Skild AI

Skild AI is an American robotics-AI foundation-model company headquartered in Pittsburgh, Pennsylvania, founded in 2023 by Deepak Pathak and Abhinav Gupta, both senior faculty members at the Carnegie Mellon University School of Computer Science and senior figures in the academic robot-learning research community. Pathak's research at CMU's Robotics Institute has emphasized self-supervised robot learning and embodied curiosity-driven exploration; Gupta has been a long-tenured CMU robotics professor with substantial publication output on visual representation learning and robotic manipulation. Skild AI develops the Skild Brain, a general-purpose robotic intelligence foundation model that the company is building to operate across diverse robot hardware platforms — humanoid robots, quadrupeds, manipulators, autonomous vehicles, and other embodied systems — rather than being specialized for a particular form factor or task. The Skild Brain positioning frames robotics as a foundation-model problem in the same way that natural language was reframed in 2018 to 2023, with the bet that a single large model trained on heterogeneous robot data will outperform per-platform per-task specialized systems. As of April 2026, Skild AI is one of the two principal robotics-AI foundation-model Insurgents globally alongside Physical Intelligence.

At a glance

• Founded: 2023 in Pittsburgh, Pennsylvania, by Deepak Pathak and Abhinav Gupta.
• Status: Private. Series A in July 2024 at $1.5 billion post-money valuation. Subsequent rounds reported through 2025 at higher valuations approaching $4 billion.
• Funding: Approximately $300 million Series A in July 2024 led by Lightspeed Venture Partners with Coatue, SoftBank, Bezos Expeditions, Sequoia Capital, Felicis Ventures, Menlo Ventures, General Catalyst, and additional investors. Cumulative capital exceeds $300 million; subsequent rounds reported through 2025.
• CEO: Deepak Pathak, Co-Founder and Chief Executive Officer. CMU School of Computer Science Assistant Professor (concurrent appointment); IIT Kanpur and UC Berkeley educational background; long-running publication record at NeurIPS, ICML, ICLR, RSS, and CoRL on robot-learning and self-supervised representation learning.
• Other notable leadership: Abhinav Gupta, Co-Founder. CMU Robotics Institute Associate Professor (concurrent appointment); long-tenured CMU faculty member with publication record on visual representation learning, embodied learning, and robotic manipulation.
• Open weights: No. Skild AI's models are closed and the company has not released the Skild Brain weights or training infrastructure publicly.
• Flagship outputs: Skild Brain — the company's general-purpose robotic intelligence foundation model, demonstrated through public videos in 2024 and 2025 across multiple robot hardware platforms including quadrupeds, humanoid robots, and manipulator arms.

Origins

Skild AI was founded in 2023 by Deepak Pathak and Abhinav Gupta with a research thesis grounded in their academic work on robot-learning at scale. Pathak had been one of the principal researchers on self-supervised robot learning, with publications including the influential "Curiosity-driven Exploration by Self-supervised Prediction" (ICML 2017) and substantial subsequent work on robot policy learning from heterogeneous data sources. Gupta had been a long-tenured CMU Robotics Institute faculty member with publication output spanning visual representation learning, scaling laws for embodied learning, and the broader question of how robotic capability scales with data, compute, and model size.

The founding thesis was that robotics had reached a moment analogous to language modeling in 2018 to 2020 — that the convergence of substantial diverse robot training data, scaled compute, and modern foundation-model architectures would enable a general-purpose robot intelligence model that could operate across hardware platforms rather than being specialized per platform. The Skild Brain positioning frames the company as a model-and-software-platform layer that hardware partners (humanoid-robot manufacturers, quadruped manufacturers, manipulator manufacturers, automotive OEMs) can integrate into their products rather than a vertically integrated robot-and-AI hardware company.

The July 2024 Series A of $300 million at $1.5 billion post-money valuation was led by Lightspeed Venture Partners with Coatue, SoftBank, Bezos Expeditions, Sequoia Capital, Felicis Ventures, Menlo Ventures, General Catalyst, and additional investors participating. The round was one of the larger AI Series A rounds of the period and established Skild AI as one of the two principal robotics-AI foundation-model Insurgents alongside Physical Intelligence (which had announced its own $400 million Series A at $2.4 billion valuation in November 2024).

The 2024 to 2026 period has seen Skild AI demonstrate Skild Brain capabilities across multiple robot hardware platforms through public videos and customer-pilot programs. Industry coverage has reported partnerships with humanoid-robot manufacturers and adjacent robotics customers, alongside continued senior research-talent recruiting from CMU, MIT, Stanford, and frontier AI labs. Subsequent funding rounds through 2025 have been reported at higher valuations, with industry coverage indicating valuations approaching $4 billion by late 2025.

Mission and strategy

Skild AI's stated mission is to build the general-purpose intelligence layer for robots — a single foundation model that can be deployed across diverse robot hardware platforms and that improves through learning from the aggregated data of all deployed platforms. The strategy combines three threads. First, foundation-model research on the Skild Brain, with explicit emphasis on cross-platform generalization (training on data from multiple robot form factors and tasks rather than specializing per platform). Second, hardware-partner integration, providing the AI-software layer that humanoid-robot, quadruped, manipulator, and adjacent robotics-hardware manufacturers integrate into their products. Third, in-house data-collection and continual-learning infrastructure that aggregates operational data across deployed Skild Brain integrations and feeds back into model improvement.

The competitive premise is that the per-platform per-task specialized approach that has dominated industrial robotics will not scale economically to the broader robot-applications market that the post-2022 humanoid-robotics wave has opened up, and that a general-purpose model layer capturing cross-platform learning effects will deliver structurally better economics for hardware partners than building per-product specialized AI from scratch.

Models and products

• Skild Brain. General-purpose robotic intelligence foundation model. Demonstrated across multiple robot hardware platforms (quadrupeds, humanoid robots, manipulator arms) in public videos and customer-pilot programs.
• Hardware-partner integration platform. Software-and-data infrastructure that hardware partners use to integrate Skild Brain into their products.
• In-house data-collection and continual-learning infrastructure. Aggregates operational data across deployed integrations.

Distribution channels are direct hardware-partner enterprise relationships rather than direct-to-consumer products. Specific named hardware partners have been disclosed through industry coverage but the company has been comparatively quiet on commercial-customer specifics.

Benchmarks and standing

Skild AI is not evaluated against horizontal AI benchmarks because the company's commercial output is robotic-policy capability rather than language-model performance. The company's standing is measured through the public videos demonstrating Skild Brain capabilities, the hardware-partner pipeline, the senior research-talent quality, and the comparative position against Physical Intelligence (the principal direct competitor).

Industry coverage has consistently grouped Skild AI with Physical Intelligence as the two principal robotics-AI foundation-model Insurgents of the 2023 to 2024 cohort. Among that pair, Skild AI's distinguishing positioning is the CMU founder-team research credibility and the Pittsburgh headquarters that anchors the company in the broader CMU robotics-research ecosystem; Physical Intelligence's positioning is the Berkeley-and-Toyota-Research-Institute founder-team research credibility and the Bay Area headquarters.

Leadership

As of April 2026, Skild AI's senior leadership includes:

• Deepak Pathak, Co-Founder and Chief Executive Officer. CMU School of Computer Science Assistant Professor (concurrent appointment).
• Abhinav Gupta, Co-Founder. CMU Robotics Institute Associate Professor (concurrent appointment).
• Senior research and engineering leadership across the company. Industry coverage has reported recruiting from CMU, MIT, Stanford, and frontier AI labs.

Both founders maintain concurrent CMU faculty appointments alongside their Skild AI leadership roles.

Funding and backers

• Seed (2023): Reported smaller seed financing.
• Series A (July 2024): $300 million at $1.5 billion post-money valuation. Led by Lightspeed Venture Partners with Coatue, SoftBank, Bezos Expeditions, Sequoia Capital, Felicis Ventures, Menlo Ventures, General Catalyst, and additional investors.
• Subsequent rounds (2024 to 2025): Industry coverage has reported additional rounds at higher valuations, with reported valuations approaching $4 billion by late 2025.
• Cumulative capital exceeds $300 million.

Industry position

Skild AI occupies a distinctive position as one of the two principal robotics-AI foundation-model Insurgents globally alongside Physical Intelligence, with the CMU founder-team research credibility, the multi-hundred-million-dollar capital base, and the cross-platform-generalist Skild Brain model positioning. Industry coverage has consistently characterized Skild AI as one of the structurally consequential robotics-AI Insurgents of the post-2022 cohort.

The structural risks are two. First, the foundation-model thesis for robotics is unproven at production scale — whether cross-platform learning produces sufficient generalization to outperform specialized per-platform models in real commercial deployments is the principal commercial question. Second, the competitive landscape includes both peer robotics-AI Insurgents (Physical Intelligence) and frontier-AI labs that have launched embodied-AI research programs (Google DeepMind RT-2 / RT-X, NVIDIA Research GR00T, Meta AI embodied research), and the timing of commercial-deployment-quality cross-platform robot intelligence remains uncertain.

Competitive landscape

• Physical Intelligence. Direct robotics-AI foundation-model competitor. Different founder lineage (Berkeley and Toyota Research Institute) and Bay Area headquarters.
• NVIDIA Research GR00T. Strategic-and-research peer; NVIDIA is also a likely hardware-platform partner for both Skild and Physical Intelligence.
• Google DeepMind. Frontier-lab embodied-AI research program (RT-2, RT-X, AutoRT, RT-Trajectory). Different commercial structure.
• 1X, Figure AI, Boston Dynamics, Tesla Optimus, Apptronik, Agility Robotics. Hardware-platform peers and potential customers. Some have in-house AI teams; Skild's commercial premise depends on selected hardware partners adopting Skild Brain rather than building in-house alternatives.
• CMU SCS, Berkeley BAIR, Stanford AI Lab, MIT CSAIL. Academic robot-learning research peers. Founder Pathak's CMU faculty role anchors the academic-research-cooperation network.

Outlook

• Continued Skild Brain development and any disclosed hardware-partner production deployments.
• The competitive dynamic with Physical Intelligence as both companies progress toward commercial deployment.
• Continued senior research-talent recruiting from CMU and frontier AI labs.
• Potential additional fundraising at higher valuations.
• Whether the cross-platform foundation-model thesis produces measurable economic advantage over specialized per-platform alternatives in commercial deployments.

Sources

• Skild AI official site. Company reference.
• Deepak Pathak CMU page. Co-Founder reference.
• Abhinav Gupta CMU page. Co-Founder reference.
• Lightspeed Skild AI announcement. July 2024 Series A reference.
• CMU School of Computer Science. Founder home institution.