Rivian Buys a UK Startup That Builds 3D Worlds From Video

Rivian has bought Third Dimension, a UK startup that uses AI to turn images and videos into realistic 3D worlds, according to TechCrunch. The price was not disclosed.
Third Dimension was founded in 2024 by Tolga Kart and Piotr Sokolski. It raised $7 million in seed funding, an early round of investment, from Abstract Ventures, MVP, Soma Capital and Solari Capital. Its 15 staff members, most of them based in the UK, will join Rivian's autonomy and AI team. That team is led by James Philbin.
Rivian plans to use the technology to speed up work on advanced driver assistance features. These are systems that help a driver with tasks such as staying in a lane. Rivian also plans to use it for the self-driving system behind the robotaxis it intends to supply to Uber. A robotaxi is a taxi with no human driver.
The robotaxi work is part of a wider deal. Uber and Rivian announced a partnership on March 19, 2026 to deploy up to 50,000 fully autonomous robotaxis. Separately, IEEE Spectrum reported on September 8, 2026 that Uber is backing Rivian with $1.25 billion. The same report said Rivian plans to deploy a fleet of Level 4 robotaxis starting in 2028. Level 4 means the car drives itself within a defined area, without a person needing to step in.
The data is the part that explains the technical logic of the deal. According to TechCrunch, Rivian receives more than 400,000 driving sequences, which are short recorded clips, from its customer cars every day. Rivian says its Rivian Drive program helps train its AI model to handle real-world situations.
Think of the fleet as a camera that never switches off. It captures the rare and unusual moments, the edge cases, that a self-driving system has to learn. The open question is how quickly that raw footage can be turned into 3D environments that engineers can use for training and simulation.
Rivian has been building its self-driving technology in-house on several fronts. On December 11, 2025, the company unveiled its first custom computer chip for self-driving. Reuters and Bloomberg reported that the chip was designed to replace Nvidia processors. Rivian shares fell after that announcement. The Third Dimension deal follows the same pattern. It moves the company's work beyond the chip and into how the car sees the road and how its software gets tested.
Rivian was founded in 2009. It is headquartered in Irvine, California, and is led by founder and CEO RJ Scaringe. The company went public on November 10, 2021, when its shares were first sold to ordinary investors. Its plant in Normal, Illinois covers 3.3 million square feet. On March 12, 2026, Rivian introduced its R2 lineup with full trims and pricing. It began R2 deliveries to public customers and opened orders to reservation holders on June 9, 2026. On October 14, 2026, Rivian and Redwood Materials expanded their partnership to cover the full life of the battery, from production to recycling.
The broader context here is that a carmaker's self-driving work now depends on three things: the chips that run the software, the data gathered from real driving, and the simulated worlds where software is tested before it reaches a road. Buying a small team with a specific skill in building 3D scenes addresses the third of these directly. Whether it speeds up development will become clearer once Rivian ships the driver assistance features and starts the robotaxi service it has described.
In my view, the more interesting question is not the purchase itself but verification. A simulated world is only useful for safety testing if engineers can show that it matches what the fleet actually encounters on the road. Pilots train in flight simulators for the same reason, since a simulator is only worth trusting if it behaves like a real aircraft. That is a difficult measurement problem, and in my view it will matter more to the 2028 timeline than the headline deal does.
If the approach works, cars already on public roads would become a source of material for building the test worlds that their next software runs in. In my view, that would be a meaningful shift in where self-driving systems get their experience, at a time when the industry has spent years debating how many real-world miles they need.


