San Francisco-based startup Bifrost is accelerating its push into South Korea, betting that the country’s deep manufacturing ecosystem will drive demand for next-generation “physical AI” systems powered by synthetic data and simulation technologies.
Speaking at the AWS Summit Seoul 2026 held at Coex in southern Seoul, co-founder and CEO Charles Wong said Korea’s industrial breadth — spanning semiconductors, electronics and advanced machinery — makes it uniquely positioned to adopt AI systems that operate in the physical world. He noted that few countries combine such a wide range of manufacturing capabilities within a single, tightly integrated ecosystem.
Bifrost, founded in 2020, develops computer vision and 3D generation technologies designed to train AI models for real-world environments, including robotics, autonomous vehicles and maritime systems. The company has already secured its first Korean client, Seadronix, an autonomous maritime startup using Bifrost’s synthetic data platform to train navigation AI for ships.
The company is now expanding beyond maritime applications, with a growing focus on robotics. Wong said this strategic shift has gained momentum as industry leaders increasingly frame robotics and physical AI as the next major frontier for artificial intelligence. He added that Bifrost recently completed a successful trial with a major Korean firm developing next-generation home robotics, using its tools to accelerate product development and unlock new capabilities.
Beyond pilot projects, Bifrost is in active discussions with several large Korean conglomerates regarding potential partnerships, with announcements expected once agreements are finalized.
Wong’s vision for Bifrost traces back to his earlier work on autonomous driving systems at NuTonomy, where he helped develop AI perception models for self-driving vehicles. That experience convinced him that autonomous technologies would eventually extend far beyond transportation into homes, factories and broader industrial applications. The key constraint, he said, was not computing power but the availability of high-quality training data.
To address this bottleneck, Bifrost built its core platform, Stardust, which enables developers to generate synthetic datasets and simulated environments without requiring specialized expertise in 3D modeling or simulation. According to Wong, the platform allows developers to create production-ready datasets in minutes, significantly reducing the cost and time required to test AI systems in real-world scenarios.
Rather than replacing real-world data entirely, Bifrost’s approach focuses on complementing it by identifying system weaknesses early in development. This enables companies to refine their AI models before deploying them in costly real-world testing environments.
Despite maintaining a relatively small team of around 30 employees, Bifrost has attracted a diverse client base spanning maritime, aerospace, defense and government sectors. Its customers include organizations such as the U.S. Air Force, NASA and defense technology firm Anduril. The company has raised $13.7 million to date across four funding rounds, including a recent $8 million investment backed by Carbide Ventures, Airbus Ventures and Peak XV’s Surge.
Underlying Bifrost’s platform is its reliance on cloud infrastructure, particularly services that enable scalable computing, storage and AI deployment. Wong highlighted tools that allow rapid testing of AI perception systems, as well as emerging capabilities that could simplify how developers build simulation scenarios. He suggested that future iterations of the platform may allow users to generate environments through natural language prompts, reducing the need for manual coding and further lowering barriers to entry.
As global industries increasingly look to integrate AI into physical systems, Bifrost’s expansion into Korea underscores a broader shift toward simulation-driven development — where synthetic data plays a central role in bringing AI out of the lab and into the real world.