Aether AI

Company Overview

Aether AI (Primordial Intelligence, Shanghai) Technology Co., Ltd., founded in Shanghai in 2026, is a cutting-edge artificial intelligence company specializing in causal world models. Founder Prof. Biwei Huang, currently an assistant professor at the Halıcıoğlu Institute for Data Science at the University of California, San Diego (UCSD), has dedicated over twelve years to the fields of causal discovery and machine learning. She has published more than 100 papers in top conferences such as NeurIPS, ICML, ICLR, and CVPR, and has led the development of Causal-Learn and Causal-Copilot, the globally recognized open-source tools that serve as industry standards in the field of causal discovery.

On June 17, 2026, Aether AI officially announced the completion of its first round of funding, raising approximately $20 million. The round was led by Matrix Partners China, with co-investors including InnoFund, SWC Global, and Jiuhe Ventures. These funds will be used for the research, development, and iteration of the Causal World Model; the construction of engineering infrastructure; the expansion of the core team; and the first commercial deployments in the field of Physical AI.

The Company's Network of Academic Advisors CoversCausal AILeading scholars in the field, including Turing Award laureate Professor Judea Pearl, Max Planck Institute Director Professor Bernhard Schölkopf, the founders of the field of causal discovery—Professors Clark Glymour and Peter Spirtes—and Professor Kun Zhang—this lineup is virtually unrivaled in the entire field of causal AI.

Development Journey: From Academic Excellence to Commercial Breakthrough

Huang Biwei categorizes the development of AI into“The Four Stages of Paradigm Evolution”: Current mainstream large language models belong to the third generation (large models + correlation), while the fourth generation (large models + causal mechanisms) that she is promoting will become the next-generation paradigm.

Products and Services: Building a ”Causal Brain” for Robots”

Aether AI has chosen the field of embodied intelligence (Physical AI) as the focus of its first application, with the goal of creating a unified causal reasoning layer for robots—the ”causal brain.”

There are three main reasons for choosing this sector:

1. AI is expanding from the digital world into the physical world, robots must be capable of performing real-world tasks;
2. There is no unified paradigm in this field yet., there is an opportunity to establish definition standards;
3. The data on embodied intelligence is relatively clean, to facilitate the verification of the validity of the causal model.

In the physical world, every action a robot takes is essentially an intervention; decisions based on statistical shortcuts, once they go wrong, immediately result in operational failure—embodied intelligence thus becomes the most direct and compelling setting for testing causal reasoning abilities.

The early validation data is extremely impressive: Causal methods have improved data efficiency in certain operational tasks20%–30%, for as little as50 high-quality causal annotation data points...which will enable missions that previously failed frequently to achieve a reliable success rate.

Huang Biwei has set clear milestones:

• Early 2027: Achieve the ”GPT-3 moment” in robotic operations, demonstrating cross-task generalization and long-range reasoning capabilities;
• Second half of 2027: To foster independent exploration and lifelong learning in an open environment.

Core Technology: A Four-Layer Architecture Driving a Causal World Model

Aether AI's technology stack consists ofFour-Tier ArchitectureThe design approach is not to start from scratch, but to transition smoothly from the existing scalable architecture and gradually introduce causal mechanisms:

Compared to the mainstream paradigm, Aether AI hasThree Fundamental Differences::

Huang Biwei used a vivid example to illustrate the difference: when an egg is cracked into a pan of hot oil, the causal model can account for the relationships between variables such as oil temperature and pan size—Structural Relationships...it can still accurately predict the results even when the oil temperature or pan size changes; whereas pure correlation models fail when these variables change.

Dr. Zhou, the technical partner, has made pioneering contributions in the field of large-scale model training, while Dr. Feng has spent many years specializing in causal world models and reinforcement learning. The team also includes several young members born in the 2000s, all of whom share the belief that causal intelligence is the inevitable direction for AI development.

Future Prospects: At the Forefront of a Paradigm Shift in AI

1. Industry Background: Bottlenecks in the Traditional Approach Have Become Apparent

In the first half of 2026, the global AI investment community began to reevaluate traditional approaches. The field of embodied intelligence had attracted significant investment over the past three years, and vision-language-action (VLA) models were the subject of high expectations; however, most demonstrations performed poorly in real-world environments. Tong Ti, a partner at Matrix Partners China, stated bluntly: ”Relying solely on data patterns and correlation-based learning is no longer sufficient to meet the demands of next-generation intelligent systems.”

Investors in Inno Fund offer a more incisive assessment: ”Causality is a unique expression of human intelligence and a massive compression of information about the world; its data efficiency and parameter efficiency are thousands or even tens of thousands of times superior to empirical correlations in statistics. The observational, action, and counterfactual systems of causal intelligence hold far greater intellectual potential in world modeling than current empirical systems.”

2. Differentiated Competition: Not Disruption, but Evolution

Compared to other world model approaches on the market, Aether AI’s key differentiator lies inExplicit Learning of Causal Variables and Structures::

Huang Biwei made it clear: ”The company’s goal is not to disrupt the existing architecture, but to gradually introduce a causal framework through a smooth transition, ultimately pioneering a next-generation AI paradigm centered on causal intelligence.”

3. A clear path to commercialization

• Short Term (2026–2027): The first commercial deployments in the field of embodied intelligence, providing robots with a causal reasoning layer;
• Mid-term (2027–2028): The causal world model has been extended from robotics to broader Physical AI scenarios;
• Long-term: Build a complete technology stack spanning from foundational models to agent systems, driving AI’s transition from ”pattern recognition” to ”mechanism understanding.”