Our first aerospike rocket thruster. 5 kN (1,125 lbf) kerolox. Aerospikes are among the hardest rocket engines to develop. Few teams have successfully hot-fired liquid-propellant versions, and none has yet flown to space. Although the geometry is radically different from a conventional bell nozzle, the governing physics are the same. That makes aerospikes an ideal fit for Noyron, which generates engines from underlying physical principles rather than from CAD drawings. An aerospike turns the conventional nozzle inside out. Its toroidal geometry eliminates the fixed outer bell wall, which forces a conventional thruster to choose one atmospheric regime. A sea-level bell nozzle sacrifices performance at high altitude. A vacuum-optimized nozzle cannot be operated safely at sea level. Aerospikes are efficient under all atmospheric conditions, making them highly relevant for reusable upper stages that fly back from space to the launch site. Cooling them is the biggest challenge. Our engine uses liquid oxygen to regeneratively cool the inner spike and kerosene fuel to cool the outer shroud. The resulting warm LOX gives the engine a nice Isp boost. We hot-fired this engine in December 2024. Its early success encouraged us to pursue aerospikes as a parallel track across our engine programs.

Clean startup and shutdown. Two things that matter in engine testing. The 6 kN (1,350 lbf) kerolox thruster from our previous post, during a nominal run. Noyron now generates first-time-right kerolox thrusters across a broad range of thrust classes, chamber pressures, materials, and nozzle architectures — from bell nozzles to aerospikes. We are getting close with methane and much larger engines. So much fun.

This 6 kN (1,350 lbf) kerolox thruster was 3D-printed in Inconel 718, a high-temperature nickel-based superalloy. Inconel is considerably stronger and more heat-resistant than the copper alloys we usually use, but it is also far less forgiving because of its much lower conductivity. That makes it a demanding test of Noyron’s thermal models. Copper can spread local hot spots through the chamber wall. Inconel doesn't. If the predicted heat flux is wrong, the wall burns through immediately. Good news: it worked.

One underappreciated aspect of Noyron is the depth of manufacturing knowledge it relies on whenever it generates a new object. Together with our global partners, we constantly explore what is possible in Additive Manufacturing and throughout the production chain, from heat treatment to post-machining. What we learn is fed back into Noyron. By now, Noyron generates designs that can be manufactured successfully on the first attempt, even when they rank among the most complex objects ever attempted in industrial 3D printing. For this 20 kN (4,500 lbf) methalox engine, printed in CuCrZr, we once again relied on Germany’s Aconity3D, a pioneer in metal Additive Manufacturing.

It is always a proud moment when a Noyron-engineered design becomes reality. Standing 1 m (3.3 ft) tall, our massive 200 kN (45,000 lbf) methalox aerospike was manufactured by HBD as a single monolithic part in Inconel 718, a high-strength, high-temperature nickel-based superalloy. This first manufacturing iteration allowed us to validate the production process and demonstrate that such a complex object could actually be built monolithically. The version intended for hot-fire testing will use an evolved design combining additive manufacturing with other production methods. Encoding the manufacturing chain end to end is one of the key challenges of building a Large Computational Engineering Model. Screenshots and renders are great. Reality feels different.

This 75 kN (16,900 lbf) open-cycle kerolox rocket engine is a Noyron-generated reference design. It is suitable for the first stage of a microlauncher or the upper stage of a larger orbital vehicle. We create reference engines across a wide range of configurations to test the robustness and versatility of Noyron’s models. Most of our work today focuses on staged-combustion methalox systems. But when time to market is critical, open-cycle kerolox remains a pragmatic choice. The combustion chamber and injector are designed for additive manufacturing in CuCrZr, a widely available high-temperature copper alloy.

All of our work is built on PicoGK (“peacock”), a compact and robust geometry kernel designed specifically for Computational Engineering. We open-sourced PicoGK to give the broader engineering community a solid foundation for automated engineering. Traditional CAD kernels use boundary representations developed around interactive, human-driven modeling. PicoGK represents geometry on a high-resolution voxel grid. This provides one coherent geometric source of truth and avoids many of the failure modes that can make conventional CAD automation brittle and ultimately unsuitable for this new engineering paradigm.

A cutaway of a work-in-progress, top-mounted oxygen-rich preburner for a methalox FFSC engine, generated by the Noyron Large Computational Engineering Model. Full-flow staged combustion is both the holy grail and the current state of the art in rocketry.

One of our earliest aerospike experiments was this conceptual test part, printed using Fraunhofer IGCV’s experimental dual-metal 3D-printing process. Multi-material manufacturing will radically expand our engineering playbook: virtual alloys, functionally graded materials, mechanisms that exploit different coefficients of thermal expansion, and fully enclosed motors and actuators printed as integrated systems. But manufacturing capability is only half the challenge. Engineering tools must evolve to compute objects whose behavior emerges from the precise spatial arrangement of multiple materials. The voxel-based structure of PicoGK (“peacock”), our open-source geometry kernel, allowed us to define exactly how copper and steel interweave throughout this engine. The part was exhibited at Formnext in 2023.

The other side of the injector head from our previous post. The embossed code links the physical part to the specific Noyron Computational Engineering Model version and parameter set that generated it. Computed hardware should be traceable hardware.

The injector head is one of the most critical parts of a rocket engine. This one was generated by Noyron for our 20 kN (4,500 lbf) methalox engine. Its injector elements must mix the propellants thoroughly for high combustion efficiency, while the injector itself must survive the intense heat of the combustion chamber and maintain its structural integrity under pressure. Noyron’s physics models must predict the state of each propellant accurately. In the methalox injector shown here, the methane is supercritical. Its density is highly sensitive to temperature, so even a small temperature deviation can significantly change injector pressure drop, mass flow, and mixture ratio. Then there are the transient conditions during startup and shutdown. These brief phases are particularly difficult to model and can impose very different conditions from steady-state operation. We test engines across different propellants, scales, and materials to expose Noyron’s models to physical reality. Each hot-fire gives us evidence to make those models more robust.

Post-fire inspection of our 20 kN (4,500 lbf) methalox bell-nozzle engine. Looking good and ready to fire again. Regenerative cooling worked as expected, with methane flowing through internal channels. The injector needs some fine-tuning because the methane arrived warmer than expected. For a first firing, we err on the side of high injector pressure drop. If it is too low, hot combustion gases can flow back into the injector. But as methane temperature rises, its density falls. Achieving the same mass flow then requires a greater supply pressure. We reached maximum methane supply pressure before achieving the intended fuel flow, so the engine ran oxidizer-rich. It still performed well. The next test will dial this in more precisely. Hot-fires ground Noyron’s models in reality. Only physical testing can provide this kind of coupled calibration data.

Hero shot of our 20 kN methalox engine firing – the engine reached its nominal 50 bar chamber pressure on the first attempt. Lots of data to crunch for Noyron.

Frozen after the pre-chill. Our Noyron-generated 20 kN (4,500 lbf) methalox engine, shortly before hot-fire.

Ready for hot-fire: the 20 kN (4,500 lbf) methalox engine from our previous post. Cryogenic methane is demanding to model. Computational Engineering can take us from specification to a manufacturable engine. Physical testing reveals how closely prediction and reality agree.

Our first methalox engine. This Noyron-generated 20 kN (4,500 lbf) pressure-fed bell-nozzle engine was hot-fired in December 2025. Printed in CuCrZr, a high-temperature copper alloy, it reached its nominal 50 bar (725 psi) chamber pressure and more than 93% combustion efficiency. From the initial specification to manufacturing and hot-fire: about three weeks. Not bad for a first attempt.

The only two humans at LEAP 71. Two founders, amplified by Noyron and a network of customers and partners. It may be an unusual company structure today, but likely a preview of how AI can augment humans to achieve what would have seemed impossible just two or three years ago.

The first AI that builds machines. But a different kind of AI, built specifically to answer a challenging question: How can we synthesize a physical device from the rules human engineers traditionally translate into geometry by hand? How can we combine logic, first-principles physics, manufacturing constraints, assembly experience, and test feedback in a process that is explainable, reproducible, and deterministic? Noyron’s Computational Engineering Models produce the same output from the same model and inputs. But the models themselves are increasingly built using LLM-based inference, directed, instructed, and supervised by humans. This creates two compounding effects. Existing engineering knowledge becomes executable, repeatable, and scalable. At the same time, candidate rules can be explored at a pace that outstrips unaided human research. Computational validation and physical testing determine which candidates become new engineering knowledge. Inference explores. Deterministic computation executes. Humans define intent, challenge assumptions, and decide what is accepted. Noyron builds machines.

This is our 200 kN (45,000 lbf) methalox aerospike, generated by Noyron for Aspire Space’s fully reusable Oryx spacecraft and printed as a single Inconel 718 structure with HBD. Eight months ago, we hot-fired its 20 kN predecessor — essentially a one-tenth-scale version. It reached 50 bar (725 psi) chamber pressure and gave us physical evidence from the same fundamental architecture before moving to ten times the thrust. Aerospikes are altitude-compensating, enabling efficient operation across a wide atmospheric regime. But their annular chambers, central spikes, and cooling systems also make them exceptionally difficult to engineer. That is why we build them. They are among the hardest tests of Noyron’s thermal models — and a direct path from computational predictions to physical evidence. @aspirespaceuae @HBD_Technology

Failure is always an option. Sometimes it teaches you what success cannot. Two years ago, we tested this 1.5 kN (340 lbf) kerolox engine, 3D printed in an experimental high-temperature aluminum alloy. It failed almost instantly. The chamber was regeneratively cooled by the kerosene. According to Noyron’s thermal models, it had ample margin. So what happened? We ignite our engines with a gaseous hydrogen/oxygen flame sized to start much larger chambers. Our sequence started and confirmed the igniter before opening the main fuel and oxidizer valves. This prevents unburned propellants from spilling onto the test stand if ignition fails. But it also created a brief period in which the igniter was firing without kerosene flowing through the cooling channels. For this small engine, the ignition flame was enormous. It weakened the chamber wall before regenerative cooling began. When the high-pressure fuel reached the channels, they yielded. The result was a spectacular spray of molten metal and a strangely artistic engine. Look closely and you can see cooling channels exposed in the melted wall. The thermal model was right within the conditions we gave it. The failure was in a startup condition we had not contemplated. Reality finds the assumptions you forget to encode.