JuliaHub Secures $65M in Series B Funding to Scale Scientific Machine Learning Platform

A certain tension is in the room right now. AI is everywhere, talking loud, moving fast, breaking things it does not fully understand. And then there’s JuliaHub, moving different, building systems that do not just generate answers but actually hold up when reality pushes back. That contrast just got sharper with a $65M Series B, and you can feel exactly where this is headed.

Out of Cambridge, Massachusetts, JuliaHub has been building a different kind of intelligence stack, one rooted in Scientific Machine Learning where physics is not a suggestion, it is law. Dr. Viral B. Shah (CEO) and Dr. Jeff Bezanson (CTO), alongside co-founders Prof. Alan Edelman (Chief Scientist), Stefan Karpinski (CPO), Deepak Vinchhi (COO), and Keno Fischer (CTO, R&D), did not start with noise. They started with the Julia language and a refusal to accept the tradeoff between speed and usability. That “two-language problem” they solved years ago is now showing up like compound interest in an AI cycle that suddenly demands both precision and scale.

Dorilton Capital led the round, with General Catalyst, AE Ventures, and Bob Muglia stepping back in. That lineup is not chasing headlines. That is capital that understands modeling a jet engine, a drug interaction, or a power grid is a different sport entirely. When the environment pushes back, your math better push harder.

The headline product, Dyad 3.0, leans all the way into that reality. JuliaHub calls it an agentic AI platform for hardware engineering. Strip the label and you get something more interesting. Machines that can reason through equations, simulate outcomes, and stay grounded in physical constraints while doing it. Not autocomplete for engineers. Actual computation that respects the rules of the universe.

Industries like aerospace, pharma, and energy are tuned in for a reason. When your margin of error is measured in lives, billions, or decades, “close enough” is not a strategy. JuliaHub is selling discipline. AI that learns from data but answers to physics. That balance is where trust starts to form, and where budgets start to move.

The business lesson is sitting right there, no translation needed. They went deeper while others went louder. Built with a technical community that does not clap for shortcuts. Let the market evolve into the problem they were already solving. Timing meets substance and suddenly the room feels different.

Now the capital is in, the product is iterating, and the signal is getting harder to ignore. The next phase of AI is not about what systems can generate. It is about what they can validate, simulate, and stand behind when the real world starts asking questions.