From Serving Tables to Physical AI: Bear Robotics…

A robot company known for carrying plates may be preparing for a much bigger financial and technological move. Bear Robotics, the Silicon Valley company controlled by LG Electronics, is reportedly seeking between 300 billion and 400 billion won in pre-IPO funding as it explores a future Nasdaq listing. That is roughly the kind of capital that can turn a successful service-robot supplier into something more ambitious: a Physical AI platform spanning hospitality, logistics, manufacturing and mobile manipulation.

The financing has not been formally announced by Bear Robotics or LG. Seoul Economic Daily and other Korean outlets attribute the figures to investment-banking sources, naming Bank of America as the arranger and describing a potential valuation of around 2 trillion won. Those details should therefore be treated as reported plans, not a completed transaction. Even so, the reports arrive at a revealing moment in Bear’s evolution.

The proposed funding is only the surface story

Bear built its reputation with Servi, a wheeled robot designed to move food and dishes through busy restaurants. Its machines later expanded into hotels, hospitals, retail spaces and other commercial buildings. According to the company, more than 16,000 robots have been shipped into commercial service across North America, Europe and Asia.

That installed base matters because real robotics businesses are not built on a single impressive demonstration. They are built on navigation reliability, charging routines, fleet monitoring, maintenance, customer support and the ability to operate around unpredictable people. A company that has already learned how to keep thousands of mobile machines working in public environments owns something that many humanoid startups still lack: operational experience.

But transport is only one half of physical work. A robot that can carry a tray or container still cannot necessarily pick up an unfamiliar object, sort mixed items or place a component accurately. Bear’s strategic question is whether its mobile fleet can become the foundation for machines that also manipulate the world.

Kinisi supplies the missing hands

Bear’s June agreement to acquire the British startup Kinisi Robotics makes the new direction explicit. Kinisi developed the KR1, a wheeled humanoid with arms designed for picking, placing, sorting and moving objects in industrial, logistics and hospitality environments. The acquisition also adds vision-language-action models, a robot foundation model, gripper technology and tools for collecting human demonstrations.

The wheeled design is significant. Legs create spectacular videos, but wheels remain more efficient and stable on the smooth floors found in restaurants, hospitals, warehouses and factories. A torso and two arms can provide access to human workspaces without forcing the machine to solve dynamic walking at the same time. Bear is effectively betting that a practical humanoid does not need to imitate every part of the human body.

Kinisi was also not an unrelated technology purchase. Bear says the KR1 was built on its navigation platform, while Kinisi founder Brennand Pierce previously co-founded Bear. That history could reduce one of the most common acquisition risks: trying to connect incompatible software, hardware and engineering cultures after a deal closes.

The data flywheel is the strongest investment thesis

Physical AI models need examples of what happens when a machine acts in the real world. Camera images alone are not enough. Training records must connect perception, instructions, robot states, actions, successes, failures and human interventions. Gathering that information is slow and expensive because robots wear out, environments change and mistakes can damage objects or equipment.

Bear argues that its deployed fleet supplies continuous operational data, while Kinisi contributes manipulation demonstrations and learning systems. In theory, the combination creates a flywheel: more deployed robots generate more useful experience; better models improve robot performance; improved performance supports additional deployments.

That is an attractive story, but the quality of the loop matters more than the number of robots shipped. Navigation data from a restaurant does not automatically teach a humanoid hand to insert a component. The company would need to show that its data is sufficiently detailed, legally usable, consistently labeled and transferable between different robot bodies. Investors should distinguish between a large fleet and a large quantity of high-value training data.

What a billion-dollar valuation would need to prove

The reported valuation represents a sharp increase from Bear’s position when LG invested $60 million in 2024. A higher price can be justified if the company has recurring revenue, growing utilization and a credible route from service robots into higher-value industrial tasks. It becomes harder to defend if the new category depends mainly on pilots and future promises.

Useful evidence would include the number of actively operating robots rather than cumulative shipments, average hours worked, customer renewal rates, intervention frequency, service costs and gross margin after maintenance. For manipulation systems, the key measurements change: task completion without human rescue, time needed to learn a new job, performance across varying objects and the cost of each successful cycle.

The financial reports also point to losses. That is not unusual for a robotics company investing in hardware, software and international support, but it makes unit economics essential. Rapid growth can destroy value if every deployment requires expensive on-site engineering. A scalable platform must make the next customer easier to serve than the previous one.

LG brings industrial reach—and a governance question

LG acquired control of Bear Robotics in 2025 as part of a broader shift toward software-defined commercial robots. The relationship gives Bear access to manufacturing expertise, components, global sales channels and enterprise customers. It could also help the company bridge the difficult gap between a laboratory prototype and a supportable product.

A separate Nasdaq listing, however, may raise questions about how value is divided between the listed subsidiary and its corporate parent. Korean reporting has already flagged the possibility of concerns around a duplicate listing. Prospective investors would need clarity on intellectual-property ownership, commercial agreements, procurement relationships and the degree of operational independence Bear would retain.

For European robotics, the Kinisi acquisition adds another dimension. The combined group gains an engineering base in Bristol and a platform aimed at warehouses and manufacturers that are struggling with labor shortages. Europe could become both a customer market and a source of manipulation research, although deployments will still have to meet local machinery-safety, workplace and data-protection requirements.

The Alpha Bionic view

The most important Bear Robotics story is not whether a serving robot company can attract a fashionable Physical AI valuation. It is whether commercial fleet experience can shorten the path from a promising manipulation demo to dependable daily work.

Bear starts with advantages: deployed machines, fleet software, customer relationships and LG’s industrial backing. Kinisi adds arms, learning models and a pragmatic wheeled humanoid. The missing evidence is operational. Can the combined platform learn new tasks quickly, complete them reliably and generate attractive economics after support and maintenance?

If Bear answers those questions with real deployment data, a pre-IPO round could fund a credible expansion rather than a change of narrative. Until then, the financing remains reported, the valuation remains prospective and the transformation from restaurant logistics to general Physical AI remains a thesis under construction.