Meta ended its 2025 by announcing the purchase of manus for more than 2,000 million dollars. It says a lot that it is a company that coordinates and directs other people’s models.
Why is it important. Manus reached 100 million in annual recurring revenue without training a single model. Use Claude and Alibaba models to do the heavy lifting. Its differentiation is not in the intelligence of the model but in the execution: planning complex tasks, invoking tools, iterating on results, delivering finished work… a purely agentic model that achieved its little viral moment at the beginning of the year.
OpenAI, Google and company focus on announcing models that fight to get tenths in the benchmarksbut Manus simply makes money by selling that ability to direct other people’s models.
The backdrop. In ranking like those of Chatbot Arena We have been seeing the same pattern for a year: every time one model is crowned the best, another surpasses it in a few weeks. None manages to make a lasting difference and there is no great difference. moatnot even the capillarity and recognition of OpenAI, or the distribution capacity of Google.
Yes, but. Yeah the models are commoditizing and they are increasingly interchangeable, where is the business? Meta just gave his answer: at the application layer. In who controls the distribution where people actually use AI.
Meta spends $70 billion a year (and counting) on AI infrastructure, but Meta AI It is not curdling and Flame 4 punctured. It lacks what Manus has proven to have: a proven ability to turn models into products that people pay to use.
The threat. Meta is facing a distribution problem:
- Google has Android, Google Search, Gmail, Google Docs…
- Microsoft has Windows and its suite of productivity, in addition to being ubiquitous in companies.
- Apple controls the iPhone and the Mac.
And Meta has social and messaging platforms, but its track record in corporate products is disastrous. Workplace never took off and closed. Their reputation with business data isn’t exactly great. Buying technology is easy, but whether companies trust Meta as a provider of work tools is another story.
Chinese arbitration. Chinese AI startups are clearly undervalued:
The Chinese number one is worth less than 1% of the American one with comparable technical capacity, and the penetration of OpenAI justifies the distance, but not at that level. Benchmark Capital saw these figures and he put in 75 million in May at a valuation of 500 million. a few months later goes for more than 2,000 million.
The new manual. Manus has just shown that there is a viable route for Chinese AI startups: turn to cheap and well-trained Chinese talent, develop a product designed for global markets (and not just Chinese) from day one, raise Western capital, move the legal headquarters outside of China (in this case, Singapore), and achieve a “clean exit.”
Other Chinese startups can now be expected to follow a similar path, and the Chinese authorities are aware of this… and are not happy with the idea, according to the wall Street Journal. They see it as a leak of technology developed with local engineers.
And now what. Meta will integrate Manus into Facebook, Instagram and WhatsApp. But the real message goes further: in AI, as happened with the Internet, the infrastructure becomes commodity and the business is in the digital “last mile”, where technology touches the user.
Meta just paid 2.5 billion to not forget that lesson.
Featured image | Manus, Xataka with Mockuuups Studio
GIPHY App Key not set. Please check settings