We’re on the brink of a robotics revolution. Imagine humanoid robots effortlessly performing complex tasks, from stocking warehouse shelves and assembling products, to assisting in hospitals or even helping around the house.

For years, the humanoid robots typified in sci-fi were constrained by expensive, cumbersome hardware and rigid software that required meticulous and manual programming. Today, China’s manufacturing boom has lowered costs, whilst the new software architectures and simulation technologies have expanded capabilities dramatically.

We are already seeing such robots rolling out in factories across the world, and at Moonfire, we are investing for an exponential rise in their use across industries, a $5 trillion economic opportunity by 2050.

Why Now? The Perfect Storm of Technological Progress

Today, sophisticated humanoid robots can cost as little as $16,000, as seen by the diminutive Unitree G1, produced in China, already stacking shelves, and competing in wrestling competitions. Whilst other models have higher costs - the Tesla Optimus is estimated to be at $50 to 60k per unit in hardware alone - these prices are within reach of broad industrial and commercial adoption.

Several key factors beyond cost have combined to push humanoid robots from the lab into the real world:

• AI Advancements: Deep learning and computer vision now enable robots to perceive, interpret and adapt to their environments autonomously. Robots no longer just follow scripts; they learn and adapt dynamically, like humans.
• Affordable, Powerful Hardware: Sensors, motors, and onboard computing have become more efficient and effective with greater motion and articulation.
• High Investment Levels: Venture capitalists and corporations are pouring resources into robotics startups, accelerating innovation and commercialisation. Over $7.4bn went to robotics startups last year, according to Crunchbase, and we are on track for a higher figure this year.
• Growing Labour Shortages and Demographic Pressures: Industries worldwide, from manufacturing to logistics, healthcare, and hospitality, are increasingly turning to automation to fill workforce gaps. This pressure will only increase as populations age, leaving more gaps in the workforce and requiring care themselves. Unlike specialised robots, humanoids can slot straight into tasks rather than requiring workplace reconfigurations.

Hardware Sets the Stage, But Software Makes the Difference

The physical capabilities of humanoid robots have dramatically improved, thanks to breakthroughs in actuators, energy-dense batteries, and advanced sensors. Companies like Tesla, Boston Dynamics, AgiBot, UBTECH and Figure AI have introduced robots with lifelike agility and dexterity. Yet, hardware alone isn’t sufficient. Without intelligent, adaptive software, even the most advanced robots remain limited.

Software is the ultimate bottleneck and the greatest opportunity for humanoid robotics. The shift from manual programming and teleoperation towards self-learning robots has become essential.

Modern robotics software combines several critical elements:

• Operating Systems and Middleware: Standardized frameworks like ROS 2 could simplify integration, allowing rapid development of robot capabilities, but to date it is mainly used by academia as firms Tesla, Figure and Boston Dynamics have avoided it.
• Perception and Sensing: Robots now use advanced sensors (cameras, LiDAR, IMUs) coupled with AI-driven vision models to interpret their surroundings in real-time.
• Control and Motion Planning: Dynamic control systems, such as reinforcement learning, enable robots to move fluidly and respond instantly to their environment, handling complex tasks like navigating cluttered spaces or adjusting balance mid-motion.
• AI-Driven Decision Making: AI techniques, especially reinforcement and imitation learning, allow robots to learn from experience or human demonstrations, adapting autonomously to changing scenarios.
• Simulation and Digital Twins: High-fidelity simulation and digital twins let robots skip slow, risky real-world trial-and-error. Tools like NVIDIA Isaac Sim, provide rapid, safe, and scalable training, boosting real-world performance whilst cutting costs and time.

In short, robots have evolved from pre-programmed, rigid machines into adaptive learners, capable of understanding context, reacting in real-time, and continuously improving their performance.

Moonfire in Robotics

Moonfire is also taking part in this robotics revolution. Using our “bionic VC” approach, we sifted through hundreds of promising startups across the stack, investing in Flexion Robotics, a small Zurich-based team. Flexion’s approach moves past the brittle, deterministic models of yesteryear, with a Vision-Language Model (VLM) driving new levels of agility, robustness, and adaptability.

Flexion’s founder and CEO, Nikita Rudin, said: “Large language models have unexpectedly accelerated the ‘top of the stack’: task planning, language understanding, and other high-level reasoning that general-purpose robots need. What remains stubbornly difficult is the bottom layer: reliable, agile and real-time whole-body coordination. We address this bottleneck by training robots in simulation with reinforcement learning, then transferring those policies to hardware. Coupling such controllers with the reasoning capabilities of LLMs brings truly general-purpose autonomy within reach."

The Future: Embodied General Intelligence & Human Freedom

Perhaps the most exciting trend is this full convergence of humanoid robotics with large-scale AI multi-modal foundation models. These giant models provide general knowledge and reasoning abilities, allowing robots to understand and execute natural language commands, generalise across tasks, and learn new skills rapidly. Projects like Google Deepmind’s Gemini Robotics are pioneering a future where robots come preloaded with human-like cognitive abilities, able to interpret complex instructions and adapt seamlessly to new tasks.

Fleet learning, where robots connected via cloud networks continuously share experiences, will further drive collective improvement. Imagine a robot encountering a novel situation in a warehouse; its experience instantly contributes to improving the entire fleet’s capabilities. Companies like Tesla and Figure AI already envision using this networked learning approach to enable robots to collectively grow smarter, faster.

As humanoid robots become increasingly autonomous, adaptive, and capable of complex decision-making, their potential applications multiply. As costs continue to fall, in manufacturing plants, they’ll handle intricate assembly; in logistics, they'll sort packages swiftly; in healthcare, they'll assist surgeries with precision. Even at home, they could become our indispensable assistants.

The next decade promises exciting developments. Thanks to advances in AI, simulation, cloud computing, and collaborative learning, humanoid robots will no longer be rigid automatons but dynamic, intelligent partners in our daily lives. The robotics revolution isn’t just about machines; it’s about empowering human potential, making life safer, more productive, and enriching for everyone.

Learn More at Moonfire Pulse

How should investors think about humanoid robotics? What industries will be most affected? Is China winning the robot race? Where will value accrue? What new business models will emerge? These are a few of the questions we will be asking at our upcoming panel, The Rise of Adaptable Robots, at Moonfire Pulse this June 10.

Once a year, Moonfire gathers 200+ of the most influential operators, investors and founders in tech and venture for Moonfire Pulse - an invitation-only afternoon of discovery and connection. We tackle the big topics of the day and always from an unexpected angle.

We’ll hear from the founders like Nikita, practitioners and academics making this sci-fi vision a reality.