Welcome, Startup enthusiasts!

Good Morning. It’s July 24, 2026.

🚀 The Startup Khan Newsletter

July 25, 2026 — Weekend Robotics Edition: Physical AI Moves From Demos to Deployment

🧠 Top Story: Humanoid Raises $152M at a $1.35B Valuation

UK robotics startup Humanoid has raised $152 million in Series A financing, reaching a $1.35 billion post-money valuation.

The round was led by Prime Movers Lab, with participation from industrial heavyweights Schaeffler and Bosch, along with Fubon Financial Holding Venture Capital and Aglaé Ventures. The financing brings Humanoid’s total capital raised to approximately $270 million. (Reuters)

Humanoid is building modular robots for industrial environments, including both wheeled and bipedal platforms. Its wheeled system is designed to perform tasks such as moving materials, handling components and operating around existing factory infrastructure.

The company plans to use the new funding to:

• Develop its next-generation robotics platform
• Expand its proprietary KinetIQ AI software
• Begin commercial deployments
• Establish mass manufacturing for its wheeled robots

Beta robots are expected to begin entering customer sites during the fourth quarter of 2026 across manufacturing, logistics, retail and other industrial environments. (Ministry of Economy, Trade and Industry)

Bosch will support contract manufacturing and hardware engineering, while Schaeffler has signed an agreement involving the planned deployment of thousands of robots across its manufacturing network. (Ministry of Economy, Trade and Industry)

Why founders should care

Robotics companies are no longer being valued only for impressive demonstrations.

Investors increasingly want evidence of a complete commercialization system:

• A repeatable use case
• An industrial customer
• A manufacturing partner
• Deployment software
• A path to operating thousands of machines

Humanoid’s strongest advantage may not be whether its robot walks on two legs. It may be that industrial partners are helping it manufacture, test and eventually purchase the machines.

👉 Founder takeaway: In physical AI, distribution and manufacturing partnerships can be as valuable as the underlying model.

---

💰 Other Robotics Moves

1️⃣ Japan builds a national physical-AI platform

Japan-backed company Noetra is developing a domestic multimodal foundation model intended to power robots and other physical-AI systems.

The initiative brings together companies including Sony, SoftBank, NEC and Honda, alongside a broader group of Japanese manufacturers, technology providers and investors. (Sony)

Noetra plans to construct an AI computing platform containing approximately 27,500 Nvidia Rubin GPUs. Construction is scheduled to begin in April 2027, with operations targeted for June 2028. (Noetra)

The wider initiative is designed to combine Japan’s industrial data and manufacturing expertise with locally controlled AI infrastructure.

FANUC announced on July 24 that it was investing in Noetra to support development of the model and strengthen Japan’s industrial competitiveness and economic security. (FANUC Corporation)

This is larger than another robotics startup.

Japan is attempting to build a national physical-AI stack spanning:

• Compute infrastructure
• Foundation models
• Industrial data
• Robot manufacturers
• Factory deployment

👉 Founder takeaway: The physical-AI race may be fought by national ecosystems, not isolated startups.

2️⃣ Foundation partners with AMD on Phantom MK-2

Robotics company Foundation Future Industries has partnered with AMD to develop the next version of its humanoid robot, the Phantom MK-2.

Foundation plans to use AMD’s Ryzen AI Embedded X100 Series processors to support onboard AI processing for autonomous industrial applications. The financial terms of the partnership were not disclosed. (Reuters)

Foundation told Reuters that it plans to open a factory in October with annual production capacity for approximately 5,000 robots. A second proposed facility would eventually have capacity to produce as many as 50,000 robots per year. (Reuters)

The important shift is edge intelligence.

Robots cannot depend entirely on distant cloud systems. Many industrial decisions require low-latency perception, control and inference directly on the machine.

👉 Founder takeaway: The robotics equivalent of cloud versus edge computing is becoming a major product-design decision.

3️⃣ BRINC raises $125M for emergency-response drones

Seattle-based BRINC raised $125 million in financing led by Motorola Solutions, with participation from Index Ventures and Figma founder Dylan Field.

The company builds drones and related systems for police, fire and public-safety organizations. BRINC says the funding will support expanded U.S. manufacturing and the development of new products. (PR Newswire)

Its long-term vision is to place emergency-response drones near police and fire stations so aircraft can be dispatched quickly when a call arrives.

The company represents a category where robots are not necessarily replacing workers. They are entering hazardous environments before human responders.

👉 Founder takeaway: The clearest robotics opportunities often begin with tasks that are dangerous, slow or expensive for people to perform.

---

🧰 Tool of the Day: LeRobot 0.6

LeRobot is Hugging Face’s open-source platform for building, training and evaluating robotics models using PyTorch.

Version 0.6, released in July, expands the platform’s model-evaluation and robot-learning capabilities. Hugging Face also added access to Nvidia Isaac Teleop and the Isaac GR00T 1.7 vision-language-action model through the LeRobot ecosystem. (Hugging Face)

LeRobot provides access to:

• Robotics datasets
• Pretrained policies and models
• Training pipelines
• Evaluation tools
• Hardware integrations
• Teleoperation workflows
• Community models through the Hugging Face Hub

This allows developers to begin experimenting with robot learning without creating an entire machine-learning stack from the ground up. (Hugging Face)

Useful founder applications include:

• Training robotic arms through demonstration
• Testing vision-language-action models
• Building warehouse automation prototypes
• Creating reusable robotics datasets
• Comparing robot policies before hardware deployment

👉 Best use: Start in simulation or with low-cost hardware, validate one repeatable task and collect structured failure data before attempting a general-purpose robot.

---

📊 Trend Check: Robotics Is Becoming a Full-Stack Business

The current robotics race is not simply a competition to build the most human-looking machine.

The emerging stack includes:

1. Compute

Chips process perception, planning and control.

1. Foundation models

Multimodal systems interpret language, video, spatial information and sensor data.

1. Training data

Robots require demonstrations, simulations and real-world interaction records.

1. Hardware

Actuators, hands, sensors, batteries and mobility systems turn decisions into movement.

1. Fleet software

Companies must monitor, update, coordinate and troubleshoot deployed robots.

1. Manufacturing

A prototype becomes a business only when it can be produced reliably and economically.

1. Deployment partners

Customers provide the real environments, workflows and feedback required to improve performance.

Humanoid’s Bosch and Schaeffler relationships, Noetra’s national industrial consortium and Foundation’s AMD partnership all point to the same conclusion:

Robotics winners will control connected layers of the stack—not merely a single robot design.

---

🌟 Founder Spotlight: Artem Sokolov and the Humanoid Team

Artem Sokolov founded Humanoid with the objective of moving humanoid robotics beyond laboratories and staged demonstrations into commercially useful industrial deployments.

The company’s leadership combines expertise across robotics hardware, artificial intelligence, product development, operations and industrial commercialization. (Humanoid)

Its strategy contains an important product lesson.

Humanoid is developing a bipedal platform, but it is also prioritizing a wheeled robot that can operate in structured factories without solving every challenge associated with human-like walking.

This reduces the amount of energy and computation dedicated to balance and locomotion, allowing more resources to be directed toward manipulation, perception and task execution.

The playbook:

👉 Do not solve the most visually impressive engineering problem first. Solve the version customers can deploy reliably.

---

⚡ Quick Win: Build a Robotics Deployment Scorecard

Before building or investing in a physical-AI product, score the proposed use case from one to five across these categories:

Task frequency

How often is the task performed?

Environment consistency

Does the workspace remain relatively predictable?

Economic value

How much labour, downtime or waste could the system reduce?

Failure tolerance

What happens when the robot makes a mistake?

Training-data availability

Can you collect enough demonstrations and edge cases?

Human supervision

Can a person intervene remotely when necessary?

Hardware complexity

Does the use case require legs and advanced hands, or would wheels and a basic gripper work?

A repetitive factory task with clear boundaries may be a stronger startup opportunity than an ambitious general-purpose household robot.

Optimize for deployment readiness—not demonstration value.

---

🤖 Weekend Build Challenge: Create a Robot Before Buying One

You do not need a humanoid robot to begin learning physical AI.

This weekend, build a simulated robot workflow:

1. Install a robotics simulation environment such as Gazebo or an accessible LeRobot-compatible setup.
2. Select one task, such as navigating to a marked location or moving an object.
3. Define the success condition.
4. Run the task repeatedly.
5. Record every failure.
6. Categorize failures into perception, planning, control and environment problems.
7. Improve the weakest stage.

The goal is not to produce a spectacular video.

The goal is to understand why reliable robotics requires far more than connecting a language model to motors.

---

🧩 Final Thought

This week’s developments show physical AI becoming an industrial race.

Humanoid is connecting robotics with manufacturing partners.

Foundation is bringing more AI processing directly onto the machine.

BRINC is deploying autonomous systems into high-value public-safety workflows.

Noetra is organizing an entire national ecosystem around robotics models, infrastructure and industrial data.

LeRobot is making development tools and models more accessible to independent builders.

The next robotics giants may not build a machine that can do everything.

They may begin by building one that performs a valuable task reliably, safely and thousands of times per day.

The robot demonstration attracts attention.

The deployment system builds the company.

— The Startup Khan

Your 5-minute startup edge — funding news, robotics tools and founder playbooks.