That lobster called "OpenClaw" is opening a vault for large model companies.

Friends, especially those following AI and Hong Kong stocks, have you been bombarded by MiniMax news these past two days? Its stock price surged 51% in two days, its market cap once surpassed Baidu, and it's up 600% year-to-date! On the surface, it looks crazy, but behind it is something called “OpenClaw” (playfully nicknamed “Lobster” within the industry), which is fundamentally changing the revenue logic for large model companies.

Simply put, OpenClaw is an “agent assembly workshop.” You no longer need to have a back-and-forth Q&A with AI. Instead, you can equip it with various skills (like searching for information, booking flights), allowing it to automatically handle complex tasks for you 24/7 like a butler. But for this butler to work, it needs to “burn tokens” (consume computing power), and users have to pay for that.

Here's the key: this precisely hits the biggest pain point for large model companies—making money is hard.

• Before (C-end): For AI products like MiniMax's, users exceeded 200 million, but paying users were less than 1.8 million, a conversion rate under 1%. People are fine with free chatting, but hesitate when it comes to actually paying.
• Before (B-end): Selling APIs or doing custom projects either saw gross margins drop to “selling cabbage” levels, or revenue growth relied solely on adding manpower, which is unsustainable.
• Now (The “Lobster” Era): For tangible automation results, users are starting to willingly pay to “burn tokens.” The more complex the task, the token consumption increases exponentially. Some users burned 30 million tokens in just 5 hours, and many spend over ten thousand yuan per month. Users are paying for “results,” not just for “conversation.”

The data is the most direct proof: The daily average token usage of MiniMax's M2 series models increased sixfold within two months, directly driving its annualized recurring revenue from $100 million to $150 million. Model performance leaderboards (like PinchBench) are no longer just news headlines; they are the direct baton guiding which model users choose and how much money a company can make.

The capital market voted with its feet, giving it a price-to-sales (PS) ratio exceeding 600 times. This means the market is willing to bet over 600 yuan on its future for every 1 yuan of its current revenue. Despite volatility from regulatory risk warnings, institutions (like Morgan Stanley, Haitong International) remain optimistic, with ratings of “Outperform” or “Buy.”

In a nutshell: This “Lobster,” OpenClaw, might be pulling large model companies out of the dilemma of “burning cash to tell stories” and into a new business cycle of “users actively paying, consumption-driven growth.” What it's prying open might not just be the application ceiling for agents, but the entire industry's profitability ceiling.