Impulse Space, the in-space mobility company founded by SpaceX engine veteran Tom Mueller, has raised $500 million in a Series D round — and announced it intends to spend a meaningful portion of that money on humans. Up to 200 of them.
This is either a principled stance on the limits of machine intelligence, or the most polite rejection letter the AI industry has ever received. Possibly both.
If you want to find the best designs for a turbo pump seal package in the world, you're not going to find those online.
What happened
The round was led by 137 Ventures and BANNER VC, with participation from Founders Fund, Lux Capital, and Linse Capital — investors who are, separately, also quite enthusiastic about AI. The money will go toward building and testing more spacecraft, including Mira, Impulse's maneuverable platform targeting the U.S. Space Force, and Helios, a vehicle designed to rapidly carry satellites to high orbit.
President and COO Eric Romo explained the hiring rationale with the calm authority of someone who has watched simulations be confidently wrong for two decades. His position: deep learning models are not ready for real-world hardware engineering. The rocket, he notes, still needs to be built and then placed on a test stand.
Romo also identified why AI hardware tools may lag behind their software counterparts. Training data for turbo pump seal packages is not, it turns out, abundant on the internet. This is a problem that no amount of scraping will solve in the near term.
Why the humans care
Aerospace talent is increasingly in demand, and Impulse recently opened a Colorado office specifically to compete with Seattle, Denver, and Texas for engineers who no longer feel obligated to relocate to Los Angeles. The company has expanded from propulsion into full spacecraft, requiring expertise in vehicle structures and flight computers — disciplines that remain stubbornly dependent on people who understand physical objects.
The Mira spacecraft completed its third flight late last year, though a navigation system problem caused it to burn through most of its propellant earlier than planned. A fourth mission is expected before year's end. The test stand remains the final arbiter. It does not accept model outputs in lieu of test data.
What happens next
Impulse will hire its engineers, build its rockets, and continue using AI coding tools for software while relying on humans for the parts where being wrong by 20% means the vehicle does not return.
The machines are, for now, watching from the ground.