While the majority of humanoid robotics developers are rushing to deliver an all-in-one package—combining the mechanical chassis, the proprietary AI control systems, and the operating software—China-based Booster Robotics is charting an entirely different course. The company manufactures the physical machinery, leaving the task of sourcing or building an intelligence to the buyer. This approach aligns strategically with the rapid advancement of freely available physical artificial intelligence foundation models from major players like NVIDIA, which are steadily growing more sophisticated and interchangeable.

“We produce the hardware,” Clarence Huang explained during a discussion at the Web Summit in Lisbon. “But we do not focus on the software.”

By stepping back from software dominance, Booster invites universities, researchers, independent developers, and tech enterprises to construct proprietary solutions directly on its mechanical platform. Users purchase the physical shell and integrate whatever cognitive architecture they prefer or develop themselves. It is a distinct strategy within a crowded global landscape that currently features at least 470 distinct humanoid robot models.

The Android Paradigm for Humanoid Robotics

Although Booster Robotics was established three years ago, its technological foundation spans roughly two decades of bipedal research originating from Tsinghua University’s premier robotics program.

Unlike competitors such as Unitree that manufacture both quadruped and humanoid machines, Huang noted that Booster concentrates exclusively on humanoids. The overarching vision is to mirror the role of Android within the smartphone industry—establishing an open hardware and software standard that empowers developers to craft diverse products without building a robotic system from the ground up.

“We provide a very open platform,” Huang stated. “We give full access to APIs and SDK.”

Application programming interfaces (APIs) allow operators to trigger machine actions, while software development kits (SDKs) guide engineers in constructing fresh capabilities. For Booster, this ecosystem is anchored by Booster Studio, a development hub introduced in June that merges coding, testing, simulation, and real-world deployment. The enterprise highlights that developers can utilize identical codebases across both virtual simulations and physical machinery, drastically smoothing the transition from inexpensive virtual training to physical execution. Furthermore, Booster Studio supplies AI-assisted code generation, real-time visualization of sensor telemetry, and agent-building utilities.

Naturally, Booster is not alone in providing open developer utilities. Unitree supplies SDKs alongside simulation environments and increasingly robust open-source AI frameworks. Furthermore, Booster is not entirely divorcing itself from software engineering; it designs proprietary operating systems and developer utilities, allocating segments of its financial funding toward embodied AI models and algorithms.

Nevertheless, the core commercial thesis remains supplying the physical shell for buyers eager to incorporate their own cognitive systems—a compelling proposition during an era marked by strict import regulations driven by anxieties over foreign electronic surveillance.

Introducing Booster T2: Packing 2,070 TFLOPS of Compute

During the Web Summit exhibition, Huang showcased the T2 humanoid, standing 1.4 meters tall and tailored specifically for academic research, developmental testing, and potential industrial deployment.

Weighing approximately 42 kilograms, the compact T2 features 31 degrees of freedom alongside a combined arm payload capacity of 10 kilograms. It accommodates various end effectors, including grippers and dexterous multi-fingered hands supplied by third parties. The Professional Edition incorporates NVIDIA’s Jetson Thor T5000 processor, generating up to 2,070 TFLOPS of FP4 sparse AI computing performance, according to Huang. This translates to roughly two quadrillion computations every second, superintended by software optimized explicitly for artificial intelligence workloads. Consequently, researchers gain substantial onboard computational power capable of handling perception, trajectory planning, and intricate neural networks.

Internal battery reserves sustain operations for roughly two hours under standard conditions, per official specifications. Remarkably, the unit can sustain continuous operation while drawing power through a magnetic tether, enabling researchers to station the robot at a fixed workbench throughout a standard eight-hour shift.

While this tethered arrangement does not yield a completely autonomous, round-the-clock worker, it affords crucial flexibility for engineering teams conducting prolonged trials.

Acquisition costs fluctuate significantly based on terminal hand selections. U.S. distributor RobotShop prices a baseline T2 Professional starting at $40,066 without complex hands, whereas a configuration featuring BrainCo Revo3 21-degree-of-freedom hands escalates to $132,660, underscoring the considerable expense of advanced robotic manipulators.

In addition to industrial variants, Booster markets a scaled-down, cost-effective model tailored for educational institutions, academic research laboratories, and competitive robotics leagues.

Cumulatively, the manufacturer reports shipping upwards of 3,000 robotic units across more than 30 global regions to support education, research, and development. Industry shipment estimates indicate that Booster stood among China's premier humanoid manufacturers during the first half of 2026, recording approximately 1,000 dispatched units.

Evaluating Market Demand for User-Supplied AI

Booster’s business model prompts a fundamental inquiry: how many buyers genuinely wish to configure their own robotic intelligence?

For academic institutions, scientific researchers, and physical AI developers, the answer is largely affirmative. They require adaptable hardware platforms suited for experimentation, customization, and deployment of proprietary models, making Booster’s open infrastructure an attractive proposition.

However, industrial end-users present an entirely different commercial prospect.

Operators of small-to-medium-sized warehouses or manufacturing facilities typically lack the desire to select foundation models, manually integrate them with mechanical frames, calibrate tasks, and sustain the resulting software infrastructure. These enterprises generally demand turnkey solutions that function instantly out of the box. This operational reality explains why innovators like Apptronik, Agility Robotics, Figure, and Tesla prioritize vertically integrated ecosystems that unite hardware, software, and intelligence into a cohesive product.

Safety considerations surrounding human-robot collaboration also remain paramount. When questioned regarding the operational safety of Booster platforms alongside human personnel, Huang offered a candid assessment.

“We cannot say it’s 100% safe,” he admitted.

System safety, he clarified, relies heavily on software maturity, task complexity, and the operational environment. Allowing buyers to integrate unverified AI layers inevitably injects another variable into an already complex engineering puzzle.

Of course, Booster rarely sells directly to traditional factories. The primary customer base comprises robotics startups, system integrators, and AI firms that acquire the baseline hardware, equip it with proprietary intelligence, and deliver fully packaged solutions to end clients. As foundational physical AI frameworks from NVIDIA and competing entities advance, assembling these turnkey packages is expected to become significantly less arduous.

Ultimately, the core debate facing the industry is whether humanoid robotics will mature along the lines of the mobile phone market—where shared hardware standards support a vast ecosystem of independent developers—or whether it will be governed by vertically consolidated entities producing tightly bound hardware and software suites.

Market dynamics suggest there is ample commercial space for both philosophies to thrive.