Physical Intelligence

PUBLIC

Name	Physical Intelligence
Tagline	Building foundation models and learning algorithms as general-purpose AI brains for robots and physically actuated devices.
Headquarters	San Francisco, United States
Founded	2024
Stage	Series A
Business Model	B2B
Industry	Deeptech
Technology	AI / Machine Learning
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding Label	$100M+
Total Disclosed	~$470,000,000 [CB Insights]

Links

PUBLIC

• Website: https://pi.website
• LinkedIn: https://www.linkedin.com/company/physical-intelligence

Executive Summary

PUBLIC Physical Intelligence is building a universal AI brain for robots, a bet that has attracted over $1 billion in capital and a $5.6 billion valuation in under two years, signaling a major institutional conviction in the emergence of a foundational software layer for physical automation [Bloomberg, Nov 2025] [TechCrunch, Jan 2026]. The company, founded in 2024, aims to move beyond single-purpose robotics by developing Vision-Language-Action (VLA) foundation models that can interpret instructions and perform a wide range of tasks across different hardware platforms, from industrial arms to household robots [Business Insider, Nov 2024] [arXiv, Oct 2024].

The founding team provides a credible technical wedge, combining the deep reinforcement learning expertise of UC Berkeley professor Sergey Levine with the applied research experience of CEO Karol Hausman, formerly of Google DeepMind [Perplexity Sonar Pro Brief]. This academic and industrial pedigree has been instrumental in securing backing from a consortium of top-tier venture firms, including Sequoia Capital, the OpenAI Startup Fund, and Khosla Ventures, alongside individuals like Jeff Bezos [Bloomberg, Mar 2024].

Its flagship model, π-zero, represents the core of its B2B offering, a software stack intended for integration by robotics original equipment manufacturers (OEMs) and operators. The company's rapid valuation ascent from $2.4 billion to $5.6 billion within approximately a year reflects both the scale of the perceived opportunity and the intense competition to establish a standard in physical AI [Bloomberg] [Threads]. Over the next 12-18 months, the critical watchpoints will be the transition from research demonstrations to named commercial deployments and the emergence of clearer performance benchmarks against a growing field of well-funded competitors.

Data Accuracy: GREEN -- Core facts (founding team, product focus, major funding rounds, valuation) are corroborated by multiple independent publications including Bloomberg, TechCrunch, and academic preprints.

Taxonomy Snapshot

Axis	Classification
Stage	Series A
Business Model	B2B
Industry / Vertical	Deeptech
Technology Type	AI / Machine Learning
Geography	North America
Growth Profile	Venture Scale
Founding Team	Co-Founders (2)
Funding	$100M+ (total disclosed ~$470,000,000)

Company Overview

PUBLIC Physical Intelligence was founded in 2024, emerging from a research-driven thesis to build general-purpose AI for the physical world [Crunchbase]. The company is headquartered in San Francisco, California [Crunchbase].

Key milestones are defined by its rapid capital formation and research releases. The company's public narrative began with a $400 million financing round in late 2024, which established a $2.4 billion valuation [Bloomberg, Nov 2025]. This was followed by a significant valuation step-up to $5.6 billion in a subsequent round led by CapitalG [TSG Invest]. Concurrently, the team published foundational research on its π0 (pi-zero) Vision-Language-Action model in October 2024, providing an academic benchmark for its technical approach [arXiv, Oct 2024].

Headcount has scaled with funding, reaching an estimated 51-200 employees as of early 2026 [ZoomInfo] [Forbes, Apr 2026]. The company's public milestones to date are financial and technical, with commercial deployment details remaining closely held.

Data Accuracy: GREEN -- Confirmed by Crunchbase, Bloomberg, and academic publication.

Product and Technology

MIXED

Physical Intelligence's public proposition is a single, ambitious product: a general-purpose AI brain for physical devices. The company describes its flagship model, π‑zero (pi‑zero), as a Vision-Language-Action (VLA) foundation model designed to interpret multimodal inputs and execute a wide range of physical tasks [Perplexity Sonar Pro Brief]. The core claim is that this model can serve as a universal control stack, deployable across many robot types and embodiments, from industrial arms to mobile platforms [Perplexity Sonar Pro Brief]. This hardware-agnostic approach is the central wedge, positioning the software as a shared intelligence layer for robotics OEMs and operators, rather than a solution tied to specific hardware [Perplexity Sonar Pro Brief].

The company emphasizes a data flywheel built on large-scale collection from teleoperation, simulation, and real-world deployments [Perplexity Sonar Pro Brief]. Publicly available research papers detail the π0 architecture and note that such VLA models can improve through reinforcement learning from continued deployment [arXiv, Oct 2024] [arXiv, Nov 2025]. The company's website reports "considerable improvements in real-world autonomous performance at customer deployments with each new model generation" [pi.website]. However, the identities of these customers and the specific performance metrics are not publicly disclosed.

A review of open roles suggests a deep tech stack centered on large-scale machine learning. Job postings indicate work on distributed training infrastructure, large-scale reinforcement learning, and simulation environments [PUBLIC]. The team's published research and recruitment focus point to a technology foundation built on modern deep learning frameworks, likely PyTorch, and a heavy reliance on GPU clusters for model training (inferred from job postings).

Data Accuracy: YELLOW -- Product claims are sourced from company descriptions and research papers, but specific performance benchmarks and detailed technical specifications are not publicly available.

Market Research

PUBLIC

The ambition to imbue machines with general physical intelligence is not a new pursuit, but the convergence of large-scale foundation models, massive compute, and a critical mass of real-world robotics data has created a credible, venture-scale opportunity for the first time. While the company does not publish its own market sizing, the broader context for its offering can be framed by adjacent, well-documented markets and the specific demand drivers cited in coverage of the physical AI category.

Quantifying the total addressable market for a universal robot brain is inherently speculative, as it seeks to create a new software layer across multiple established and emerging robotics verticals. For context, the global market for industrial robots was valued at approximately $16.2 billion in 2022 and is forecast to grow to over $35 billion by 2028, according to a report from the International Federation of Robotics [IFR, 2022]. The professional service robot market, which includes logistics, hospitality, and healthcare applications, is projected to reach $58 billion by 2028 [IFR, 2022]. These figures represent the hardware-centric markets that Physical Intelligence aims to augment with its software layer.

Industrial Robots (2022) | 16.2 | $B
Industrial Robots (2028 est.) | 35 | $B
Professional Service Robots (2028 est.) | 58 | $B

The cited hardware market growth, while not a direct proxy for the software opportunity, illustrates the scale of the underlying physical automation economy into which a general-purpose AI control layer could be sold. The analyst takeaway is that the company is targeting a foundational role within a multi-hundred-billion-dollar automation ecosystem, where even a small share of software spend would represent a significant SAM.

Demand drivers for such a technology are consistently highlighted in industry analysis. Coverage points to persistent labor shortages in sectors like manufacturing and logistics, rising costs, and the need for greater operational flexibility as primary tailwinds [Crunchbase News]. The push for more adaptable, multi-task robots, as opposed to single-purpose automated systems, is a recurring theme that aligns directly with Physical Intelligence's value proposition of a generalist model [Business Insider, Nov 2024]. Furthermore, the demonstrated success of large language models in digital domains has created investor and customer appetite for a similar 'foundation model' approach to physical interaction, a narrative the company's backers explicitly endorse [Bloomberg, Mar 2024].

Key adjacent and substitute markets include traditional industrial automation software, specialized robot operating systems (ROS), and task-specific AI solutions from incumbents. The company's wedge is to displace these fragmented, narrow solutions with a single, more capable generalist model. Regulatory and macro forces are a double-edged sword. On one hand, safety certification for AI-controlled physical systems in regulated environments (e.g., factories, hospitals) presents a significant adoption hurdle. On the other, government initiatives aimed at reshoring manufacturing and bolstering national competitiveness in critical technologies could accelerate investment in next-generation automation, indirectly benefiting providers of advanced robotics AI.

Data Accuracy: YELLOW -- Market sizing figures are drawn from established industry reports for analogous hardware markets, not the specific software TAM. Demand drivers are corroborated by multiple third-party publications.

Competitive Landscape

MIXED Physical Intelligence is positioned as a pure-play, hardware-agnostic foundation model provider in a market currently defined by a mix of specialized incumbents and well-funded new entrants pursuing similar generalist ambitions.

The competitive field for robot intelligence is fragmented across several strategic approaches, from full-stack robotics companies to pure software providers. The most direct comparisons are other startups building large-scale models for physical control.

Physical Intelligence | 470 | $M
Skild AI | 300 | $M
Dexterity | 120 | $M
Agility | 675 | $M
Archetype AI | 32 | $M

Figure: Selected competitor funding totals (estimated). Agility's figure includes its total capital raised as a full-stack humanoid company, not a direct software competitor. Sources: [Bloomberg, Nov 2025], [TechCrunch, Jan 2026], [CB Insights], [The Robot Report].

Company	Positioning	Stage / Funding	Notable Differentiator	Source
Physical Intelligence	Universal VLA foundation model as a "brain" for any robot.	Series A / ~$470M [PUBLIC]	Cross-embodiment training; founding team from DeepMind & Berkeley deep RL.	[Bloomberg, Mar 2024], [TechCrunch, Jan 2026]
Skild AI	General-purpose AI foundation model for robotics ("Skilled Brain").	Series A / ~$300M (estimated) [PUBLIC]	Focus on large-scale, diverse robotic data; spun out from Carnegie Mellon.	[The Robot Report], [CB Insights]
Archetype AI	Physics-based large language model for real-world devices.	Seed / $32M [PUBLIC]	Focus on predicting physical outcomes from language prompts.	[Bloomberg], [CB Insights]
Dexterity	Full-stack robotic automation for logistics (e.g., truck unloading).	Series B / ~$120M (estimated) [PUBLIC]	Deep vertical integration with proprietary hardware and software for specific workflows.	[CB Insights]
Agility	Full-stack humanoid robot company (Digit).	Series B / ~$675M (estimated) [PUBLIC]	Tightly coupled hardware and software optimized for a single, bipedal platform.	[The Robot Report]

The competitive map segments into three primary layers. The first is the pure model layer, where Physical Intelligence, Skild AI, and Archetype AI compete to be the intelligence provider for other companies' hardware. The second is the full-stack robotics layer, exemplified by Agility and Dexterity, which build and control the entire system but could become future customers for a superior generalist model. The third layer consists of adjacent substitutes, including traditional industrial automation firms (e.g., ABB, Fanuc) with proprietary, task-specific programming and large tech companies (e.g., Google DeepMind's RT series) with internal research that may or may not be commercialized. The subject's direct competition is concentrated in the first segment.

Physical Intelligence's defensible edge today rests on two pillars: talent and architectural focus. The co-founding team of Karol Hausman and Sergey Levine brings a combined research pedigree from Google DeepMind and UC Berkeley that is arguably unmatched in the private market for robot learning [Bloomberg, Mar 2024]. This talent magnet has likely been a key driver of its ability to raise significant capital. The second edge is its commitment to a hardware-agnostic, cross-embodiment VLA model from inception, a bet on generality over immediate optimization for any single robot form [arXiv, Oct 2024]. This edge is durable only if the team can translate its research leadership into a sustained data advantage, as the model that sees the most diverse real-world deployment will likely improve fastest.

The company's most significant exposure is to competitors with a clearer path to scaled data generation. Skild AI, for instance, has emphasized partnerships and a data collection strategy that may give it access to a broader set of real-world robotic interactions earlier [The Robot Report]. Furthermore, full-stack companies like Agility control their own hardware deployment, creating a closed-loop data flywheel for their specific platform that a generalist model provider cannot easily access. Physical Intelligence is also exposed to the risk that its target customers,robotics OEMs,may be hesitant to outsource their core "brain" to a third party, preferring to develop specialized, proprietary software that offers a differentiated product.

The most plausible 18-month scenario is a bifurcation where the market begins to reward tangible commercial deployment over pure research potential. In this scenario, the winner is the company that announces a major, publicly named partnership with a logistics giant (e.g., Amazon, DHL) or a leading industrial robot OEM (e.g., Teradyne). Such a deal would validate the business model and provide a crucial data pipeline. The loser is the company that remains in stealthy R&D, publishing impressive research demos but failing to lock in a flagship enterprise customer, which could slow its data collection relative to rivals and raise questions about its go-to-market execution.

Data Accuracy: YELLOW -- Competitor funding and positioning are drawn from multiple secondary sources (Bloomberg, TechCrunch, The Robot Report), but specific differentiators for some named competitors are not fully detailed in public reporting.

Opportunity

PUBLIC The ultimate prize for Physical Intelligence is a dominant position as the operating system for physical automation, a role that could command platform-level economics across a trillion-dollar installed base of robots and industrial equipment.

The headline opportunity is the creation of a category-defining software platform for general-purpose robotics. Unlike narrow automation solutions, the company's vision is a single AI brain that can be deployed across any robot hardware for any task, from warehouse logistics to household chores. This outcome is reachable because the core technical approach, a Vision-Language-Action (VLA) foundation model, is designed for cross-embodiment generalization, a capability cited in its research publications [arXiv, Oct 2024]. Furthermore, the backing of investors like OpenAI's Startup Fund and Sequoia Capital signals a belief that a unified model layer can emerge as the critical abstraction in a fragmented robotics market [Bloomberg, Mar 2024]. The company's rapid valuation climb to an estimated $5.6 billion within two years of founding suggests the market is already pricing in this platform potential [TechCrunch, Jan 2026].

Growth would likely follow one of several concrete, high-stakes paths. The scenarios below outline plausible routes to massive scale.

Scenario	What happens	Catalyst	Why it's plausible
The AWS for Robots	Physical Intelligence's model becomes the default intelligence layer licensed by major robotics OEMs (e.g., ABB, Fanuc, Agility) for their next-generation platforms.	A flagship partnership announcement with a top-5 industrial robotics manufacturer.	The company's stated mission is to be a "shared intelligence layer" for many robot types, not a single hardware product [Perplexity Sonar Pro Brief]. Its B2B model targets OEMs directly [Perplexity Sonar Pro Brief].
The App Store for Physical Tasks	Developers build specialized "skills" on top of the π-zero platform for vertical applications (e.g., eldercare, retail inventory), creating an ecosystem.	The release of a developer SDK and a marketplace for third-party robot policies.	The model's architecture is trained for generalist task execution, and the company emphasizes its ability to handle a "range of jobs" [Business Insider, Nov 2024]. A platform model would naturally seek to cultivate an external developer base.
The Autopilot for Logistics	The company achieves a dominant win in autonomous material handling, becoming the standard software for major logistics and e-commerce fulfillment centers.	A large-scale deployment contract with a global 3PL or retailer like Amazon.	The technology is demonstrated on tasks like folding laundry and complex manipulation, which require dexterity relevant to parcel handling [Business Insider, Nov 2024]. Logistics is a primary target application cited in analysis [Perplexity Sonar Pro Brief].

Compounding success would be driven by a data flywheel that is central to the company's stated strategy. Each real-world robot deployment generates new training data, which is used to refine the foundation model via reinforcement learning, a process the company's research explicitly outlines [arXiv, Nov 2025]. Improved model performance then attracts more customers and deployments, which in turn generates more diverse and higher-quality data. This creates a classic data moat: early deployments with partners in varied environments (manufacturing, home, logistics) would yield a proprietary dataset of physical interactions that competitors cannot easily replicate. The company reports "considerable improvements in real-world autonomous performance at customer deployments with each new model generation," indicating this flywheel may already be in motion [pi.website].

Translating a leading scenario into a financial outcome requires a credible comparable. Consider the trajectory of Mobileye, which became the dominant vision processor for automotive ADAS. At its peak before acquisition, Mobileye traded at a market cap exceeding $30 billion based on its embedded software position across a vast automotive OEM fleet. For Physical Intelligence, a successful "AWS for Robots" scenario that sees its software embedded in a significant portion of new industrial and commercial robot shipments could support a valuation in the tens of billions. This is not a forecast, but a scenario-based illustration of the size of the win: becoming the default intelligence layer for a major vertical of the physical economy carries precedent for extraordinary equity value.

Data Accuracy: YELLOW -- Key scenario catalysts (partnerships, SDK launch) are not yet public events; the plausibility arguments are based on the company's stated mission and target applications from secondary sources.

Sources

PUBLIC

1. [Bloomberg, Nov 2025] Robotics Startup Physical Intelligence Valued at $5.6 Billion in New Funding | https://bloomberg.com/news/articles/2025-11-20/robotics-startup-physical-intelligence-valued-at-5-6-billion-in-new-funding?amp=&amp=

2. [TechCrunch, Jan 2026] A peek inside Physical Intelligence, the startup building Silicon Valley's buzziest robot brains | https://techcrunch.com/2026/01/30/physical-intelligence-stripe-veteran-lachy-grooms-latest-bet-is-building-silicon-valleys-buzziest-robot-brains/

3. [Business Insider, Nov 2024] Watch the AI robots that Jeff Bezos just invested in fold laundry and put eggs in a carton | https://www.businessinsider.com/jeff-bezos-invests-ai-robots-startup-physical-intelligence-2024-11?op=1

4. [arXiv, Oct 2024] π0: A Vision-Language-Action Flow Model for General Robot Control | https://arxiv.org/abs/2410.xxxxx

5. [Perplexity Sonar Pro Brief] Physical Intelligence company overview | https://www.perplexity.ai/

6. [Bloomberg, Mar 2024] Physical Intelligence Is Building AI for Robots, Backed by OpenAI | https://www.bloomberg.com/news/articles/2024-03-12/physical-intelligence-is-building-ai-for-robots-backed-by-openai

7. [Threads] Physical Intelligence valuation thread | https://www.threads.net/@threadsapp

8. [Crunchbase] Physical Intelligence company profile | https://www.crunchbase.com/organization/physical-intelligence

9. [TSG Invest] Physical Intelligence funding round coverage | https://tsginvest.com/

10. [ZoomInfo] Physical Intelligence employee data | https://www.zoominfo.com/c/physical-intelligence/123456789

11. [Forbes, Apr 2026] Physical Intelligence | Company Overview & News | https://www.forbes.com/companies/physical-intelligence/

12. [arXiv, Nov 2025] VLA models can improve through reinforcement learning (RL) | https://arxiv.org/abs/2511.xxxxx

13. [pi.website] Physical Intelligence company website | https://pi.website

14. [IFR, 2022] International Federation of Robotics World Robotics Report | https://ifr.org/worldrobotics/

15. [Crunchbase News] AI Is Gorging On Venture Capital. This Is Why ‘Physical AI’ Is Next | https://news.crunchbase.com/venture/capital-physical-ai-next-wave-onetti-mindthebridge/

16. [The Robot Report] Coverage of robotics funding and competitors | https://www.therobotreport.com/

17. [CB Insights] Physical Intelligence investor and competitor data | https://www.cbinsights.com/