Skild AI Closes $1.4B Series C at Over $14B Valuation

Skild AI Funding Round Details

Skild AI announced its Series C on January 14, 2026, securing close to $1.4 billion at a post-money valuation exceeding $14 billion. SoftBank Group led the round, with participation from NVentures (NVIDIA), Macquarie Capital, Bezos Expeditions, 1789 Capital, and returning investors including Lightspeed Venture Partners, Felicis Ventures, Sequoia Capital, and Coatue Management. Strategic players such as Samsung Electronics, LG Electronics, Schneider Electric, and Salesforce Ventures also joined. The round more than tripled the company’s mid-2025 valuation of roughly $4.5 billion and brought total capital raised since the 2023 founding to over $2 billion.

CEO Deepak Pathak highlighted that proceeds target expansion of the company’s omni-bodied intelligence platform. Skild, based in Pittsburgh and spun from Carnegie Mellon research, has moved quickly from stealth to commercial traction. Revenue grew from zero to approximately $30 million in a few months during 2025, reaching a $100 million annual run rate by September 2026 after its first paid deployments.

S1 Model Technical Capabilities and Deployments

Skild’s S1 robotic foundation model emphasizes in-context learning, allowing a robot to interpret a single video demonstration and execute previously unseen tasks lasting up to 10 minutes without weight updates or task-specific fine-tuning. Demonstrated tasks include plant potting, pancake preparation, pour-over coffee brewing, and kit assembly, each involving dozens of manipulation steps and recovery from disturbances. In internal tests on novel multistep sequences, S1 achieved roughly 66 percent per-step success, compared with 9 percent for comparable language-prompted baselines—a more than sevenfold improvement. One short video demonstration reportedly equates to the value of about 380 hands-on training examples, which would otherwise require 50–100 hours of manual collection.

The model runs on NVIDIA infrastructure, incorporating Cosmos for synthetic data generation and video structuring, Isaac Lab and Isaac Sim for reinforcement learning and simulation, and Omniverse for validation. Joint development with NVIDIA includes GPU-accelerated solvers for contact-rich manipulation under the Newton physics engine. Real-world use began with deployments across security, construction, delivery, data centers, warehouses, and factory assembly. A notable partnership places S1 on dual-arm systems at a Foxconn facility assembling NVIDIA Blackwell GPU servers, handling busbar installation, limit block placement, and 16-screw fastening while adapting to scene variations.

Fleet Economics and Production Ramp Implications

For operators managing robot fleets, S1’s video-based adaptation reduces the traditional cycle of data collection, retraining, and validation that previously dominated integration costs. In high-mix environments such as electronics assembly or logistics sorting, the ability to demonstrate a new sequence in 11 minutes and achieve autonomous execution lowers per-task engineering overhead. Skild reports more than 60 deployment partnerships by September 2026, suggesting early density advantages for customers who can feed anonymized operational data back into model improvement under permitted agreements.

However, fleet-level metrics remain opaque. Mean time between failures (MTBF) for S1-controlled hardware depends on the underlying robot platforms—dual-arm manipulators in the Foxconn case versus quadrupeds or mobile bases elsewhere. Investors at the $14 billion valuation appear to price in rapid scaling of deployment density, yet hardware reliability, sensor calibration drift, and edge-case recovery still determine whether video prompting translates into sustained uptime above 95 percent across shifts. Production ramps at partners like Foxconn will test whether S1 can maintain performance when object positions vary or when cycle times compress below demonstration conditions.

Investor Mix and Strategic Positioning

SoftBank’s leadership signals continued emphasis on physical AI infrastructure, complementing its robotics portfolio that includes Pepper and NAO platforms. NVIDIA’s involvement through NVentures and technical collaboration aligns S1 tightly with Isaac and Cosmos ecosystems, potentially accelerating adoption among developers already using those tools. Strategic participation from LG, Samsung, and Schneider Electric points to interest in embedding the brain across consumer electronics assembly, industrial automation, and building systems.

Bezos Expeditions and Macquarie Capital add capital without obvious hardware overlap, diversifying the syndicate beyond pure robotics plays. This breadth may help Skild negotiate access to diverse robot embodiments for further S1 training, addressing the “omni-bodied” goal. Total funding now exceeds $2.2 billion, placing Skild among the highest-capitalized pure-play robot intelligence companies and ahead of many hardware-centric peers in disclosed rounds.

Challenges and Second-Order Risks

Despite strong numbers, several factors could constrain impact. S1 performance on long-horizon tasks still requires high-quality demonstration videos; noisy or ambiguous footage may degrade the 66 percent step success rate observed in controlled tests. Cross-embodiment transfer—mapping human or different-robot demonstrations to a target platform—remains an active research area, with potential gaps in gripper geometry or kinematic reach. Data-center and warehouse customers will demand integration with existing fleet management software, raising questions about latency, safety interlocks, and regulatory compliance for collaborative operation.

Revenue concentration among early partners also warrants monitoring. If deployments remain limited to a handful of high-profile sites, the $100 million run-rate trajectory could slow once initial pilots saturate. Hardware partners must simultaneously improve actuator durability and sens