"Lobster" Rewrites the Hundred-Billion Hardware Arena

Lobster has broken the closed - loop camp of hardware companies and brought them the possibility of industrial chain reconstruction.

How to enter the world of Lobster (OpenClaw) is becoming a must - answer question for hardware manufacturers.

Recently, mobile phone manufacturers such as Xiaomi, Huawei, OPPO, and Honor have collectively launched system - level "Mobile Phone Lobster", regarding it as the killer weapon to subvert the future experience.

On March 6th, Xiaomi launched a small - scale closed - beta test of Xiaomi miclaw, which is based on Xiaomi's self - developed MiMo large - model and can call more than 50 system tools and Xiaomi IoT ecological devices.

"By integrating the entire ecosystem of vehicles, homes, and people, miclaw can achieve operations that were previously unimaginable." On March 19th, after the Spring New Product Launch Conference, Lei Jun told media such as China Entrepreneur, "There is no doubt that this era requires us to actively embrace AI. The mobile phone lobster can be regarded as the prototype of Xiaomi's AIOS with the built - in MiMo (Xiaomi's large - model)."

Recently, Wang Teng, the former person - in - charge of Redmi at Xiaomi, also said that it is difficult for mobile phone hardware to have major innovations at this stage, and the efficiency of the traditional iteration model is relatively low. He suggested that mobile phone manufacturers should launch fewer new phones and instead focus more on developing the mobile phone version of Lobster.

Beyond mobile phones, from AI glasses manufacturers such as Rokid, Thunderbird, and Liweike, to AI earphones, smart wearables, and even robot companies, hardware manufacturers are flocking to "raise lobsters". They are deeply involved in the homogeneous competition of competing on parameters, performance, and price. Many manufacturers are worried that if they miss the opportunity to enter the Lobster field, they may miss the turning point in the hardware industry.

However, while "Lobster" brings active AI capabilities, it also brings new problems such as security and Token consumption. How to make "Lobster hardware" bring new interaction methods will reconstruct the hardware landscape.

Flocking to "Raise Lobsters"

Rokid is the first AI glasses company in the industry to announce its embrace of Lobster.

Zhu Mingming, the founder of Rokid, told China Entrepreneur that currently, the company has dozens of Lobster open - source projects, and the access work is spontaneously completed by developers. The applications cover scenarios such as stock trading reminders, paper summarization, and code debugging. "We will make OpenClaw the default standard configuration in the near future, and users can activate it with one click on their glasses or mobile phones."

Zhu Mingming believes that accessing Lobster is expected to transform AI glasses from light - task "tools" to heavy - task "partners".

"Traditional smart glasses can perform light tasks such as translation and cueing, but they are inadequate when it comes to data analysis and code writing. Lobster can hand over heavy tasks to the cloud or computer, and the hardware only needs to be responsible for interaction, adjustment, and notification. This division - of - labor model will also return the core value of hardware to human - machine interaction itself, rather than computing power competition."

Thunderbird has taken the route of combining self - research with the ecosystem. As early as the second half of 2025, Thunderbird established a large - model team of dozens of people to develop an active AI module similar to Lobster.

"Our developer platform has accessed the Lobster interface, and we are also polishing our own active AI applications." A relevant person in charge of Thunderbird told China Entrepreneur. Thunderbird's Lobster deployment uses a direct cloud - connection model without PC transfer, and it is currently exploring the implementation of scenarios such as smart home control.

The active AI capabilities of Lobster coincide with Thunderbird's R & D direction over the years. "The biggest feature of Lobster is that it can take over devices and actively execute tasks. For example, it allows users to order coffee or takeaways directly through glasses or control smart home appliances. Without multiple voice commands, a single trigger can complete a closed - loop operation."

Guangfan Technology, an AI earphone company, recently completed a financing of 300 million yuan. Its launched AI earphones and watches have also accessed Lobster.

"In this era, people are gradually realizing that the model is the product, but perhaps more fundamentally, the scaling law is the product. There are still great opportunities in the integration and organization of tool - level connections and the migration from Internet narratives to large - model closed - loops. The software ecosystem will have full freedom compared with the previous generation of experience." Guangfan Technology told China Entrepreneur.

However, Guangfan Technology also said that the core challenge of Lobster deployment lies in Token management and experience. "If users do not make local deployments themselves, the experience will not be as smooth."

Although there are many challenges, no one dares to miss this wave of opportunities. Hardware companies are rushing to make arrangements at a "head - start" speed.

According to the report of "New Cortex" by Yicai: miclaw was developed by a small experimental team in the software department of Xiaomi's mobile phone team. Initially, there were less than 5 engineers in the team. Xiaomi's intelligent assistant "Super Xiaoai" has already tried to cooperate with miclaw, and there is a possibility of merger in the future.

On March 11th, He Gang, the CEO of Huawei's Consumer BG, also showed the interface of Huawei's Lobster "Xiaoyi Claw Beta" at the press conference. Different from Xiaomi, Huawei's solution focuses more on cross - device collaboration. "Xiaoyi Claw" can switch across devices such as mobile phones, tablets, and smart screens in the HarmonyOS ecosystem.

A person familiar with the matter told China Entrepreneur: "Xiaoyi Claw can be regarded as the intelligent ability that has grown in Huawei's HarmonyOS and ecosystem. Huawei has brought the Agent competition back to the system and security, and has the support of the HarmonyOS capabilities." The person also revealed that Huawei will announce more news about Xiaoyi in March.

Chen Xi, the person - in - charge of OPPO ColorOS, recently also exposed "Xiaobu Claw" on social media. It can read the content of call recording files, actively prompt subsequent operations, and perform cross - device operations.

Honor also announced an ecosystem - level "Lobster Universe" at the Magic V6 launch conference. It allows users to "raise" and "control" lobsters with one click on Honor devices through the YOYO assistant and is compatible with the access of Lobster from other ecological devices. This function will be open for internal testing in the near future.

Open - Source Equal Rights

Behind the collective "lobster - raising" of hardware manufacturers is the industry's redefinition of intelligence.

Chen Yuetian, the founder of Huofeng Capital, said that the core attraction of Lobster lies in "open - source", which is the key logic for hardware manufacturers to compete for access. "Developers around the world are constantly contributing code to OpenClaw, almost with a new version every day - Manus can't iterate that fast." Chen Yuetian told China Entrepreneur.

Lobster has built a low - cost, high - iteration, and barrier - free intelligent upgrade platform for hardware manufacturers, turning the active AI capabilities that originally required heavy investment into an industry dividend accessible to everyone. Hardware manufacturers do not need to bear the R & D, debugging, and iteration costs of active AI functions alone, which also brings them the possibility of industrial chain reconstruction.

For example, in the past, the robot industry had to independently develop everything from the brain (decision - making system), cerebellum (motion control) to large and small models. This not only had high R & D costs and long iteration cycles but also had compatibility problems caused by the technical closed - loop.

After accessing OpenClaw, robot manufacturers can focus more on core capabilities such as the hardware body and motion control, and hand over the "intelligent decision - making" part to the mature AI intelligent agent framework. This division - of - labor and cooperation model allows robots to obtain AI capabilities more quickly and reduces the R & D threshold and costs.

The more important significance of Lobster's open - source lies in breaking the closed - loop camp of AI technology. Previously, the AI systems of large companies were naturally "exclusive", often requiring hardware manufacturers to exclusively bind models and ecosystems. Lobster gives hardware manufacturers the right to pursue "ecosystem freedom".

For example, Rokid allows users to freely choose different base models such as Doubao, Qwen, DeepSeek, GPT, and Claude; Thunderbird connects its self - developed large model and third - party intelligent agents on the basis of OpenClaw. For them, accessing the open - source Lobster no longer requires attachment to a single AI platform but can join a more open technology ecosystem.

At the same time, traditional AI solutions are mostly limited to cloud - based interactions and are disconnected from the local operation system of hardware, resulting in a fragmented intelligent experience. Lobster can be combined with the local file system in a cross - way, allowing the core value of hardware to be more fully released.

Chen Yuetian compares OpenClaw to "early Linux". Its open - source ecosystem replicates the successful paths of Linux and Android - by opening up technology to gather industrial forces, form a technical foundation, and then drive the large - scale development of the entire industry. He predicts that a more mature ecological system will be born based on Lobster in the future, and manufacturers who test the waters early will pre - occupy the technical standards for the next - generation AI hardware.

Multiple Tests

Although the ecological prospects of Lobster are generally optimistic, hardware manufacturers also face multiple challenges such as security, cost, and experience during the access process.

Among them, security and privacy issues are undoubtedly the biggest concerns of users. Zhu Mingming said bluntly: "I highly recommend that ordinary users do not casually open Lobster permissions, especially in terms of privacy and finance." He gave an example that some users' Lobsters even complained that the computer's response speed was not fast enough and suggested that users buy a new GPU to "upgrade".

Chen Yuetian also expressed concerns about the potential technical risks of Lobster: "Lobster can be 'persuaded'. All along, there has been a possibility of 'jailbreaking' in large language models. By using extremely long instructions to squeeze out the original context, it can break through permission restrictions, which is a problem that the industry has not been able to fully solve at present."

"The memory system of OpenClaw uses Markdown documents, and each system prompt is regenerated. The construction of context is a very long process. As users use it, Markdown files will accumulate more and more and become larger and larger, and there will definitely be a problem of memory forgetting."

To deal with security risks, hardware manufacturers have taken protective measures. Thunderbird uses dual verification of "voiceprint + hardware" in payment scenarios to ensure that Lobster cannot operate payments without permission; Rokid restricts the permissions of sensitive words and clearly prohibits the system from touching functions related to finance and privacy; Xiaomi and Huawei's mobile phone versions of Lobster are deployed in a security sandbox and isolated from users' core data.

In addition, the cost problem caused by Token consumption cannot be ignored. Some users reported that "they used up all the Tokens overnight and had to recharge the next day to continue using". Zhu Mingming said that his friend tried to use Lobster for stock trading, but the profit was not enough to cover the Token fees for the small account.

For manufacturers still in the exploration stage, the current cost - sharing model for Tokens is not clear. Thunderbird provides free Token support for developers, but whether the product end will charge in the future is still under evaluation. Rokid chooses to believe in the power of technological evolution - "The cost of Tokens will definitely gradually decrease, and then OpenClaw will also gradually optimize the Token consumption speed."

The consistency of experience and the difficulty of adaptation also test the manufacturers. The person - in - charge of Guangfan Technology pointed out that the experience of non - local deployed Lobster is not smooth enough, while local deployment requires a relatively high threshold for user operations. Zhu Mingming also found that there are significant differences in the experience of different base models. "The effect will be better if you choose GPT or Claude, but the experience is very poor if you choose models like Llama." Therefore, he plans to ensure consistent user experience by officially recommending base models.

Hardware manufacturers generally believe that adhering to the essence of hardware is the core of being a good "carrier in the AI era".

Zhu Mingming emphasized: "Rokid positions itself as a human - machine interaction company, not an AI company. We don't develop models, but only focus on being a good carrier." In the future, the competition in hardware will focus on the naturalness of interaction, portability, and scenario adaptation ability, rather than the stacking of computing power.

Chen Yuetian compares OpenClaw to the "printing press" and believes that it is breaking the technological monopoly and popularizing intelligent capabilities. The essence of this "lobster - raising" movement will be the transformation from "selling hardware" to "providing intelligent services".

When smart hardware can be "de - centralized" and become more active, robots, desktop devices, and wearable devices may all become execution nodes of Agents. When cloud - based AI can directly understand voice, visual, or environmental information through various "distributed hardware", mobile phones may also retreat from being the "control center" to one of many terminals.

This article is from the WeChat public account "China Entrepreneur Magazine" (ID: iceo - com - cn), written by Lin Qiuyi and Wang Yijie, edited by Li Yuan and He Yifan, and published by 36Kr with authorization.