Flow Engineering raises $50M at $750M valuation in Series B led by Valar and Atreides

Flow Engineering, a three-year-old San Francisco startup, announced yesterday it has closed a $50 million Series B at a $750 million valuation. The round was co-led by Antonio Gracias of Valar Equity Partners — a fund best known for backing Elon Musk's companies, chiefly SpaceX — and Gavin Baker of Atreides Management, a hedge fund that also holds stakes in Musk-affiliated ventures and in AI chipmaker Cerebras. Sequoia Capital, which led the Series A last October, participated again, and former Sequoia partner Roelof Botha joined as a private investor, taking a board seat.

Flow builds AI agents designed to close the loop between CAD drawings, product requirements, simulation results, and physical tests. Instead of engineers manually updating models every time a requirement shifts or a simulation reveals a failure, the agents are meant to align geometry automatically against the spec and incoming data. The company has not published comparative benchmarks against manual workflows or existing tools, so real-world effectiveness still hinges on customer validation.

The customer list carries weight: Anduril, Rivian, Joby Aviation, the GM–TWG Motorsports joint venture (GM PPU), the Rivian–Volkswagen joint venture (RV Tech), and Stoke Space. That companies developing safety-critical systems with long development cycles are willing to adopt an automation tool this early signals an unusual level of trust, even if the internal scope of deployment isn't public.

The new capital is earmarked for expanding the research team and accelerating integrations with widely used CAD and simulation tools. Sequoia's Series A bet, combined with the entry of Valar and Atreides — both deeply exposed to the Musk ecosystem — points to a strategic vector aimed at the SpaceX, Tesla, and Starlink supply chains, though none of those companies appear on the official customer roster. Botha's board seat adds significant organizational experience, but the real test will be whether the agents can shrink design iterations on live projects without introducing silent errors.