Impulse Space has raised $500 million with a clear message that cuts against a growing trend in tech: the company wants to build rocket engines and other flight-critical hardware by expanding its human engineering bench, not by betting everything on “AI-first” autonomy. In a conversation with TechCrunch, Impulse president Eric Romo framed the funding as fuel for a very old-fashioned kind of progress—designing, testing, breaking, learning, and repeating—while acknowledging that modern tools, including software and machine learning, can still play a role. But the center of gravity, he argued, remains people.
That distinction matters, because in aerospace, the difference between a promising prototype and a reliable engine is rarely a single breakthrough. It’s a long chain of decisions made under constraints: materials that behave differently at temperature, manufacturing tolerances that stack up in unexpected ways, sensors that drift, valves that stick, combustion instabilities that appear only after you’ve pushed the system into the regime where they matter. AI can help with parts of that chain—data analysis, anomaly detection, optimization of certain parameters—but it cannot replace the engineering judgment required to define what “good” looks like, to interpret failure modes correctly, and to decide what to change next when the data is incomplete or misleading.
Impulse’s funding announcement is therefore less about rejecting AI and more about rejecting the idea that AI can substitute for the human work of turning physics into flight hardware. The company’s approach reflects a broader reality in the space industry: autonomy is attractive in theory, but the path to autonomy runs through the same bottlenecks as everything else—test capacity, instrumentation, manufacturing quality, and experienced teams who know how to translate results into design changes.
A rocket engine startup is not a software company with a deploy button. It’s a manufacturing and test company with a software layer. Even if you have sophisticated simulation, you still need physical validation. And physical validation is expensive, slow, and unforgiving. That’s why hiring is not just a growth strategy for Impulse—it’s a way to increase throughput across the entire engineering pipeline. More engineers means more parallel work: multiple subsystems designed concurrently, more test campaigns planned and executed, more integration work done without waiting for a single team to become a bottleneck, and more time spent iterating rather than merely surviving the next milestone.
Romo’s comments point to a specific kind of expertise that is hard to automate: the ability to make sense of messy reality. In rocket development, the “truth” is often distributed across hardware, telemetry, and the test environment itself. A sensor reading might be correct but misinterpreted because the calibration assumptions were wrong. A performance shortfall might be real but caused by something upstream—plumbing, flow conditioning, thermal gradients—rather than the component you initially suspect. Engineers learn these patterns over years, and they develop instincts for what to trust and what to question. That instinct is not a feature you can simply bolt onto a model.
Impulse’s $500 million round also signals investor confidence in the idea that scaling a physical engineering organization can be a competitive advantage. In many industries, capital can be used to buy speed through automation. In aerospace, capital buys speed only if you can convert it into test time, manufacturing capacity, and engineering bandwidth. Hiring is one of the most direct levers. It expands the number of people who can write requirements, build test plans, design fixtures, interpret results, and manage the iterative loop that turns prototypes into engines that can survive repeated use.
There’s another nuance here: “AI-first” autonomy is often discussed as if it’s a single capability. In practice, autonomy is a stack. It includes perception, decision-making, control, verification, and safety constraints. Each layer has to be validated in the real world, and each layer introduces new failure modes. In a rocket engine context, the stakes are extreme. If an AI system makes a wrong call during a test or a firing sequence, the cost isn’t just downtime—it can be hardware destruction, schedule slips, and sometimes safety risks. That pushes companies toward conservative architectures where humans remain responsible for defining objectives and approving changes, even if AI assists with analysis.
Impulse’s stance can be read as a pragmatic response to this verification problem. The more autonomy you introduce, the more you must prove that it behaves correctly across edge cases you may not have encountered yet. For a startup, that proof burden can become a distraction from the core mission: building engines that work. By focusing on human-led engineering, Impulse is likely trying to keep the development loop tight and interpretable. When something goes wrong, the team needs to understand why quickly enough to change the design before the next test window closes.
The company’s emphasis on human talent also highlights a less glamorous but crucial aspect of rocket development: integration. Many technologies can be designed in isolation. Rocket engines are not isolated. They interact with feed systems, turbomachinery, ignition systems, thermal management, structural loads, and control electronics. Integration is where theoretical performance meets practical constraints. It’s also where “AI can do it” narratives often break down, because integration requires coordinating multiple disciplines and reconciling conflicting requirements. An AI tool might optimize a parameter within a subsystem, but it can’t easily arbitrate tradeoffs across the entire system unless it has a complete, accurate model of everything—and those models are never complete early in development.
Impulse’s hiring focus suggests the company is investing in the connective tissue between disciplines: systems engineers who can translate between mechanical design, combustion behavior, controls, and manufacturing realities. It also implies investment in test engineering—the people who build the measurement infrastructure, ensure data quality, and create repeatable procedures. In many organizations, test engineering is treated as support. In rocket development, it’s a primary driver of learning. Better tests produce better data; better data produces faster iteration. If you want to accelerate progress, you don’t just need more designers—you need more people who can run the experiments that teach you what to do next.
There’s also a cultural dimension. When a company says it’s hiring people rather than relying on AI autonomy, it’s making a bet on organizational learning. Teams that iterate successfully develop shared mental models: how the engine behaves, what failure signatures look like, which anomalies are noise and which are signals, and how to structure experiments to reduce uncertainty. Those mental models are built through collaboration and repetition. They’re not easily transferred from a model trained elsewhere because the relevant data is unique to your hardware, your manufacturing process, and your test environment.
This is where Impulse’s message becomes more than a slogan. It’s a statement about the nature of engineering progress in complex physical domains. In software, you can often ship, observe, and patch quickly. In aerospace, you ship by building hardware that survives. The feedback loop is slower, so the quality of each iteration matters more. Human teams can compensate for incomplete information by using judgment, cross-checking hypotheses, and designing experiments that isolate variables. AI can assist, but it still depends on the human-defined structure of the problem.
At the same time, it would be inaccurate to interpret Impulse’s position as anti-technology. Modern rocket development already uses advanced computational tools: CFD for fluid dynamics, FEA for structural analysis, digital twins for certain subsystems, and data-driven methods for diagnostics. Machine learning can be useful for identifying patterns in telemetry, predicting component wear, or flagging anomalies that might indicate a developing issue. The key difference is whether AI is treated as the driver of the engineering process or as a tool within it.
Romo’s framing suggests Impulse sees AI as subordinate to the engineering workflow. The company is likely using software and analytics to improve efficiency, but it’s not outsourcing responsibility for design decisions to an autonomous system. That’s a subtle but important distinction for investors and customers. In a sector where reliability is everything, accountability matters. If a model recommends a change that later proves harmful, who owns the decision? If the system fails, how do you explain the failure and prevent recurrence? These questions are not just legal—they’re technical. They determine how quickly you can recover and how confidently you can scale.
The $500 million figure also raises the question of what Impulse intends to do with the money beyond hiring. While the announcement emphasizes people, large rounds typically fund multiple categories: facilities, test infrastructure, supply chain expansion, and long-term development programs. Hiring alone doesn’t create test capacity. If Impulse is serious about accelerating iteration, it likely needs to expand the physical infrastructure that supports testing and manufacturing. That could include additional test stands, improved instrumentation, upgraded data pipelines, and investments in production tooling that reduce variability.
In aerospace, variability is the enemy of predictability. Even small differences in manufacturing can shift performance and stability margins. A team that understands how to manage variability—through process control, inspection, and iterative refinement—can turn uncertainty into a manageable parameter. That again points back to human expertise. Process engineers, quality specialists, and manufacturing technicians are essential to translating designs into repeatable outcomes. They are also the people who can interpret why two “identical” parts behave differently and what to change to reduce the gap.
Impulse’s approach may also reflect a strategic view of talent as a moat. Many startups can access off-the-shelf software. Fewer can assemble a team with deep experience across combustion, turbomachinery, structures, controls, and test operations. Talent is difficult to recruit quickly, and it’s even harder to retain when competitors are also scaling. By raising a large round, Impulse can compete for that talent and build a team capable of executing multiple development tracks simultaneously.
There’s another angle: the space industry is increasingly crowded with companies that claim to be “AI-native” or “autonomy-first.” Some of those claims are marketing. Others may reflect genuine progress in specific areas, such as mission planning or ground operations. But rocket engines are not a mission-planning problem. They are a physics problem with a manufacturing and testing backbone. If Impulse’s investors believe the company can win by focusing on the fundamentals—engineering execution,