Announcing Engineering Superintelligence.
A frontier AI lab founded by two Stanford professors and engineers working at the cutting edge of engineering who sets out to build a new class of intelligence: AI that can design, simulate, test, and build the physical systems civilization depends on.
MENLO PARK, Calif. — September 13, 2026 — Apex Ultra today announced its mission: Engineering Superintelligence (ESI), artificial intelligence capable of designing, simulating, testing, and building physical systems — from launch vehicles and fusion reactors to race cars and robots — by reasoning across the disciplines that engineering requires and learning from the consequences of its decisions. The lab, headquartered at SRI International in Menlo Park with teams in San Francisco, Boston, London, Zurich, and Paris, has assembled AI researchers, engineers, and computer scientists from the Stanford AI Lab and Stanford Aerospace, MIT, Brown, UC Berkeley, Imperial College London, Oxford, Cambridge, Google DeepMind, OpenAI, Anthropic, NASA, the U.S. Space Force, and the Mercedes-AMG and McLaren Formula 1 teams. It trains and evaluates its models in partnership with organizations operating at the extremes of engineering.
Why engineering
The hardest problems in front of humanity are engineering problems. Abundant energy, affordable access to space, transport that does not cost lives, resilient infrastructure, machines that can work alongside people — each depends on our ability to design and build physical systems better and faster than we do today. That ability is constrained not by ambition but by the number of people who can hold a whole machine in their head: its physics, its materials, its manufacturing, its mission. Modern systems have outgrown that number. Design cycles are measured in years, tools separate the disciplines that must reason together, and the cost of being wrong is enormous.
Apex Ultra was founded on the conviction that this is the most consequential place for artificial intelligence to go next — and the least explored.
What ESI is
Engineering Superintelligence is intelligence that works in the language of engineering rather than the language of description: physics, geometry, materials, tolerances, loads, thermal limits, manufacturability, cost, and mission. It designs under constraints, simulates its designs, proposes tests, and revises when the tests disagree. It produces artifacts engineers can verify — geometry, analyses, specifications — inside the tools engineers already use. Its output is not a document about a machine. It is the machine, made checkable.
ESI is not a chatbot and not a general assistant. It is an engineer.
How ESI differs from language models and world models
Large language models are trained on what people have written. They are remarkable at description, explanation, and code, and they will remain part of any engineering toolchain. But engineering truth is not stored in text. A language model predicts what someone would say about a bracket; it cannot know whether the bracket holds. Its errors are cheap in conversation and catastrophic in a load path.
World models learn how the physical world evolves from observation — video, sensors, interaction — and they are essential to robotics. Engineering, however, is not the prediction of a world that exists. It is the creation of things that do not yet exist, to meet requirements, within constraints, in a form that can be manufactured and verified. That demands intent, constraint satisfaction, and a closed loop with testing that observation alone does not provide.
ESI sits at the intersection the other two leave open: goal-directed design, grounded in physics and simulation, operating in real engineering systems, and corrected by real outcomes.
The conditions we believe ESI requires
We do not believe ESI can be trained in the abstract. Our work rests on a set of conditions we consider necessary, and our approach is organized around meeting them.
- Real engineering environments. The intelligence must be trained and evaluated where physical systems are actually designed, built, and tested — where outcomes are measurable and being wrong has a cost.
- Real engineering data. Test results, simulation runs, design histories, manufacturing outcomes, and the reasons decisions were made. This data is proprietary, unlabeled, and absent from the open internet; it exists inside organizations working at the limit of their discipline.
- Immersion in real workflows and systems. ESI must operate the tools engineers use — CAD, simulation and analysis, product lifecycle systems, test infrastructure — and produce artifacts those systems accept and engineers can check.
- A closed loop between design and consequence. Predictions must be compared with tests; simulation must be reconciled with reality; the system must learn from the gap.
- Expert humans in the loop. The best engineers in each field, as teachers, reviewers, and the final authority on what ships.
- Isolation and stewardship by design. Sovereign data boundaries, traceable decisions, and human control are not features to add later; they are conditions for being trusted with the work at all.
- Measurement over demonstration. Progress judged by engineering outcomes in the field, not by demonstrations.
How Apex Ultra is pursuing it
Apex Ultra began as a collaboration between two professors at Stanford along with top engineers working at the most extreme levels of engineering — who approached the same problem from different disciplines. The team they assembled spans fundamental research and the demands of building real systems.
To meet the conditions above, the lab trains and evaluates its engineering foundation models inside a small number of strategic sandbox partners operating at the extremes of their fields — Formula 1, hypersonics, aerospace and defense, fusion energy, automotive, and robotics — against those partners' real problems and under their real constraints. Every deployment is isolated; no partner's data or intent is pooled with another's. Partners receive early access to the systems trained in their environment and help shape what is built and how it is tested.
"Engineering is where AI meets consequence. A model can be wrong about a sentence at no cost; it cannot be wrong about a load path. We started Apex Ultra because we believe intelligence that can build is the most important thing AI can become — and because it will only be earned in the environments where being wrong matters."
Co-founder, Apex Ultra
About Apex Ultra
Apex Ultra is a frontier AI lab building Engineering Superintelligence (ESI): AI that designs, simulates, tests, and builds complex machines and physical systems, trained in the most demanding engineering environments in the world. Founded by two Stanford professors and built by researchers and engineers from the world's leading laboratories, aerospace programs, and Formula 1 teams, Apex Ultra is headquartered at SRI International in Menlo Park, California, with teams in San Francisco, Boston, London, Zurich, and Paris. apexultra.com · @apex_ultra_ai
Media contact
Apex Ultra Communications
ai@apexultra.com
Press assets, including the Apex mark and share images, are available on request.
Note to editors
The company name is written "Apex Ultra" — two words, capital A, capital U. Engineering Superintelligence is abbreviated ESI. The lab's mission, team, and approach are described at apexultra.com.