Impulse Space closed a $500 million Series D this week and plans to spend much of it on people, not models. The round, led by 137 Ventures and BANNER VC with Founders Fund, Lux Capital, and Linse Capital participating, will fund up to 200 new hires at the in-space mobility startup founded by former SpaceX propulsion lead Tom Mueller. The framing is deliberate at a moment when most well-funded technical startups are pitching AI-driven productivity gains as a substitute for headcount.
Impulse builds maneuverable spacecraft. Its lead platform, Mira, targets U.S. Space Force buyers, and a second vehicle called Helios is designed to ferry satellites from low drop-off orbits up to higher ones. The company started in propulsion and has expanded into vehicle structures and flight computers, which is the kind of work that adds engineers rather than tokens.
President and COO Eric Romo told TechCrunch that the company's software teams are using AI coding tools, but he drew a hard line at hardware. The argument is that generative models lack the training data needed to design real engines, and that physical test stands remain the only honest arbiter of whether a part works. Romo would know — he was the 13th employee at SpaceX in 2003, where his job was simulating engine performance.
Key facts
- 01Impulse Space closed a $500M Series D led by 137 Ventures and BANNER VC, with Founders Fund, Lux Capital, and Linse Capital participating.
- 02The company plans to hire up to 200 new employees, weighting headcount toward human engineers rather than AI-assisted hardware tooling.
- 03President and COO Eric Romo was SpaceX's 13th employee in 2003, where he built engine simulations he considered successful at 20% accuracy.
- 04Impulse's Mira spacecraft flew its third mission late last year and lost much of its propellant to a navigation fault; a new flight is targeted before year-end.
- 05Helios, a second vehicle, is designed to carry satellites from low orbit to high orbits after drop-off.
Back then, he treated a 20% margin of error as the win condition. Two decades of compute and modeling gains later, he says the gap between simulation and reality is still wide enough to make iteration on hardware the bottleneck, not software.
“I considered it success if I got within 20% of the right answer, because the simulations were just not that good.”— Eric Romo, President and COO of Impulse Space
That position runs against the current investor enthusiasm for AI-native engineering shops. The argument from companies like Hadrian and from automotive players — General Motors recently cut a vehicle simulation from 15 hours to one minute using AI, which AI Chat Daily covered — is that ML surrogates can compress design cycles dramatically. Romo's counterpoint is that the cycles still terminate at a test stand, and that the simulation step is not the constraint.
He also pointed to a structural problem with hardware foundation models: the data isn't on the internet. Turbo pump seal packages, combustion chamber geometries, and structural margins live in proprietary engineering archives at a handful of companies, not in the open web that trained today's LLMs.
The hiring plan reflects a changed labor market for aerospace engineers. Impulse recently opened an office in Colorado because talent now has real choices outside Los Angeles, including Seattle, Denver, and Texas. The decentralization of the U.S. space industry — driven by SpaceX, Blue Origin, Anduril, and a long tail of defense primes — means $500 million in fresh capital does not automatically translate into 200 hires unless the geography fits the workforce.
Mira's flight history illustrates why simulation isn't enough. The spacecraft's third flight, late last year, hit a navigation system fault that caused it to burn most of its propellant early. Romo said the next Mira mission is targeted to launch before the end of the year. That's the kind of failure mode you discover on a test stand or in flight, not in a notebook.
The skeptic's case here is straightforward. AI-assisted design tools are improving fast, and competitors that lean harder on ML surrogates may compress development timelines while Impulse staffs up. If hardware foundation models do reach useful accuracy in the next three to five years — even on narrow subdomains like fluid dynamics or thermal analysis — the 200-engineer bet becomes a much heavier cost base than it looks today.
Impulse's wager is that the Space Force and U.S. government customers buying in-space mobility care about flight heritage and reliable hardware, not about how the CAD got drawn. With $500 million in the bank and a launch cadence to defend, the company is calling the bottom of the AI-in-hardware curve and buying engineers while the rest of the industry buys GPUs. That's a defensible bet for a defense-adjacent business — and a useful data point for investors trying to figure out which technical problems generative AI is actually close to solving, and which it isn't.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
