In just 20 days since its launch, Muse has garnered approximately 4 million total users and between 600,000 and 800,000 DAU (Daily Active Users). Keep in mind, it is currently only available for download in select markets, such as the United States and Canada. During this period, Meta's stock price has surged by around 20%, transforming it from an 'AI industry laggard' to a 'new leader in the AI sector'.

Recall two months ago, during Meta's Q2 earnings call, analysts repeatedly questioned Zuckerberg, 'Why is Meta still investing heavily in developing the Muse Spark model? Why not simply rely on third-party or open-source models?' Now, Zuckerberg can proudly declare to Wall Street, 'This is precisely why!'

Muse is a robust, user-friendly AI Agent designed for the general consumer market. Many are inquiring: How does it differ from competitors like OpenClaw and Manus, which Meta once considered acquiring? The distinction is significant. From a market positioning perspective, Muse's strategy is clear: it targets the average consumer and addresses consumer-end issues, whereas the majority of popular Agents currently focus on productivity and office tasks.

More importantly, from a technological standpoint, Muse and OpenClaw differ greatly. OpenClaw is an Agent runtime environment that operates on the user's own computer, with all data stored locally, granting users absolute control and the ability to choose any model to complete tasks. In contrast, Muse is a virtual machine running an Agent, packaged and secured by Meta, with data stored on the virtual machine and only able to utilize Meta's proprietary Muse Spark model.

When Meta states that 'each Muse user has a dedicated computer,' it does not imply that each user is assigned a physical PC. Rather, it means that a pre-configured virtual machine is allocated to each user in the cloud. Being pre-configured, users are spared the cumbersome installation and debugging processes. Since Meta handles security, complex operations involving money can be executed. Additionally, as it does not rely on local computing power, it can operate on smartphones.

This represents the most significant difference between Muse and other Agent products. Competitors such as OpenAI, Anthropic, and even Google have yet to launch similar offerings. Domestic products like WorkBuddy, Doubao, and Qianwen Office follow the OpenClaw approach. Lobster (OpenClaw) requires 'nurturing,' whereas Muse does not—it is foolproof, literally 'usable by anyone with hands,' reducing the barrier to entry for AI Agents to nearly zero.

So, here's the question: When OpenClaw emerged, Chinese tech giants followed suit and achieved considerable success (WorkBuddy comes to mind). Should they follow Muse's lead this time?

Given the style of Chinese tech giants, the answer would likely be yes. This could represent the next market with 100 million DAU, boasting a much broader user base and far greater commercial potential than Lobster. The issue is that the computing power consumption is on an entirely different scale—it could escalate to the point of burdening a tech giant.

Local Agent runtime environments like OpenClaw and WorkBuddy run Loops locally on the user's device, with most data also stored locally. External computing power is only utilized when burning Tokens. Specifically:

Agent orchestration, including task allocation, coordination, and maintaining context, is executed by the user's CPU, with relevant data running in the user's memory. Long-term storage data generated during this process is primarily stored on the user's hard drive or a remote storage location specified by the user (paid for by the user). When calling external large models, the computing power required for Tokens is provided by third-party model vendors or cloud providers, with users simply paying for the results. However, if the user deploys a model locally, even this computing power is the user's responsibility.

Muse completely reverses this model: Agent orchestration is handled by the Muse virtual machine, data is stored in the virtual machine, and only the Muse Spark model can be used. All computing power and storage are provided by Meta, with the user's device acting as a thin client. Even if the device is completely powered off and disconnected from the internet, the Agent continues to run in the virtual machine—it just requires user approval at key points.

Just imagine the additional computing power and storage this model would necessitate. At least 50 million computers in China have installed OpenClaw, WorkBuddy, or their competitors. If tech giants were to adopt Muse's model, it would entail taking over the CPU computing power and storage space of these 50 million computers from users. The sheer scale is mind-boggling and somewhat terrifying.

If this vision had emerged a year ago, Chinese tech giants might have embraced it without hesitation, as capital markets then favored seeing tech giants invest heavily in computing power and storage. When Alibaba announced an increase in its capital expenditure targets, investors celebrated. However, times have changed. When Tencent and Alibaba announced record quarterly capital expenditures, the market reacted with indifference, even skepticism.

The same trend is evident in the United States. When Meta announced a firm increase in capital expenditures in Q2, the market remained unimpressed, prompting Zuckerberg to declare, 'If there's excess computing power, we can rent it out.' Meta's recent surge is largely attributable to its previous decline, and investors, having accepted the seemingly inevitable vision of 'you have to burn money,' were pleasantly surprised to discover that 'burning money can still yield results.' Even so, Meta has not yet recovered its all-time high from a year ago.

Any tech giant, whether Chinese or American, that decides to replicate Muse's model will have to invest heavily once again, and it is unlikely to be achieved by merely adjusting existing capital expenditure plans. Muse requires massive CPU computing power and long-term storage, which do not align with tech giants' established data center hardware structures. Of course, this is excellent news for CPU vendors like Intel and AMD—AMD even surpassed the $1 trillion market cap barrier as a result. For internet giants, it means opening another hole in their already overburdened wallets.

Not to mention that China's user base is significantly larger than North America's. Muse has been a hit in North America but has not yet surpassed 1 million DAU. In China, reaching tens of millions of DAU should not be difficult, implying ten times the computing power and storage requirements. It turns out that Chinese users, regardless of whether their work is AI-related, exhibit the strongest FOMO (Fear of Missing Out) regarding AI Agents globally. An AI entrepreneur friend of mine lamented that only in China would you witness elderly people going out to have Lobster installed—something unimaginable in Silicon Valley. Just envision the scene if a Chinese version of Muse were created.

With such vast, nearly bottomless costs, Chinese tech giants must first determine one thing if they choose to follow: Can domestic users swiftly develop payment habits, or can they immediately generate sufficient revenue through advertising, cross-promotions, etc.? In short: The revenue must come from the users. Who can be certain of that? What if you charge and competitors do not? Chat apps like Doubao and Kimi offer paid options, but the results so far have not been particularly impressive. Will the fate of paid personal Agents be any different?

I sense that during this National Day holiday, the management and AI product teams of tech giants will not be able to rest easy. Everyone will have to seriously consider—should I follow this time? If not, and competitors seize the high ground, who takes the blame? If yes, and a massive additional capital expenditure is required, where will the money come from?

No matter the answer, capital markets probably will not be particularly enthusiastic. Only the computing power and storage supply chain can continue celebrating and reveling.

Wishing everyone a joyful National Day holiday!