Sanctuary AI
Last reviewed
Sources
34 citations
Review status
Source-backed
Revision
v8 · 6,003 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
34 citations
Review status
Source-backed
Revision
v8 · 6,003 words
Add missing citations, update stale details, or suggest a clearer explanation.
Sanctuary AI (formally Sanctuary Cognitive Systems Corporation) is a Canadian artificial intelligence and robotics company headquartered in Vancouver, British Columbia, that builds general-purpose humanoid robots powered by its proprietary Carbon AI control system.[1] Founded in January 2018 as a spin-off from Kindred Systems,[5] Sanctuary AI is best known for Phoenix, a humanoid robot that the company says can learn a new task in under 24 hours, roughly a 10x improvement over earlier generations, positioning it as embodied labor aimed at a global labor shortage the company estimates exceeds US$10 trillion in economic impact.[21][10] Sanctuary AI's stated mission is to create the world's first human-like intelligence in general-purpose robots, with the goal of addressing labor shortages and advancing research toward artificial general intelligence (AGI) through physical embodiment.[17]
The company is best known for its Phoenix robot platform, which has gone through eight generations of development as of early 2025.[10] Phoenix was recognized as one of TIME magazine's Best Inventions of 2023[7] and was the first humanoid general-purpose robot to be deployed in a commercial setting.[18] The company was co-founded by Geordie Rose, Suzanne Gildert, Olivia Norton, and Ajay Agrawal.[5] As of early 2026, Sanctuary AI employs approximately 163 people and is led by interim CEO James Wells following leadership transitions in 2024.[32]
Sanctuary AI traces its origins to Kindred Systems Inc. (formally Kindred AI), a Vancouver-based robotics company that used reinforcement learning in production robots for warehouse automation. Kindred was ranked number 29 on MIT Technology Review's Smartest Companies list in 2017 and was eventually sold to UK-based online grocery company Ocado Group in 2020.
Geordie Rose, a theoretical physicist who earned his PhD from the University of British Columbia, co-founded D-Wave Systems in 1999 and led the company as it pioneered commercial quantum computing, selling systems to organizations including NASA and Google. In 2014, D-Wave was ranked number 40 on MIT Technology Review's list of the 50 Smartest Companies. After leaving D-Wave, Rose became CEO of Kindred AI.
Suzanne Gildert, who earned a PhD in experimental physics from the University of Birmingham, had worked at D-Wave developing machine learning applications and became fascinated by the idea that humans could provide data to teach robots to think and act like people by remotely operating machines.[28] While both Rose and Gildert were at Kindred, they grew increasingly convinced that the commercial robot business and the pursuit of artificial general intelligence (AGI) belonged in separate entities.
In January 2018, Kindred announced the spin-off of its AGI research division into a new entity called Sanctuary.[5] The rationale for the split was that Kindred's product division and AGI research division had been operating increasingly independently, and separating them would maximize the likelihood of success for both. An amicable split was negotiated, and Sanctuary Cognitive Systems Corporation was incorporated in January 2018. Sanctuary licensed some of Kindred's patents and software, and Kindred maintained a minority ownership stake in the new company.[5]
The four co-founders brought complementary expertise. Rose served as CEO, bringing his experience in deep-tech entrepreneurship. Gildert took on the role of Chief Technology Officer, leading the technical vision around cognitive architectures. Olivia Norton, who had led the Artificial General Intelligence group at Kindred AI and previously worked at ETH Zurich's Institute for Biomechanics, became Chief Product Officer. Ajay Agrawal, a professor of strategic management at the University of Toronto's Rotman School of Management and founder of the Creative Destruction Lab, joined as a board director, bringing academic rigor and connections to the broader Canadian AI ecosystem.
During its first several years, Sanctuary AI focused on developing its core technology stack, including the Carbon AI control system and early prototypes of humanoid robots. The company built approximately 50 robots during this period, including five humanoid models across its first several generations. Sanctuary operated largely in stealth mode while refining its approach to embodied AI and developing the mechanical systems needed for human-like dexterity. The company pursued a vertically integrated approach, designing and building its own motors, hydraulic actuators, sensors, and AI algorithms in-house.
In March 2022, the company closed an oversubscribed Series A funding round of C$75.5 million (approximately US$58.5 million).[3] Investors included Bell Canada (BCE Inc.), Evok Innovations, Export Development Canada, Magna International, SE Health, Verizon Ventures, and Workday Ventures.[3] The diversity of investors reflected the broad potential applications of general-purpose robots across telecommunications, healthcare, manufacturing, automotive, and enterprise sectors.
Later that year, in November 2022, Sanctuary received a C$30 million contribution from the Government of Canada's Strategic Innovation Fund (SIF).[4] The SIF contribution supported a C$120 million project to design general-purpose industrial robots with human-like physical and cognitive features.[4] This government backing signaled confidence in Sanctuary's potential to position Canada as a leader in the emerging humanoid robotics industry.
In early 2023, Sanctuary AI achieved a significant milestone by completing the first commercial deployment of a humanoid general-purpose robot.[18] The deployment took place at a Mark's retail store (owned by Canadian Tire Corporation) in Langley, British Columbia. During a week-long pilot, the robot successfully completed more than 110 retail-related tasks, including picking and packing merchandise, cleaning, tagging, labelling, and folding garments, representing roughly 40% of all tasks performed at the store.[18] A second pilot followed at a SportChek store, also owned by Canadian Tire Corporation.
On May 16, 2023, Sanctuary publicly unveiled Phoenix, its sixth-generation humanoid robot, which attracted widespread attention.[2] Standing at 170 cm (5 feet 7 inches) tall and weighing approximately 70 kg (155 pounds), Phoenix was described as the world's first humanoid general-purpose robot powered by Carbon.[6] Later that year, Phoenix was named one of TIME magazine's Best Inventions of 2023, the only humanoid general-purpose robot on the list.[7] Sanctuary AI was also recognized by LinkedIn as a Canadian Top Startup of 2023.
In 2024, the company accelerated its pace of development and partnerships. In March 2024, Accenture Ventures made a strategic investment in Sanctuary AI through its Project Spotlight program, which connects emerging technology startups with Accenture's Global 2000 client base.[14] While the investment amount was not disclosed, Accenture identified post and parcel, manufacturing, retail, and logistics warehousing as key sectors where Sanctuary's robots could complement human workers.[14]
In April 2024, Sanctuary announced a strategic partnership with Magna International, one of the world's largest automotive suppliers, to deploy Phoenix robots in Magna's manufacturing facilities.[9] Magna had been an investor since 2021 and deepened the relationship through a multi-faceted collaboration, also making a strategic equity investment in the company.[22] Alongside the Magna announcement, Sanctuary unveiled its seventh-generation Phoenix robot, which featured faster task learning (under 24 hours to automate new tasks, down from weeks in previous generations), improved range of motion, lighter weight, and a lower bill of materials cost.[21]
In May 2024, Sanctuary AI announced a research collaboration with Microsoft to accelerate AI development for general-purpose robots, leveraging Microsoft Azure cloud infrastructure.[8]
In July 2024, additional strategic financing from BDC Capital's Thrive Venture Fund and InBC Investment Corp., a strategic investment fund created by the Province of British Columbia, brought total investment in the company to over C$140 million.[15]
Sanctuary AI experienced significant leadership transitions in 2024. In April 2024, co-founder and Chief Technology Officer Suzanne Gildert departed the company to pursue her interest in AI safety, AI ethics, and robot consciousness.[24] She subsequently founded Nirvanic Consciousness Technologies Inc., described as a quantum-AI deep-tech venture focused on consciousness technology.[29] Fellow co-founder Olivia Norton assumed the roles of CTO and Chief Product Officer.
In November 2024, the company's board of directors removed co-founder and CEO Geordie Rose from his position.[13] In an email to investors, CFO Philip Smith stated that the board had decided to remove Rose and "separate him from the company," though no specific reasons were made public.[27] James Wells, the company's Chief Commercial Officer who had been with Sanctuary for five years, was appointed as Interim CEO.[27] Following Rose's departure, approximately 30 employees were laid off, including software engineers and marketing staff.[12] The board formed a search committee to find a permanent CEO replacement.
The departures of both co-founders within a seven-month period represented a significant turning point for the company, though the board's actions suggested active governance and a determination to steer the company in a new direction.
Following his departure, Rose began work on Snowdrop Quantum Applications, a quantum computing research project, and was appointed as a strategic advisor to ExperienceFlow.AI in November 2025.[33]
In January 2025, Sanctuary AI released the eighth generation of its Phoenix robot, optimized for high-quality data capture and improved manufacturing efficiency.[10] The new generation featured improved cameras and telemetry systems, enhanced person-robot interaction capabilities, significant hardware improvements increasing range of motion in the wrists, hands, and elbows, miniaturized hydraulics reducing weight and power consumption, and improved visual acuity and tactile sensing.[10] The company also pursued additional funding, undertaking a US$10 million convertible note offering to finance operations through fiscal 2025.
In February 2025, Sanctuary integrated new tactile sensors into Phoenix's finger pads, with each pad containing a seven-cell touch sensor array using micro-barometers capable of detecting forces as low as 5 millinewtons, approaching human-level sensitivity.[19]
In March 2025, Sanctuary announced its use of NVIDIA Isaac Lab, an open-source framework for robot learning built on NVIDIA Isaac Sim, for accelerating dexterous learning in simulation, training thousands of virtual hands simultaneously through reinforcement learning.[11]
By early 2026, signs of recovery emerged as the company posted 19 open positions ranging from Staff Research Scientists in dexterous manipulation to Machine Learning Engineers, with most roles based on-site in Vancouver.[32] As of early 2026, the company employs approximately 163 people.[32]
| Person | Role | Background |
|---|---|---|
| Geordie Rose | Co-founder, former CEO (departed November 2024) | PhD in theoretical physics from the University of British Columbia. Previously co-founded D-Wave Systems (1999), the world's first commercial quantum computing company, and served as CEO of Kindred Systems.[13] |
| Suzanne Gildert | Co-founder, former CTO (departed April 2024) | PhD in experimental physics from the University of Birmingham. Experimental quantum physicist who worked at D-Wave Systems before co-founding Kindred Systems. Departed to found Nirvanic Consciousness Technologies.[29] |
| Olivia Norton | Co-founder, CTO and CPO | Took over as CTO after Gildert's departure in April 2024. Former AGI group lead at Kindred AI; previously worked at ETH Zurich's Institute for Biomechanics. MEng from UBC; BSc from University of Calgary. |
| James Wells | Interim CEO (appointed November 2024) | Former Chief Commercial Officer at Sanctuary AI with five years at the company. Appointed interim CEO following Rose's departure.[27] |
| Ajay Agrawal | Co-Founder, Board member | Professor at the University of Toronto's Rotman School of Management and founder of the Creative Destruction Lab. Member of the Order of Canada (2022). |
Geordie Rose's career trajectory is notable for spanning three pioneering technology companies. He co-founded D-Wave Systems in 1999, which became the first company to sell quantum computers to organizations such as NASA and Google. He then co-founded Kindred Systems, which became the first robotics company to deploy reinforcement learning in a production environment. Sanctuary AI represented his third major venture, applying lessons from both quantum computing and AI robotics to the challenge of building general-purpose humanoid robots.
Phoenix is Sanctuary AI's flagship humanoid robot platform, designed as a general-purpose robot capable of performing a wide range of work tasks. The robot has gone through eight generations of hardware iterations since the company's founding, with each generation introducing improvements in capability, manufacturing cost, and learning speed.
| Specification | Details |
|---|---|
| Height | 170 cm (5 ft 7 in) |
| Weight | 70 kg (155 lbs) |
| Payload capacity | 25 kg (55 lbs) |
| Hand degrees of freedom | 20 to 21 per hand |
| Hand actuation | Hydraulic |
| Locomotion | Wheeled base (not bipedal) |
| Sensors | Depth cameras, vision cameras, force-torque sensors, proprietary tactile/haptic sensors |
| AI control system | Carbon |
| Task learning time | Less than 24 hours (Gen 7+) |
Phoenix stands at 5 feet 7 inches tall and weighs 155 pounds, dimensions deliberately chosen to approximate average human proportions.[6] The robot can carry up to 55 pounds, making it suitable for a range of physical tasks in retail, manufacturing, and logistics environments.[6]
Unlike many competing humanoid robots that prioritize bipedal locomotion, Phoenix uses a wheeled base for movement. This design decision was informed by customer feedback indicating that bipedal legs are too fragile to support the strong upper body needed for precise and safe industrial work. As Geordie Rose explained in an interview with IEEE Spectrum: "Legs are nowhere near as important as hands, so in our strategy for rolling out the product, we're perfectly fine using wheels."[16] The company has stated that "value as a worker is not primarily defined by legs," emphasizing that manipulation capability is more critical than walking ability for most practical applications. By using a wheeled platform, Sanctuary prioritizes stability, safety, and dexterity over bipedal locomotion, focusing engineering resources on the upper body, arms, and hands where fine manipulation occurs.
One of Sanctuary AI's most significant technical achievements is its robotic hands, which the company considers industry-leading. Each hand features 20 to 21 degrees of freedom with hydraulic actuation, rivaling the dexterity and fine manipulation capabilities of the human hand. The hands incorporate proprietary haptic technology that enables the robot to perceive tactile feedback, mimicking the human sense of touch.
In February 2025, Sanctuary integrated a new generation of tactile sensors into Phoenix's finger pads.[19] Each finger pad contains a seven-cell touch sensor array built with micro-barometers. These sensors can detect forces as low as 5 millinewtons (mN), which approaches human-level tactile sensitivity.[19] This tactile capability is critical because, without it, robots must rely solely on visual feedback to interact with objects. Vision-only systems often struggle with tasks requiring precise grip control, leading to inefficient grasping and re-grasping attempts. The hands can sense pressure down to 5 millinewtons, enabling tasks that require distinguishing between objects of different fragility, such as handling a glass cup versus a plastic one of the same appearance.[31] Sanctuary has demonstrated in-hand object reorientation under disturbance, including recovering after an unexpected 500-gram load, a manipulation feat the company says it was the first commercial humanoid to achieve.[20]
The company's use of hydraulic actuation in the hands is distinctive among humanoid robot developers, providing a combination of high strength, speed, and controllability. Sanctuary has stated that its hands possess kinematics "beyond human capability," meaning the mechanical range of motion in certain axes exceeds what human hands can achieve. Sanctuary's solutions are designed, built, owned, and patented by the company through vertical integration, spanning from motor design to generative AI algorithms.
Phoenix has evolved through eight generations of hardware.
| Generation | Approximate Date | Key Developments |
|---|---|---|
| Gen 1 to 4 | 2018 to 2022 | Early prototypes and research platforms developed during the company's stealth period |
| Gen 5 | January 2023 | First commercially deployed humanoid general-purpose robot; completed 110+ tasks at Mark's retail store[18] |
| Gen 6 | May 2023 | Publicly unveiled Phoenix brand; named one of TIME's Best Inventions of 2023; 5'7", 155 lbs, 20 DOF hands[2] |
| Gen 7 | April 2024 | Task learning reduced from weeks to under 24 hours; 10% weight reduction; miniaturized hydraulics; 100x more AI parameters; improved vision and tactile sensing; lower BOM cost[21] |
| Gen 8 | December 2024/January 2025 | Optimized for high-quality data capture; improved cameras and telemetry; enhanced person-robot interaction; further cost and manufacturing improvements; integrated advanced tactile sensors[10] |
The pace of iteration accelerated significantly starting in 2023, with the company releasing three new generations (6, 7, and 8) within approximately 20 months.
Carbon is Sanctuary AI's proprietary AI control system that serves as the "brain" of the Phoenix robot. Named to evoke the carbon-based intelligence of the human brain, Carbon is designed as a cognitive architecture that mimics subsystems found in the human brain, including memory, sight, sound, and touch. Carbon translates natural language instructions into physical actions in the real world.
Carbon is not a single algorithm but rather a suite of interconnected AI models working together. The system combines multiple approaches to artificial intelligence.
| Component | Function |
|---|---|
| Symbolic reasoning | Structured task planning and explainable decision-making |
| Large language models | General knowledge and natural language processing |
| Deep learning | Perception, visual understanding, and pattern recognition |
| Reinforcement learning | Motor control and skill acquisition |
| Physics simulation | Training in virtual environments before physical deployment |
This hybrid approach enables Carbon to process information through four key stages: understanding natural language instructions, planning task sequences, executing physical actions, and learning from experience. The reasoning and task plans generated by Carbon are both explainable and auditable, which is a differentiator from purely neural network-based approaches. When Phoenix makes a decision (reaching for a part, grasping it in a particular way, placing it at a specific location), Carbon can explain why it chose that plan. This explainability is valuable in industrial settings where understanding robot behavior is important for safety and process optimization.
Carbon enables Phoenix to understand natural language instructions, plan task execution sequences, control fine motor movements, adapt to new objects and environments, and provide explainable reasoning for its decisions. The system integrates visual and haptic data through Sanctuary's patented tactile sensor technology, allowing Carbon to understand the physical properties of objects through touch and adjust its grip and manipulation strategy accordingly.
Building on the foundation of large language models (LLMs), Sanctuary AI is developing what it calls "Large Behavior Models" (LBMs). While LLMs process and generate text, LBMs are designed to ground AI in the physical world by enabling systems to understand and learn from real-world experience. These models are trained on high-quality behavioral data collected from Phoenix robots operating in real environments and through teleoperation.
The concept reflects Sanctuary's broader thesis that achieving human-like intelligence requires physical embodiment.[17] By interacting with the physical world, the AI system can develop a richer understanding of cause and effect, spatial relationships, and the properties of objects.
Both Carbon and the dexterous robotic hands are designed to be modular, allowing for potential integration with a range of non-humanoid and humanoid robots beyond Phoenix. Because Carbon mimics subsystems of the human brain in a modular fashion, it is theoretically vendor-agnostic, meaning the control system could be deployed on humanoid platforms built by other manufacturers. This positions Carbon not only as the control system for Phoenix but as a potentially licensable platform for the broader robotics industry.
Sanctuary AI's approach to building robot intelligence is built on a layered autonomy strategy that begins with human-controlled teleoperation and progressively adds autonomous capabilities over time.
In teleoperation mode, a human pilot controls a Phoenix robot using a rig that transmits the pilot's physical movements to the robot. The robot moves in the same way as the operator, while the robot's senses (visual, tactile, and force feedback) are transmitted back to the pilot. This creates a bidirectional loop where the human provides motor intelligence while experiencing the robot's sensory environment.[16]
Teleoperation serves two purposes simultaneously. First, it is a production mode that allows skilled operators to handle complex tasks remotely, providing immediate economic value. Second, every teleoperated session generates training data that can be used to improve autonomous performance. This dual function means that productive work and AI training happen in parallel.
From the grasping data collected through teleoperation, Sanctuary's researchers have discovered that common hand movements emerge as recurring patterns called Eigengrasps. These are fundamental building blocks that, when combined in different ways, allow hands to perform a vast range of manipulation tasks. Identifying these Eigengrasps enables more efficient training of autonomous behaviors by reducing the dimensionality of the grasping problem.
Sanctuary AI employs a combination of teleoperation and simulation to train its robots. Human operators can remotely control Phoenix robots to demonstrate tasks, generating behavioral data that the AI system learns from. Additionally, the company uses NVIDIA Isaac Lab, an open-source simulation framework, to simulate dexterity-focused training environments, enabling online reinforcement learning in simulation before transferring learned policies to physical robots.[11] This simulation-to-reality (sim-to-real) approach allows Sanctuary to train thousands of simulated hands concurrently, dramatically accelerating the learning process compared to training on physical robots alone.[20]
Using simulation, Sanctuary can train thousands of virtual hands simultaneously, dramatically accelerating the learning process compared to training on physical robots alone.[11] This approach also enables the learning algorithms to explore the full range of the hands' mechanical capabilities, including movements that exceed what human teleoperators can demonstrate. The company has reported achieving hand kinematics in simulation that go "beyond human capability," suggesting that the simulated training environment allows the discovery of manipulation strategies that human operators might not intuitively perform. In Generation 7, these improvements reduced the time required to teach Phoenix a new task from weeks to under 24 hours, which the company has described as roughly a 10x leap over previous generations.[21]
Sanctuary's roadmap envisions a gradual transition from primarily teleoperated systems to fully autonomous robots. Rather than attempting to solve full autonomy in a single step, the company adds increasing layers of autonomous capability on top of the teleoperation foundation. This pragmatic approach allows Sanctuary to deliver value to customers at every stage of the autonomy spectrum while continuously collecting the data needed to push toward greater independence.
Sanctuary AI has built one of the most significant patent portfolios in the humanoid robotics industry. In February 2025, Morgan Stanley's Research division ranked Sanctuary AI third globally in published humanoid-related patent filings, though the report identified only 51 of the company's approximately 140 published US patent filings.[30]
The patent portfolio covers several key technology areas:
| Patent Area | Description |
|---|---|
| Software-compensated robotics | Techniques for using software to improve mechanical precision |
| Multi-purpose robot grasping | Systems, devices, and methods for grasping by multi-purpose robots |
| Haptic photogrammetry | Methods for mapping between visual and haptic data in robots |
| Visual servoing | Real-time visual feedback for guiding robot manipulation |
| Grasping simulation | Real-time simulation of the grasping process |
Sanctuary has strengthened its IP position through strategic acquisitions, including purchasing the entire patent portfolio of Giant.AI and acquiring assets from Tangible Research. These acquisitions deepened the company's intellectual property and expertise in hand dexterity and fine manipulation, areas that Sanctuary considers central to the commercial viability of general-purpose robots.
The company's leadership has emphasized that intellectual property will be a key competitive advantage in the physical AI market, drawing parallels to how IP assets became immensely valuable during the smartphone revolution.[34]
Sanctuary AI's primary business model is "labor-as-a-service," in which general-purpose robots integrate into existing workforces to fill gaps caused by labor shortages.[23] Rather than selling robots outright, this model allows organizations to deploy robotic labor flexibly. This model mirrors the shift in software from licensed products to subscription services and could lower adoption barriers for customers by converting large upfront capital expenditures into ongoing operational costs. Bell Canada's 5G network has been identified as a key enabler for this model across Canada, providing edge computing capabilities and secure connectivity for remote robot operation.
The estimated price for a Phoenix robot is approximately $40,000 depending on custom integration, though the company's focus remains on the service model rather than unit sales. The company targets industries facing persistent labor shortages, including manufacturing, retail, logistics, and warehousing. By designing Phoenix as a general-purpose robot that can learn new tasks in under 24 hours, Sanctuary aims to offer flexibility that traditional single-purpose industrial robots cannot match.
Sanctuary AI's philosophical approach to artificial general intelligence is rooted in the belief that true human-like intelligence requires physical embodiment. The company's thesis is that a machine capable of performing all economically valuable work that humans do must possess a mind with many, if not all, of the properties of the human mind.[17] By building humanoid robots that interact with the physical world in the same way humans do, Sanctuary aims to create the conditions necessary for AGI to emerge.
This perspective positions the humanoid robot as a means to an end rather than the end itself. The company's goal is not merely to build useful robots but to develop human-level intelligence by grounding AI systems in physical reality. This approach aligns with a growing body of research in embodied AI and embodied cognition, which argues that intelligence is fundamentally shaped by having a body that interacts with the environment.
In April 2024, Sanctuary AI announced a strategic partnership with Magna International, one of the world's largest automotive parts suppliers.[9] Magna manufactures and assembles vehicles for automakers including Mercedes-Benz, Jaguar, and BMW. Magna had been an investor in Sanctuary since 2021. The collaboration encompasses several elements: deployment of Phoenix robots in Magna's manufacturing operations; a joint assessment of how to improve the cost and scalability of robots using Magna's automotive engineering and manufacturing capabilities; a strategic equity investment by Magna in Sanctuary AI; and plans for Magna to manufacture Phoenix robots under contract in the future.[22]
This partnership is significant because it provides Sanctuary with both a major commercial customer and a potential manufacturing partner with deep expertise in high-volume production.
In May 2024, Sanctuary AI announced a collaboration with Microsoft focused on accelerating AI development for general-purpose robots.[8] The partnership involves Sanctuary leveraging Microsoft Azure infrastructure for training, inference, networking, and storage.[8] Microsoft Research is collaborating with Sanctuary on embodied AI research in areas including reasoning, planning, and human-agent collaboration.
The collaboration is particularly focused on advancing Large Behavior Models (LBMs) for the Carbon AI system, with Microsoft providing the cloud computing resources needed to train and deploy these models at scale.
Canadian Tire Corporation (CTC) was Sanctuary AI's first commercial partner. Through CTC's retail brands Mark's and SportChek, Sanctuary conducted pilot deployments in 2023 that demonstrated Phoenix's ability to perform retail tasks autonomously. The pilot at the Langley, British Columbia Mark's store demonstrated Phoenix performing over 110 retail tasks, providing real-world validation of the robot's capabilities outside of laboratory settings.[18] These pilots were designed to test whether general-purpose robots could alleviate monotonous tasks and free employees for higher-value work such as customer service.
Accenture's investment through its Ventures arm and Project Spotlight program connects Sanctuary with Accenture's global consulting network and client base.[14] Accenture has identified manufacturing, retail, logistics warehousing, and post and parcel operations as sectors with high potential for humanoid robot deployment.[14]
Sanctuary AI uses NVIDIA Isaac Lab for accelerating dexterous robot learning in simulation.[11] This partnership enables Sanctuary to train its robotic hands in physics-realistic virtual environments, running thousands of simultaneous training instances to develop manipulation skills more efficiently than would be possible with physical robots alone.[20]
Sanctuary AI has raised over C$140 million in total funding from a diverse group of investors, including private venture capital, strategic corporate investors, and government sources.[15]
| Date | Round/Source | Amount | Key Investors |
|---|---|---|---|
| 2018 to 2021 | Seed and early rounds | Undisclosed | Early-stage investors |
| March 2022 | Series A | C$75.5 million (US$58.5 million) | Bell Canada, Evok Innovations, Export Development Canada, Magna International, SE Health, Verizon Ventures, Workday Ventures[3] |
| November 2022 | Government of Canada Strategic Innovation Fund | C$30 million | Government of Canada[4] |
| March 2024 | Strategic investment | Undisclosed | Accenture Ventures[14] |
| April 2024 | Strategic equity investment | Undisclosed | Magna International[22] |
| July 2024 | Strategic financing | Undisclosed | BDC Capital (Thrive Venture Fund), InBC Investment Corp.[15] |
| January 2025 | Convertible note offering | Up to US$10 million (target) | Various investors |
In January 2025, the company's CFO indicated that Sanctuary was undertaking a US$10 million convertible note offering to finance operations through fiscal 2025, reflecting the capital-intensive nature of humanoid robotics development.
While the total exceeds $140 million, this figure is modest compared to some U.S.-based competitors in the humanoid robot space. Figure AI, for example, has raised over $675 million, and Tesla has committed significant internal resources to its Optimus humanoid robot program.
Sanctuary AI operates in the rapidly growing humanoid robot market, which was valued at approximately $2.92 billion in 2025 and is projected to reach $15.26 billion by 2030. The company faces competition from more than 20 companies actively developing humanoid robots as of 2026.
| Company | Robot | Headquarters | Key Differentiator |
|---|---|---|---|
| Sanctuary AI | Phoenix | Vancouver, Canada | General-purpose AI intelligence (Carbon); industry-leading dexterous hands; wheeled base; labor-as-a-service model |
| Tesla | Optimus | Austin, USA | Massive scale ambitions (targeting 100,000 units by 2026, 1 million long-term); leveraging Tesla's manufacturing and AI expertise |
| Figure AI | Figure 02 | Sunnyvale, USA | Bipedal; partnerships with BMW and OpenAI; raised over US$1.6 billion |
| Agility Robotics | Digit | Corvallis, USA | Bipedal; RoboFab factory in Oregon capable of 10,000 units per year; deployed at Amazon warehouses |
| Boston Dynamics | Atlas | Waltham, USA | Advanced bipedal mobility; research-focused heritage; transitioned to electric Atlas in 2024 |
| 1X Technologies | NEO | Moss, Norway | Backed by OpenAI; consumer-oriented humanoid design; household and commercial applications |
| Apptronik | Apollo | Austin, USA | Designed for logistics and manufacturing; partnership with Mercedes-Benz |
| Unitree | G1/H1 | Hangzhou, China | Low-cost humanoids starting at US$16,000 |
Sanctuary AI differentiates itself from competitors through several factors: its focus on hand dexterity over bipedal locomotion, its hybrid Carbon AI architecture combining symbolic reasoning with deep learning, its labor-as-a-service business model, its extensive patent portfolio (ranked third globally by Morgan Stanley),[30] its data-driven teleoperation approach to building autonomous skills, and its philosophical commitment to achieving AGI through embodiment. The company's Canadian headquarters and government support also position it as a national champion in the Canadian AI ecosystem. However, the company's relatively limited funding compared to competitors like Figure AI and Tesla, combined with the leadership upheaval of 2024, presents challenges for scaling.
| Year | Award/Recognition |
|---|---|
| 2023 | TIME's Best Inventions of 2023 (Phoenix robot)[7] |
| 2023 | LinkedIn Top Startup (Canada) |
| 2025 | Morgan Stanley Research: ranked third globally in humanoid-related patent filings[30] |
Sanctuary AI faces several challenges as it looks ahead. The departure of two of its three co-founders in 2024, followed by staff layoffs, introduced uncertainty about the company's direction and stability.[12] The capital-intensive nature of humanoid robotics means that continued fundraising will be essential, and the company's total funding of approximately $140 million, while significant, trails several U.S.-based competitors.
On the other hand, the company's strategic partnerships with Magna International and Microsoft provide both commercial deployment pathways and access to world-class manufacturing and cloud computing resources. The Magna partnership in particular could prove transformative, as it offers a path to high-volume manufacturing of Phoenix robots using Magna's automotive production expertise.
The broader humanoid robot industry entered a period of rapid growth in 2025, with multiple companies moving from prototypes to commercial deployments. Sanctuary AI's early mover advantage in commercial deployment (first in January 2023) and its sophisticated Carbon AI system position it as a notable player in this emerging market, though the company will need sustained investment and stable leadership to compete against increasingly well-funded rivals.