Analysis
Three of 2026's largest early-stage rounds have gone to companies selling physical hardware, not software:
- Stoke Space — $1B Series E for reusable rockets
- Castelion — $800M equity + $250M debt at a $13B valuation for hypersonic strike missiles
- Atoms — $1.7B, Travis Kalanick's six-month-old robotics company
None of the three has meaningful commercial revenue. All three raised at valuations and speeds that, three years ago, were reserved for software companies with recurring revenue already growing 3x a year.
“All three raised at valuations and speeds that, three years ago, were reserved for software companies with recurring revenue already growing 3x a year.”
What changed is not the hardware, it's the underwriting
Capital-intensive hardware has always required large checks -- building a rocket or a missile costs real money regardless of the financing environment. What is new is the speed and the founder-driven pricing. Castelion and Atoms both closed their rounds within months of founding or relaunching. Stoke priced its Series E off two years of engineering credibility and a 2027 flight date, not a demonstrated commercial contract. That is the AI-lab financing playbook -- price the team and the thesis, not the trailing revenue -- applied to metal.
Three forces are driving it. First, defense and aerospace budgets are genuinely expanding, and government-adjacent hardware now carries a credible acquirer or customer of last resort in a way consumer hardware never did. Second, a small number of crossover and growth investors -- Andreessen Horowitz, Point72, Sequoia -- have decided capital-intensive physical infrastructure is where the next decade of outsized outcomes sits, and they are moving fast to avoid being priced out the way many funds were on AI labs in 2023 and 2024. Third, founder scarcity: Kalanick, Stoke's Blue Origin-trained team, and Castelion's ex-SpaceX leadership are all repeat operators with track records that let investors skip the traditional proof points.
The risk profile is genuinely different from software
A software company that raises ahead of revenue burns cash on payroll and cloud bills; if the thesis is wrong, the company can pivot cheaply. A hardware company that raises ahead of revenue burns cash on physical test articles, fabrication and flight hardware that cannot be repurposed if the thesis is wrong. Stoke's Nova Pathfinder, Castelion's hypersonic test vehicles and Atoms' robotics platforms are all single-purpose capital, and a failed test is both a technical setback and a cash event in a way a failed software sprint never is.
For funds without deep hardware diligence experience, that is the trap in following this round-sizing pattern into deep tech: the team-and-thesis underwriting that works for AI labs assumes the founder can iterate cheaply on failure, and physical hardware founders cannot. The funds writing these checks -- Andreessen Horowitz, Point72, Sequoia -- have specific partners who have underwritten aerospace and defense before, and that specialized diligence bench is the actual protection this category has against a speculative bubble forming underneath it.
The scoreboard that will settle it
None of this is resolved by the size of a round. Rocket Lab took eight years and dozens of missions to get Neutron to the pad after raising comparably large sums earlier in its history, and Relativity Space has raised billions across a decade without a commercial flight of its flagship vehicle. Stoke, Castelion and Atoms are all still in the phase where the check is bigger than the evidence. The next twelve to eighteen months of test flights, missile trials and robotics demonstrations -- not the next funding announcement -- are what will show whether 2026's hardware round sizes were prescient or just early.
