Show Notes
Some of the smartest engineers alive are designing the physical world with software older than their interns. Brake discs, axles, medical devices, aircraft components — all built on tools that can take twenty minutes just to open a file, and a knowledge base that walked out the door when the industry shipped its expertise overseas. Hugo Nordell saw this up close, and it bothered him the way only a theoretical physicist turned operator could be bothered by a problem.
Hugo is the co-founder and CEO of Encube, which exited stealth in October 2025 with a $23 million round led by Kinnevik alongside Promus Ventures and Inventure, validated through R&D programs with Volvo Group, Scania, and Beyond Gravity. Before starting the company with co-founder Johnny Bigert in late 2021, Hugo spent years in Silicon Valley's drone and autonomous driving era, then led digital transformation at Sandvik and Aker. He kept watching brilliant hardware teams fight their own tooling every day while production costs quietly ballooned. So he built the thing he wished he had.
Encube is a browser-based, collaborative design platform that sits between your CAD system and your release management, then layers AI on a foundation almost nobody else is building: a deterministic engine that actually understands manufacturability. Think of it as a shared whiteboard where complex CAD models and heavy engineering drawings load in two to three seconds on a standard laptop — no expensive graphics card required. People thought he was cheating. He was not. That speed comes from a custom geometry and physics compute engine Hugo's team built from scratch.
The deeper insight is one every founder building AI products should internalize. Generative AI is rewriting software engineering because software has forty years of validation infrastructure: compilers, linters, unit tests, CI/CD, and stack traces that let agents self-correct. Hardware has none of that. There is no compiler for atoms. So Encube is building one — deterministic where it must be, with large language models bolted on only at the edges where stochastic answers are safe. Get that order right and you could reimagine hardware design the way Lovable, Bolt, and Claude Code reimagined software. Get it wrong and you ship slop into the one place slop kills people.
Frameworks from This Episode
The Three KPIs That Actually Matter in Hardware
Most hardware teams track the wrong numbers. Hugo's framework focuses on the three levers that actually compound into competitive advantage.
- ›Time to market: the single biggest lever for competitive advantage. Nvidia wins partly on release cadence, not just silicon.
- ›Engineering productivity: critical because hardware is facing its worst talent scarcity since the second industrial revolution.
- ›Marginal cost of production: ignore your cost of revenue while scaling and you dig your own early grave.
Shift Left
Surface trade-offs and manufacturability problems early in the design process, when they are still cheap to fix. By the time you reach production, the cost of a change can be measured in months.
- ›Up to 80% of a product's final cost is locked in during the design phase, long before manufacturing problems become visible.
- ›Catching a problem at design costs cents on the dollar versus catching it at production.
- ›A six-month production delay is almost always a design review that didn't shift left far enough.
- ›Concurrent engineering — exploring many design and manufacturing paths in parallel early on — is only possible when tooling is fast enough to let you iterate.
The Hardware Compiler
Software AI agents are powerful because they can self-correct — there is a right answer and a fast feedback loop. Hardware never had that. Encube is building the equivalent.
- ›Software has compilers, linters, CI/CD pipelines, and unit tests that give agents the ground truth they need to iterate.
- ›Hardware has had none of that — until now. Encube's deterministic engine answers: can you make this, what drives cost, what breaks under pressure, what is the von Mises stress.
- ›Build the compiler first, then bolt agents on top. Do it the other way and you generate expensive, dangerous nonsense.
- ›This is why Encube is not just a collaboration tool — it is the validation infrastructure that makes agentic hardware design safe.
Deterministic Core, LLM Edges
Run the same compute twice and get the same answer twice. Physical products demand reproducibility. Large language models, left unguarded, cannot guarantee it.
- ›All geometry, physics, and manufacturability computation runs deterministically on Encube's GPU-accelerated spatial compute engine.
- ›LLMs are reserved for translation and explanation: 'explain it like I'm five,' 'do a deep dive,' 'summarize this drawing.'
- ›LLMs never touch the math, the stress calculations, or anything irreversible.
- ›Stochastic answers that drift by 0.1 in software produce a slightly quirky chatbot reply. In hardware, they produce a faulty part.
Zero to One Before Zero to One Hundred
The dream of AI capturing decades of tribal engineering knowledge is real — but it is a future state. You cannot skip the unsexy infrastructure work that makes it possible.
- ›First you fix the basics: messy data, scattered files, broken processes, no single source of truth.
- ›You cannot one-shot from nothing to a knowledge-capturing AI. The foundation has to exist before the intelligence is worth building.
- ›This is true in every category: the AI is only as good as the data and process hygiene underneath it.
The Conductor Model
When computation takes over the heavy lifting, the engineer stops being a button pusher and becomes an orchestrator — exploring many design paths in the time it used to take to explore one.
- ›Old model: the engineer spends 80% of time running simulations and waiting for files to load.
- ›New model: the engineer directs five parallel design explorations simultaneously and evaluates the results.
- ›The productivity gain is not incremental — it is a category change in the nature of the work.
- ›This is the hardware equivalent of what vibe coding did for software: the loop between idea and tested output collapses from days to minutes.
Key Terms
Tools from This Episode
Encube
Browser-based collaborative design platform that sits between CAD and PLM, loads large engineering assemblies in two to three seconds without a graphics card, and layers AI on a deterministic manufacturability engine — so hardware teams can explore five design paths in the time it used to take to explore one.
Q&A
What is Encube?
Encube is a browser-based collaborative design platform for hardware engineering teams. It sits between CAD and PLM, connects disparate engineering data into one visual workspace, and layers AI on a deterministic manufacturability engine. Large assemblies load in two to three seconds on a standard laptop without a graphics card.
Who is Hugo Nordell?
Hugo Nordell is the co-founder and CEO of Encube. A trained theoretical physicist, he spent years in Silicon Valley's drone and autonomous driving era before leading digital transformation at Sandvik and Aker. He started Encube with co-founder Johnny Bigert in late 2021.
Why is non-deterministic AI dangerous in hardware engineering?
Because physical products require reproducibility, and a stochastic answer that drifts from 2 to 2.1 can mean a faulty part rather than a quirky chatbot reply. In software, a hallucination crashes an app. In hardware, it can injure someone. Encube keeps LLMs entirely away from the math.
What is a 'compiler for hardware'?
A deterministic execution environment that validates manufacturability, cost drivers, and failure modes — giving AI agents the fast feedback loop hardware has historically lacked. Software AI works because compilers and unit tests make it easy to validate outputs. Hardware never had that equivalent until now.
How does Encube differ from Autodesk, PTC, or Siemens?
It does not replace CAD head on. Encube sits between CAD and PLM, connecting engineering data that incumbents leave scattered, and layering a manufacturability intelligence layer that the incumbents do not offer. The white space is the integration and collaboration layer, not the 3D modeling core.
Why does most of a product's cost get locked in during design?
Because as much as 80% of final cost is determined by early design decisions — tolerancing, material choices, geometry — often before manufacturing problems are even visible. By the time an issue surfaces in production, the cost to fix it can be a hundred times higher than catching it at design review.
Where should founders actually put AI in their product?
Only where there is a clear right answer and a fast feedback loop. Map every AI touchpoint and ask: is this task deterministic-safe, or am I trusting a stochastic guess on something irreversible? Wherever an LLM is doing math, judgment, or anything that can't be validated, you have found your next failure point.
Where can I learn more about Encube and Hugo Nordell?
Visit getencube.com. Hugo Nordell is active on both LinkedIn and X.
Links from This Episode
- Encubehttps://www.getencube.com
- AI for Foundershttps://aiforfounders.co