Silicon Valley startup Physical Intelligence aims to build 'ChatGPT for robots'

SAN FRANCISCO – A startup called Physical Intelligence is developing a general-purpose artificial intelligence model designed to teach robots how to perform everyday chores, ranging from folding laundry to peeling vegetables.

The company’s approach has been described as "ChatGPT for robots." At its headquarters, robotic arms are currently undergoing testing to master mundane tasks such as flipping shirts and peeling squash.

Sergey Levine, a company co-founder and associate professor at UC Berkeley, said the firm collects data from robotic stations across multiple locations to train its foundation models. These models are then evaluated and refined at specialized testing stations.

Physical Intelligence’s strategy focuses on using sophisticated software to compensate for simpler, less expensive hardware. The robotic arms used in their research cost approximately $3,500, and they can be manufactured with material costs of less than $1,000.

Investor Lachy Groom, an early employee at Stripe, backed the venture after being impressed by the academic work of Levine and other researchers from Stanford and Google DeepMind. To ensure the robots can adapt to diverse challenges, the company is testing them in various real-world environments, including a kitchen.

Physical Intelligence is positioning itself as a disruptor in the automation space, signaling a fundamental paradigm shift from hardware-centric engineering to adaptive, software-driven intelligence. The company’s core thesis—that sophisticated AI can compensate for low-cost hardware—is a high-stakes gamble that aims to democratize robotics in much the same way Large Language Models (LLMs) have transformed generative AI.

For decades, the industrial robotics sector has been defined by high-cost, bespoke hardware designed for singular, repetitive tasks. Physical Intelligence is upending this capital-intensive model. By leveraging affordable, off-the-shelf robotic arms and focusing resources on a centralized "foundation model," the startup is building a digital brain capable of learning and generalizing physical movements across disparate environments. If successful, this unified model could bridge the gap between niche automation and general-purpose utility, controlling everything from food preparation in commercial kitchens to logistics in high-volume warehouses.

The firm’s strategic pedigree is equally significant. Backed by Lachy Groom—an investor with a track record of identifying early-stage winners like Figma and Notion—and led by a founding team of elite researchers from Berkeley and Stanford, the company represents a potent blend of deep-tech academia and Silicon Valley’s commercial pragmatism.

However, the path to scalability remains fraught with technical hurdles. The primary challenge lies in the "data bottleneck": the massive volume of physical interaction data required to train these models to a level of real-world reliability that meets industrial standards. Should Physical Intelligence overcome this barrier, the implications will extend far beyond the factory floor, potentially restructuring global labor markets and the future of the consumer services sector.

While these advanced robotics have yet to directly impact Vietnamese-American small businesses, the long-term trend toward low-cost automation could eventually reach labor-intensive sectors like the nail salon industry or the phở restaurants that anchor Little Saigon. For now, this is more of a tech story to keep an eye on rather than an immediate economic threat to the community’s livelihood, our remittances, or the stability of those here on F2B, H-1B, TPS, and EB-5 visas.