Foundation Models & Agent Frameworks · B2B SaaS Provider
Physical Intelligence
Robotics foundation models and hosted inference for general-purpose robot control.
Available information varies by company and source.
Profile record updated:
Company facts
- Official name
- Physical Intelligence (Pi), Inc.
- Entity type
- COMPANY
- Founded
- 2024
- Headquarters
- United States
- Company size
- 50–200
- Market cluster / sector
- Foundation Models & Agent Frameworks
- Market role
- B2B SaaS Provider
- Official website
- physicalintelligence.company
What Physical Intelligence does
Physical Intelligence develops robot foundation models, distributes selected models and tooling through the open-source OpenPi repository, and provides managed model training and cloud-hosted inference through the Pi API beta. It creates value by allowing robotics teams to adapt pre-trained vision-language-action models to their own robot data and embodiments without building model-training and inference infrastructure from scratch.
Category differentiation
Physical Intelligence is a B2B robotics foundation-model and inference provider, not a robot hardware manufacturer or a general consumer AI assistant. It differs from conventional industrial automation vendors by supplying adaptable learned robot-control models.
Strategic context
AI-supported assessment from the existing company research; distinguish interpretation from sourced facts.
Physical Intelligence (Pi), Inc. is a private US robotics-AI company founded in 2024. It develops general-purpose vision-language-action foundation models and learning algorithms that enable robots and other physically actuated devices to perform diverse tasks. Its product portfolio includes the open-source OpenPi package, Pi0 model family, Pi0.7 robotic foundation model and the Pi API beta for managed training and cloud-hosted inference. The company serves robot manufacturers, industrial automation teams, robotics start-ups, engineering teams, researchers and universities. OpenPi expands technical adoption through open-source code and model weights, while the Pi API beta creates a commercial route for hosted compute, customer-data fine-tuning and deployed model inference. Public pricing is not disclosed, and the company reported no generated revenue as of October 2026.
Company news briefing
Briefing updated:
Physical Intelligence continues to cement its position as a leading model partner across the embodied AI ecosystem, supplying foundational robot-control models integrated by emerging modular platforms like Feather Robotics and training-data providers such as Microagi. Technical deployments increasingly highlight hybrid compute trade-offs, balancing off-robot datacenter inference architectures (such as π0.7 on H100 hardware) with local execution layers. Meanwhile, the broader physical AI landscape sees continued executive talent flow and substantial venture investment, with former alumni helping seed competing specialized sensing startups.
Business model & monetisation
OpenPi is distributed as an open-source research and development package. Commercial monetisation is centred on the Pi API beta, which provides managed compute for customer-data uploads and model fine-tuning plus cloud-hosted robot inference. Physical Intelligence has not disclosed standard pricing, usage rates or contractual terms.
- Pi API managed model training and fine-tuning
- Usage-based managed compute service
- Pi API cloud-hosted inference
- Usage-based API and inference service
- OpenPi open-source package
- Open-source distribution and developer acquisition
Products & capabilities
| Product | Description | Source |
|---|---|---|
| OpenPi | OpenPi is Physical Intelligence's open-source robotics package containing vision-language-action models and supporting training, fine-tuning and inference code. It provides the Pi0, Pi0-FAST and Pi0.5 model variants, with base checkpoints trained on more than 10,000 hours of robot data and examples for adapting models to custom robot platforms and datasets. |
|
| Pi API | Pi API is Physical Intelligence's hosted robotics model platform for training and inference. The beta allows robotics teams to upload their own data, fine-tune Physical Intelligence models using managed compute and deploy cloud-hosted inference on their own robots. |
|
| Pi0.7 | Pi0.7 is a steerable general-purpose robotic foundation model designed to control robots across diverse tasks and environments. It supports language-guided behaviour, multi-stage manipulation and zero-shot cross-embodiment transfer, and is made available through Physical Intelligence's Pi API beta. |
|
POLARIS7 WORKSPACE
Go deeper on Physical Intelligence and its market context
Access extended corporate structures, historical M&A and partnership links, technical footprint data, and automated Watchlist briefings grounded in primary sources.
- Corporate & subsidiary hierarchy
- Competitor & partner network
- Automated Watchlist briefings
Competitors & alternatives
Track competitor shifts
Receive automated telemetry on strategic moves, acquisitions, and market share shifts across Physical Intelligence's competitive landscape.
Private enterprise data labs for long-running AI agents.
Cloud desktops and APIs for AI agents to operate computers.
Enterprise runtime for authorised, governed AI-agent actions.
Physical-AI developer for engineering and manufacturing workflows.
Alibaba Cloud platform for multimodal AI inference, APIs and deployments.
Recent recorded signals
Dates refer to the source publication. Older entries are historical context, not evidence of a new event.
Robot AI Inference: On-Device vs Datacenter Compute Trade-offs
Infrastructure · Recorded impact score: 3/5
This analysis examines the computational architecture for embodied AI, weighing on-device inference (e.g., NVIDIA Jetson Thor) against off-robot datacenter inference for generalist robot models. Key trade-offs include real-time latency, cost, and silicon efficiency. While on-device compute ensures determinism, it limits model size; offloading enables larger models but introduces network latency and security issues. The article argues that a hybrid cascade is inevitable, with hierarchical models placing heavy planning in the cloud and fast action layers locally. Examples include Figure running Helix on-robot, Physical Intelligence's π0.7 off-robot on H100, and Boston Dynamics using onboard Jetson Thor with Google TPUs off-robot. Benchmarks show offloading to a B300 offers ~46% of on-device TCO per PFLOP at 40% utilization, and one B300 can serve seven robots with a p99 latency of 1.16 seconds. However, the network wall—uplink, handoff, scheduling—remains the main hurdle, requiring co-designed hardware and access point improvements.
- Frontier robot models range from 5 to 14 billion parameters (π0.7, DreamZero); Jetson Thor has ~1/10th the compute of a B200 and ~1/14th of a B300.
- Offloading to a shared B300 GPU serves up to 7 robots at $0.17/hr/PFLOP vs on-device Jetson Thor at $0.37/hr/PFLOP (40% utilization), achieving ~46% of on-device TCO.
Explore company relationships
Questions about Physical Intelligence
What is Physical Intelligence?
Physical Intelligence is a US robotics-AI company building general-purpose foundation models, open-source tooling and hosted inference for robot control.
Who uses Physical Intelligence?
Robot manufacturers, industrial automation teams, robotics start-ups, developers, universities and research teams use its models and tooling.
How does Physical Intelligence make money?
It commercialises managed model training, fine-tuning compute and cloud-hosted inference through the Pi API beta; public pricing is undisclosed.
Sources & coverage
This profile uses public, official and technically observable information. Missing information does not prove that a product or relationship does not exist. The list below does not imply that every profile statement has been verified.
1 publicly documented primary sources and citations linked across the market graph.
Start with a company or market. Track what changes.
Create your Polaris7 workspace to explore extended company profiles, organize relevant players into Watchlists, and receive focused briefings when material changes occur.
Physical Intelligence
Open extended company details, ownership links, and automated signal monitoring in Polaris7.
