Meta rings opening bell in age of AI agents

As 2025 drew to a close, US-based Meta completed a multibillion-dollar acquisition of Butterfly Effect, the Chinese startup behind the artificial agent product Manus. The deal, though faces potential antitrust assessment and risks, has forced the global tech industry to recalibrate.

I remember my first reaction was not of surprise at the price — thought to be around $2 billion, according to some reports, but at the timing. This was not a defensive acquisition made under pressure, nor a speculative bet on a distant future. It was decisive. Meta was buying a ready-to-deploy AI agent company at precisely the moment when the industry narrative was shifting — from competing over model parameters to competing over real-world application.

Inside the industry, the transaction made an immediate impact. This was Meta's third-largest acquisition ever. More importantly, it was a signal that the AI race has entered a new phase. The era of "who has the bigger model" is giving way to a far more brutal contest: who can turn intelligence into action, at scale, for users who are not AI engineers.

Manus sits squarely in that transition. Unlike traditional chat-based AI products, it operates as an agent — planning tasks, calling multiple models, executing workflows and consuming orders of magnitude more inference resources in the process. Research firms estimate that a single Manus task can require up to 100,000 tokens, roughly 100 times the inference load of a standard conversational query.

That number matters. It explains why Meta was willing to pay billions, and why this deal is not simply about acquiring talent or technology, it is about controlling the next layer of AI consumption — the layer that will determine future demand for computing power, cloud infrastructure and downstream services.

Among Chinese investors and founders, the reaction was more conflicted. Some described it as Mark Zuckerberg "buying a ticket onto the AI agent ship". Others lamented yet another Chinese AI company being absorbed by a US tech giant. But reducing the deal to capital arbitrage misses the deeper issue.

Manus followed a familiar path. It was founded by a Chinese team, backed early by top domestic funds including ZhenFund, Hongshan and Tencent, and grew rapidly with a global user base. What is less discussed is that earlier acquisition offers from Chinese tech firms reportedly valued the company at only tens of millions of dollars — two orders of magnitude below Meta's final price.

That gap reflects a structural mispricing of AI application value inside China's tech ecosystem. For years, attention and capital flowed overwhelmingly toward foundation models and infrastructure. Application-layer innovation was treated as secondary, incremental, or easily replicable. Meta's move suggests the opposite: whoever controls agent-level intelligence may ultimately dictate how models are used, monetized and scaled.

From an industry perspective, the implications are stark.

For China's tech ecosystem, it shows that the country can produce world-class AI application teams. What remains uncertain is whether it can retain them. Capital exits are not failures in themselves. But when the most valuable outcomes consistently flow outward, it raises questions about long-term industrial depth and strategic autonomy.

This deal also effectively sets the tone for the AI agent sector. Meta has declared agents a strategic battleground. It is difficult to imagine Google, OpenAI, ByteDance or Tencent standing still. For smaller startups, the choice will narrow quickly: be acquired, or retreat into deep vertical niches with defensible domain expertise.

Still, Meta's logic is clear. In the AI era, tickets to the future are not free. They are purchased with capital, computing power and control over how intelligence is deployed in the real world.

As I step back from the headlines, one conclusion stands out. This acquisition is not an ending, it is the opening bell for the AI agent age. Over the next year, consolidation will accelerate, boundaries will harden and the gap between model builders and application owners will widen.

And somewhere, Chinese investors are already asking the next question: where will the next Manus be born — and will it stay?