Enigma $71M funding backs robot foundation models
Enigma launched with a $71 million seed round to build foundation models and interfaces for robots, with valuation and revenue undisclosed.
By Dominic Okoye · Staff Writer
· 3 min read
Enigma $71M funding was announced as the San Francisco startup launched with a plan to build foundation models for robots. Index Ventures and Ribbit Capital co-led the seed round, with Conviction Partners and angel investors from Google DeepMind, Anthropic and OpenAI also participating.
The company did not disclose its valuation, revenue, customer count or headcount. It said the capital will go toward hiring engineers and buying compute capacity, two cost centers that define how quickly a robotics AI company can train and test models before it has broad commercial proof.
Enigma was founded last year by cybersecurity researchers Jonathan Jacobi and Gal Niv. Jacobi, the company’s chief executive, previously joined Microsoft at age 17, which Enigma describes as making him the company’s youngest employee at the time. He later worked at a startup acquired by Wiz, the cybersecurity business owned by Google.
What is Enigma building for robots?
Enigma says it is developing foundation models meant to run across many types of robots and reduce the amount of task-specific training data required to adapt software to a new machine. In robotics, a foundation model is a general-purpose AI model that can be tuned for different robots and tasks instead of being built from scratch for each deployment.
The practical target is the programming burden in industrial robotics. A robot arm typically needs custom code for a new task, and that code often has to be changed for each physical setup. Even two matching robotic arms can require different configuration scripts if they sit beside conveyor belts with different dimensions.
AI models can shorten that setup process. Nvidia’s GR00T, for example, is designed to let engineers configure supported robots using natural language prompts. Enigma is making a broader claim: that its models can operate on practically any robot and can be adapted with far less training data than is usually needed. The company did not provide benchmark results, deployment metrics or the size of the claimed reduction.
Training data is a hard cost in robotics because it often requires recording robots performing tasks in physical settings. In some cases, companies build mock production lines to collect the footage needed to teach models how to act. If Enigma can lower that requirement, the product would address a real deployment bottleneck rather than a generic AI feature.
Who else is trying to cut robotics training data?
Meta is working on a related problem. Last year it released V-JEPA, a model the company says can be adapted to a new robot with under 100 hours of explanatory footage. Meta has said the model uses a neural network architecture introduced in 2022 that helps AI systems predict future events in a factory setting and improve decision-making.
Enigma is also building a human-robot interface alongside its models. The company launched a website where users can interact with more than 100 robots configured for different tasks. Enigma said it will use data from those interactions to improve the interface.
Jacobi said the company’s view is that AI will move beyond chatbots and screens into systems that can understand, adapt to and operate in the physical world. That is the standard thesis behind the current wave of robotics AI funding. Enigma’s next test is whether it can show that its models work across real machines and environments, using less data, outside a controlled demo.
This story draws on original reporting from SiliconANGLE.
