A $500 Million Bet That Humans Still Beat AI at Building Rockets

In a year when AI companies are raising billions to replace human workers, Impulse Space just raised $500 million to do the opposite. The rocket engine startup, founded by SpaceX's thirteenth employee and engine guru Tom Mueller, announced a Series D round led by 137 Ventures and BANNER VC, with participation from Founders Fund, Lux Capital, and Linse Capital. The money will go toward hiring as many as 200 new employees — engineers who build physical things in the real world.

This is not a retrograde decision made by people who do not understand AI. Impulse's president and COO, Eric Romo, was the 13th employee at SpaceX in 2003, where his job was creating computer simulations of rocket engine performance. He knows exactly what computational tools can and cannot do. And his assessment is blunt: when it comes to solving engineering problems in physical systems, AI is not ready.

"I considered it success if I got within 20% of the right answer, because the simulations were just not that good," Romo said of his early SpaceX work. "They've improved, but they've not improved that much, and so there's not really any substitute for designing the thing, analyzing the thing, building it, and then getting it on the test stand."

The data problem AI cannot solve Romo's reasoning gets at a structural limitation of current AI systems that most coverage of the AI revolution ignores. Large language models are powerful because they were trained on vast quantities of text and code available on the internet. But the internet does not contain the world's best turbo pump seal designs. Proprietary engineering knowledge — the kind that determines whether a rocket engine blows up or delivers a satellite to orbit — lives inside the heads of experienced engineers and the proprietary databases of aerospace companies.

"If you want to go find the best designs for a turbo pump seal package in the world, you're not going to find those online," Romo points out. This is why AI coding assistants can help Impulse's software teams but cannot replace the mechanical engineers who design the spacecraft itself. The training data simply does not exist in sufficient quantity or quality for AI to learn from.

This is a crucial distinction that the current AI hype cycle often glosses over. AI excels at tasks where the relevant information is digitally abundant: writing code, generating text, processing documents, analyzing data. It struggles at tasks where the relevant information is locked in physical experience, proprietary knowledge, or situations where the only way to learn is to build something and test it until it either works or explodes.

What Impulse is actually building The hiring push is driven by Impulse's expanding product lineup. The company started with propulsion systems and has evolved to build complete spacecraft. Its Mira vehicle, designed for highly maneuverable in-space operations, has completed three flights. The most recent flight encountered a navigation system problem that caused the vehicle to expend much of its propellant early on — precisely the kind of real-world failure that simulations struggle to predict and AI cannot currently debug.

Impulse is also building Helios, a vehicle designed to carry satellites rapidly to high orbits after they are dropped off closer to Earth. Both products target U.S. Space Force buyers at a moment when the Pentagon is pouring money into space mobility and national security applications.

The broader context is telling. The Series D comes amid a surge of investor interest in space and defense tech, driven by U.S. government spending on national security problems and anticipation of SpaceX's IPO. Impulse is positioning itself as a nimble alternative to larger, slower-moving prime contractors — and it is betting that the nimbleness comes from having more talented humans, not more capable AI.

Why this matters beyond aerospace The Impulse story is a useful corrective to the narrative that AI is about to make human expertise obsolete across every domain. In fields where the relevant knowledge is digitally encoded and abundant — software development, legal document review, content creation — AI is already transforming how work gets done. In fields where the relevant knowledge is physical, proprietary, or experiential, the human bottleneck remains.

This is not a permanent condition. AI will eventually get better at physical engineering as simulation tools improve and more engineering data becomes digitized. But the timeline is longer than the hype suggests, and the path is more uncertain. In the meantime, companies like Impulse that invest in human talent in domains where AI still falls short may have a durable competitive advantage — not because they are anti-AI, but because they are clear-eyed about what the technology can and cannot do today.

What this means for you If you work in software, AI tools are already changing your job — use them. If you work in physical engineering, manufacturing, or any field where the best knowledge lives in people's heads rather than on the internet, your expertise is more valuable than the hype cycle suggests. The companies winning investment dollars right now are not just the ones building AI — they are also the ones building things AI cannot yet build. The $500 million flowing into Impulse is proof that the market still knows the difference.