Summary of Key Points
After disbanding its robotics team in 2020 due to a lack of real-world data, OpenAI has restarted its robotics efforts six years later by establishing the OpenAI Robotics team, led personally by CEO Sam Altman and key developers from Sora. This reboot is not just a simple return; it represents a shift from investing in others to taking matters into its own hands. The short-term goal is to assist workers with infrastructure tasks such as construction sites and data centers, while the long-term vision is to provide everyone with a personal robot. Behind this initiative is OpenAI’s “World Simulation” technology, which enables AI to understand physical laws in virtual environments. The motivation also includes the closure of Sora (due to high costs and copyright issues) and the breakdown of cooperation with Figure AI (over differences in technical approaches). As OpenAI prepares for its potential trillion-dollar IPO, the robotics business serves as a new growth story to attract investors. However, it faces competition from global rivals such as Tesla, Figure AI, and domestic companies.
Detailed Analysis
#### 1. Six Years of Rebirth: From Abandonment to Restart
Six years ago, OpenAI disbanded its robotics team because AI lacked the knowledge of real-world physics—things like how much force to use when picking up a cup without breaking it or how to maintain balance while walking required extensive training with real data, which was not available at the time. Today, things have changed. OpenAI has a project called “World Simulation,” which allows AI to practice in virtual environments—simulating scenarios where a cup will break if dropped, a door needs to be turned to open, and how a person’s center of gravity shifts while walking. This technology, developed by the Sora video team (which can create coherent videos by understanding time, space, and causality), is now being applied to robotics, transforming AI’s understanding of the virtual world into practical applications in the real world.
The lead of the team is Adia Ramesh, a key developer behind DALL·E and Sora, who excels at helping AI understand the logic of how the world functions—essential skills for robotics.
#### 2. What Will the Robots Do?
OpenAI’s robotics strategy involves two phases:
- Short-term: Creating industrial “tool robots” that assist workers in tasks such as construction, data center maintenance, and energy facility inspections, where robots can collaborate with humans to perform heavy and dangerous work more efficiently.
- Long-term: Developing personal assistants that can help with daily tasks like grocery shopping, cooking, and cleaning, shifting the focus from serving businesses to serving individuals. This would bring AI out of screens (smartphones, computers) and into everyday life.
#### 3. Why Go It Yourself?
OpenAI’s decision to develop its own robots was driven by two events:
- The Shutdown of Sora: Sora, a popular AI video generation tool, was too costly to operate (15 million USD daily) and generated limited revenue (2.1 million USD), with a poor user retention rate (1% after 30 days). It also faced copyright issues and potential deepfakes concerns. With an IPO approaching, OpenAI needed to get rid of this unprofitable project and shift resources to robotics.
- The Breakdown of Cooperation with Figure AI: In 2024, OpenAI invested in Figure AI and helped develop its models. However, Figure’s founders believed that general-purpose large models (like GPT) were too slow and imprecise for specific tasks, leading to the need for dedicated, end-to-end robot models. The disagreement led to the end of their partnership. This experience taught OpenAI that relying on others meant losing control of core technologies.
#### 4. A Lifeline Before the IPO?
OpenAI’s current valuation is 852 billion USD, but it faces challenges: projected losses of 14 billion USD in 2026 and a low gross margin of 33% due to high AI development costs. Its competitor Anthropic has already secured more corporate clients (34.4% vs OpenAI’s 32.3%). An IPO requires a compelling growth story, and robotics offers this opportunity. By entering the physical world (with a potential global robotics market worth trillions), OpenAI aims to show investors that it can not only understand the world but also transform it, thus expanding its profit potential.
#### 5. Global Competition
OpenAI is not alone in the robotics race:
- International Competitors: Tesla’s Optimus (a home assistant robot), Figure AI (for logistics), Boston Dynamics (advanced robotic capabilities), and NVIDIA (recently unveiled a humanoid robot design in collaboration with Chinese company Yuzhu).
- Domestic Competitors: Chinese startups like Zhiyuan, Yuzhu, and Ubot, as well as giants like Alibaba, ByteDance, and Huawei, are all working on “embodied intelligence” technologies.
OpenAI’s strength lies in its AI models, but it needs to develop hardware and real-time control capabilities, which poses a significant challenge.
#### In Conclusion
OpenAI’s restart of robotics is a natural progression based on its technological progress. It represents a strategic move towards becoming more than just a digital AI giant; however, the path ahead is challenging, as competitors have been waiting for its entry for six years.