Back to Blog
May 6, 2026 • Anna Heim

Genesis AI Goes Full Stack with Human-Like Robotic Hands

Genesis AI Goes Full Stack with Human-Like Robotic Hands Genesis AI Goes Full Stack with Human-Like Robotic Hands Genesis AI Goes Full Stack with Human-Like Robotic Hands

Genesis AI Goes Full Stack with Human-Like Robotic Hands

Overview

Genesis AI, a robotics startup backed by a $105 million seed round from Khosla Ventures and Eclipse, has unveiled its first foundational AI model (GENE-26.5) alongside a surprising hardware development: human-sized robotic hands designed in-house. The company's decision to go "full stack" represents a strategic pivot to gain better control over both AI intelligence and physical embodiment.

Key Technical Achievements

The GENE-26.5 Model

  • First foundational AI model specifically designed for robotics manipulation
  • Named for May 2026, with continuous iteration planned
  • Trained on massive amounts of human-based internet videos
  • Evaluation bottleneck solved through custom simulation system

Human-Scale Robotic Hands

  • Same size and shape as human hands (not traditional two-finger grippers)
  • Reduces the "embodiment gap" between simulation and real-world conditions
  • Enables direct use of human demonstration data without translation
  • Demonstrated complex tasks:
    • Cooking (cracking eggs, slicing tomatoes)
    • Preparing smoothies
    • Playing piano
    • Solving Rubik's cube
    • Lab work

Data Collection Innovation: The Sensor Glove

  • Lightweight, wearable data collection device
  • Works as a real-life double of the robotic hand
  • As light and easy to wear as security gloves used in industry
  • Relatively cheap to manufacture
  • Allows workers to collect training data during normal job activities
  • Complemented by egocentric video (workers filming themselves)

Strategic Architecture

Full-Stack Approach

"The model has always been the goal, because a better model means better intelligence," said CEO Zhou Xian. "But we decided to go full stack" to maintain control over hardware.

Three-Pillar Data Strategy

  1. Simulation system for rapid model iteration and evaluation
  2. Sensor glove for real-world human demonstration data
  3. Internet-scale video data for foundational training

Competitive Positioning

  • Unlike Physical Intelligence and Skild AI, Genesis controls the entire stack
  • Unlike ~50-100 robotic hand companies, Genesis uses human-scale hands
  • Combines hardware, AI model, and data collection in integrated system

Commercial Applications & Roadmap

Target Industries

  • Pharmaceutical labs (lab technician work)
  • Manufacturing (assembly and manipulation tasks)
  • General-purpose automation where human-like dexterity is required

Near-Term Plans

  • Full-body robot reveal coming shortly (not just hands)
  • Partnerships with customers for data collection via glove technology
  • Building a "human skill library" through third-party data partners

Team & Resources

Funding & Investors

  • $105M seed round (July 2025)
  • Led by Eclipse and Khosla Ventures
  • Additional backers: Bpifrance, HSG, Eric Schmidt, Xavier Niel, Daniela Rus, Vladlen Koltun

Team Structure

  • 60 employees across three locations
  • 40-45% in Europe (Paris, London)
  • 50-55% in the U.S. (California)
  • Currently hiring in all locations
  • Co-founders: Zhou Xian (CEO) and Théophile Gervet (President, former Mistral AI research scientist)

Technical Differentiators

Why Human-Scale Hands Matter

"If we could design a robotic hand that tries to mimic a human hand as much as possible, we can instantly unlock huge amounts of human data without having to worry about what people call the 'embodiment gap' in robotics research," Zhou explained.

Simulation as Competitive Advantage

The real bottleneck for model iteration is evaluation—Genesis' simulation system dramatically speeds up training cycles.

Concerns & Open Questions

  • Worker consent for data collection: Whether employees will agree to wear gloves that train their potential replacements
  • Data compensation: No details yet on extra pay for workers who collect training data
  • Data sharing: Customers may choose not to share collected data with Genesis
  • Genesis is also pursuing third-party data collection partners as alternative

Bottom Line

Genesis AI's full-stack approach—combining foundational AI models, human-scale robotic hardware, and innovative data collection methods—positions them uniquely in the robotics space. By solving the embodiment gap through hardware design rather than software adaptation alone, they're betting on faster iteration and better real-world performance. The upcoming full-body robot will test whether this integrated approach delivers on the promise of "the most capable robotic system."