Daedalus built automated factories for paying customers — and still went insolvent

Daedalus, a Karlsruhe startup founded by Jonas Schneider — who spent three years as one of the first engineers inside OpenAI's robotics team before leaving in 2019 — built fully automated precision-manufacturing lines that ran for paying customers in semiconductors, defense, energy, e-mobility and medical devices. It raised an $11.5M seed in 2021 led by Addition with Khosla Ventures alongside, and a $21M Series A in February 2024 led by NGP Capital with both earlier backers returning — more than $40M in total. This month it entered insolvency proceedings.

The bet was the hardest version of physical AI: not selling robot arms into factories, but owning the factory itself — the capital expenditure, the hardware, the software and the operating risk. It was not vapourware; customers across five industrial verticals signed on. What never arrived was a third round. As the RobotAIGeek analysis puts it, capex at this scale is balance-sheet risk rather than technology risk, and venture capital — even from patient firms — wants years, not decades, to see a return.

The same week Daedalus's insolvency became public, Brooklyn warehouse-robotics startup Ultra closed a $50M Series A led by Framework Ventures for its dual-arm OP1 robot, which has packed more than 500,000 orders for paying customers — built on licensing Physical Intelligence's foundation model instead of owning an AI stack. The contrast cuts both ways: Cyngn, which also owns no foundation model, disclosed a thin revenue base and an August 2026 cost reorganization. Renting intelligence trades capex risk for dependency risk.

More than $40M raised across two rounds; insolvency proceedings despite running production lines for paying customers.

Owning the factory made Daedalus the most capital-intensive version of the physical AI bet — plant, hardware, software and operating risk on one balance sheet.

Revenue from five industrial verticals never outran the capex, and no third round appeared in the two years after the February 2024 Series A.

Venture structures are built for years-scale returns, not the decade-long funding profile a self-owned factory network needs.

Deployment is not a business model: owning the factory converts a technology bet into a balance-sheet bet, and venture timelines don't fund decade-long infrastructure.

Insolvency proceedings opened in October 2026, per court filings cited in trade coverage. The analysis frames the deeper signal: capital is now rewarding 'which layer of the stack you refuse to own' — Physical Intelligence accumulates leverage across many hardware customers without bearing any single customer's capital expenditure. What it flags for next year is whether the companies quietly moving capex onto others' balance sheets — and the suppliers accumulating it — get priced honestly for that risk.