This is a Founders’ Note rather than a product announcement. For us, choosing where to enter the field of robotics and embodied intelligence is a decision that must be deliberate from day one.

At HanabiAI, we chose to begin with the chip and the compute platform, rather than starting by building a visible, full robot.
Below, we share the reasoning behind this choice and, along the way, how we think about long-term, generational iteration.

1. Why We Did Not Start by Building a Full Robot

For many people, “building a robot” naturally brings to mind something with a body a shell, wheels or legs, movement, speech.
But from an engineering standpoint, a full robot implies:

• mechanical structure, actuation, sensors, industrial design, and safety certification; supply chain, manufacturing, and after-sales support;

• heavy customization for a single scenario, where much of the investment cannot be reused for the next product;

• and the constant risk of being pulled toward short-term demos, spending most effort on the “visible” parts.

If we had put all our energy into the hardware form factor from day one, we would likely find ourselves consumed by mechanical details, scenario-specific edge cases, and supply-chain complexities leaving too little bandwidth for what we believe is the true core: the robot’s brain.

In our view, the robot brain has far greater reusability and scalability:

• Robots in different environments can share the same foundational intelligence stack.

• Chips, compute platforms, and models can accumulate over time and power multiple generations of robot forms.

• When collaborating with robot manufacturers or research institutions, we can be a long-term partner providing the core intelligence, rather than a direct competitor on the full robot.

This is why we chose not to start with a finished robot. We chose instead to make the brain clear, solid, and future-proof before anything else.

2. What We Mean by the “Robot Brain”

When we talk about the robot brain, we don’t mean a single chip, nor a standalone software service. What we refer to is an interconnected technology stack that works together:

• Chip Layer Embodied AI SoC A dedicated SoC designed specifically for embodied-intelligence robots. We plan and develop it in-house, focusing on system-level concerns such as compute distribution, energy constraints, real-time behavior, and safety boundaries.

• Platform Layer Embodied AI Compute Platform. Running on the cluster/data-center side, it powers simulation, training, evaluation, and model engineering. On top of general HPC&AI capabilities, it is purpose-built to support embodied-intelligence workloads.

• Models and Toolchains. Policy, perception, and control models for embodied scenarios, along with the export, adaptation, and evaluation tools surrounding them.

• Modules and Future Forms. As the SoC matures, the brain can be packaged into modules that are easier to integrate providing different robot manufacturers with a unified “brain interface.”

A simple way to think about it is this: the Embodied AI Compute Platform “prepares and trains” the brain, the SoC/module “runs the brain” on the robot, and the robot body carries these capabilities into the real world.

3. Why We Start from the Chip and the Platform

Better Reusability Across Scenarios

Full robots are often deeply tied to specific environments shopping malls, hospitals, factories, campuses each requiring heavy customization.
In contrast, when designed well, chips, platforms, and model toolchains can be reused across both robots and scenarios:

• The same Embodied AI SoC can adapt to multiple robot form factors.

• The same Embodied AI Compute Platform workflow can support different embodied-intelligence tasks.

• When new robot forms emerge, as long as they follow certain interface standards, they can quickly plug into the same brain.

Complementing Ecosystem Partners

What we want is to:

• Work with robot manufacturers, system integrators, and research institutions as long-term partners;

• Validate the value of embodied-intelligence robots across diverse real-world scenarios—together;

• Instead of competing directly with them on complete robot products.

By starting from the chip and the platform, we can take on a role in the ecosystem that is both clearer and more sustainable:
focus on building the best possible brain, so the robot’s body can better perceive the physical world, interact with it more effectively, and ultimately create more value.

4. How We Think About Long-Term Iteration

For HanabiAI, embodied intelligence is not something that can be achieved within a single product cycle. This is not just a question of technical difficulty it is fundamentally a question of time scale.

We are very clear about one thing: a truly valuable robot brain is not a one- or two-year project. It is a roadmap that must be planned on a five- to ten-year horizon. Every generation of chips, every reconstruction of the platform, every iteration of models and toolchains each one serves as groundwork for the next, rather than an attempt to satisfy short-term trends.

This is also why we chose to begin with the chip and the platform. They determine how far embodied-intelligence robots can go in the long run, not “whether the next demo looks impressive.” Along this path, we would rather spend more time refining architecture and foundational capabilities than constantly shifting direction for short-term visibility or resetting progress again and again.

For us, long-term investment is not a slogan it is a form of discipline:

• We do not present exploratory directions as if they were mature, fully-developed products.

• We will not abandon the core mission of building the robot’s brain just because of short-term shifts in market sentiment.

• Within the limits of our capabilities and resources, we will continue to invest time and effort, building this work generation by generation.

We treat embodied intelligence and the robot brain as something worth spending years not months to build. This is not a “trend to ride” or a “story to tell.” It is both our vision, and a commitment we make to ourselves.

5. What This Means for Potential Partners

If your current interests include:

• research topics or pilot projects related to embodied intelligence;

• mid- to long-term planning for robot roles in campuses, industrial parks, or public spaces;

• the future possibility of using dedicated SoC or modules in robotic systems;

then we hope you can understand HanabiAI in the following way:

• We are not yet providing plug-and-play full robots.

• We are better positioned to be your long-term partner for the robot’s brain and compute infrastructure.

• What is immediately ready for real deployment is the Embodied AI Compute Platform’s capabilities in general HPC&AI scenarios.

Capabilities specific to embodied-intelligence workloads will continue to be validated and refined through pilot collaborations and co-creation projects.

If any of these ideas intersect with what you are thinking about right now, we would be glad to talk when the time is right.