specifications the questions the IVAS case has made urgent: where things are made, who owns the intellectual property, what happens if the supplier changes priorities or ownership.

What to do, if you buy critical technology

  1. Distinguish product from programme in the tender specification: an off-the-shelf product should be assessed on evidence, not promises — demand operational trials before committing to a multi-year deal.
  2. Clarify ownership of data, models and interfaces: what remains yours, in what format, with what documentation.
  3. Write human control into the contract: who authorises the system’s actions, how it is interrupted, who is accountable for it — and how this is demonstrated to an auditor.
  4. Assess the supplier’s production capacity as a risk factor: where the factories are, what the bottlenecks are, what the alternatives are.

Assessing a supplier that arrives with the product already built? Half an hour with one of our experts to work out what to ask before you sign.

Sources