Skild AI
The most heavily funded independent robot-brain startup — $2B raised at $14B+ valuation, backed by SoftBank, NVIDIA, and Bezos, building a universal AI that controls any robot.
1. Core Product / Service
Skild AI builds the Skild Brain — a general-purpose AI foundation model designed to control any robot across tasks and hardware types (quadrupeds, humanoids, robotic arms, mobile manipulators). Founded in May 2023 by two CMU robotics professors who previously led Meta's robotics lab, the company takes a software-only approach: it doesn't build robot hardware, it builds the intelligence layer that any robot can use.
The Skild Brain is trained on diverse robot data and designed to generalize across embodiments — a "unified model" philosophy similar to physical-intelligence's approach, contrasting with nvidia-groot's per-embodiment tuning. In early tests, the model was deployed on a Unitree humanoid robot and achieved results within a single day [1].
By late 2025, Skild's software was being used at NVIDIA's Houston GPU factory for automation tasks, demonstrating real-world industrial deployment [1]. The company reports revenue growing from $0 to ~$30M in 2025, serving 8+ customers.
2. Target Users & Pain Points
- Robot manufacturers: Integrate Skild Brain into their hardware rather than building perception-to-action AI stacks from scratch. This is the "Intel Inside" model applied to robotics.
- Industrial automation: Factory operators who want to deploy different robot types for different tasks but control them all through a single intelligence layer. NVIDIA's Houston GPU factory deployment is the flagship case [1].
- Logistics & warehousing: Mobile manipulation across diverse environments and task types.
The core pain point: robot hardware companies are great at mechanics but struggle with the AI software layer. Skild offers a drop-in "brain" that abstracts away the complexity of perception, planning, and action generation across robot types.
3. Competitive Landscape
| Competitor | Approach | Valuation |
|---|---|---|
| physical-intelligence|Physical Intelligence | Closed unified VLA (π series) | $5.6B → $11B target |
| nvidia-groot|NVIDIA GR00T | Open VLA platform + WBC + SONIC | Part of $3.5T+ NVIDIA |
| Galaxea (星海图) | Open G0.5 SOTA VLA + hardware | $4.6B |
| unitree|Unitree | Open VLA + hardware sales | $6.2B IPO target |
Skild's differentiators: pure software play (no hardware conflict with potential robot-maker customers), the strongest institutional backing (SoftBank + NVIDIA, 27 investor firms), and the highest independent valuation ($14B vs PI's $5.6B). The pure-software model means it can theoretically partner with any robot OEM without competing against them.
4. Unique Observations
- Valuation divorced from revenue. $14B+ valuation on ~$30M revenue is a ~467× revenue multiple — even by AI startup standards, this is extreme. The bet is not on current revenue but on becoming the "operating system for physical intelligence," analogous to what Android/iOS did for mobile.
- SoftBank-NVIDIA-Bezos alignment is the signal. When the three most aggressive AI investors (SoftBank, NVIDIA's NVentures, Bezos Expeditions) co-invest in a robotics AI company, it signals a consensus that the "robot brain" is the next platform layer worth tens of billions. Samsung, LG, and Schneider Electric joining as strategics further validates the industrial deployment thesis.
- The CMU → Meta → Skild pipeline. Both founders (Deepak Pathak and Abhinav Gupta) came from Meta's FAIR robotics lab then CMU, then left academia to start Skild. This mirrors the academic → big-tech → startup trajectory of many successful AI founders, and gives them both research credibility and deployment experience.
5. Financials / Funding
- Founded: May 2023, Pittsburgh, PA
- Total raised: ~$2B across 3 rounds [1]
- Series A (Jul 2024): $300M at ~$4.5B valuation; led by SoftBank, Lightspeed, Coatue, Bezos Expeditions, Sequoia
- Series B (Jun 2025): $135M; SoftBank, Samsung, NVIDIA, LG Tech Ventures
- Series C (Jan 2026): $1.4B at >$14B valuation; led by SoftBank, joined by NVIDIA (NVentures), Macquarie, Bezos Expeditions, 1789 Capital, and return investors [1][2]
- Revenue: $0 → ~$30M (2025); 8+ customers [1]
- Team: 100+ people, hiring from Meta, Tesla, NVIDIA, Google, Amazon [1]
- Investor count: 27 firms, including SoftBank, NVIDIA, Lightspeed, Sequoia, Coatue, Bezos Expeditions, Samsung, LG, Salesforce Ventures, Schneider Electric, Amazon (Alexa Fund & Industrial Innovation Fund), General Catalyst, Felicis, CMU [1]
6. People & Relationships
- Co-founder & CEO: Deepak Pathak — former CMU Robotics Institute Assistant Professor, previously at Meta's robotics lab
- Co-founder & President: Abhinav Gupta — CMU Robotics Institute tenured professor, founding member & Research Lead at FAIR Robotics (Meta)
- Key investors: SoftBank Group (lead across all rounds), NVentures (NVIDIA), Bezos Expeditions, Lightspeed, Sequoia, Coatue, Samsung, LG, Schneider Electric, Amazon
- Key partners: NVIDIA (Houston GPU factory deployment), LG CNS (humanoid robotics partnership)
- Key competitors: physical-intelligence, nvidia-groot
- Robot partners: unitree (humanoid robot used in early deployment demos)